KEY TAKEAWAYS
- The Anyway Ventures technical due diligence checklist covers twelve areas, from architecture and code to AI claims and the first 100 days, with 107 numbered data-room requests and an Excel workbook.
- Every technical due diligence finding needs a pound figure and a route, and annual costs count at the multiple: an ignored £80k yearly inference bill is £960k of enterprise value at 12 times EBITDA.
- Key-person risk and IP ownership are the most common price-moving findings on lower mid-market deals, so make access transfer and confirmatory IP assignments conditions of completion, not 100-day actions.
- Black Duck's 2026 audit of 947 commercial codebases found open source licence conflicts in 68%, up from 56% a year earlier, partly from AI tools reproducing code without its licence.
- Front-load technical due diligence by sending the P1 data-room requests on day one, interviewing engineers in week one and giving the deal team a red-flag note with rough costs by the end of week one.
This is a complete technical due diligence checklist and data-room request list for private equity, venture and corporate acquirers buying or investing in software and tech-enabled businesses. It covers twelve areas, from architecture and code to AI claims and the first 100 days, and for each one sets out what to check, the evidence to request, what good looks like, the red flags and where each finding should land in the deal. It comes with 107 numbered requests ready to paste into a tracker, an interview plan by role, a costing method, sector variants and a worked example, plus an Excel workbook that totals findings by area, severity and route.
1. Executive summary
- Every finding needs a pound figure and a route, or it is an observation. The routes are price (including debt-like items and run-rate EBITDA), a condition of completion, a specific warranty or indemnity in the SPA, structure (retention, escrow, deferred consideration, earn-out terms), the 100-day plan, or the file. A checklist that ends in a red, amber and green grid has not done the job.
- Annual costs are worth the multiple, not their face value. An inference bill of £80k a year that the plan ignores is £960k of enterprise value at 12 times EBITDA. One-off remediation is worth roughly what it costs. Most deal teams treat the two the same, and the gap is usually where the money is.
- Ask for evidence, not documents. A policy says what should happen. A deployment log, a cloud bill, a commit history and a restore test show what did. Where only the policy exists, that is a finding.
- Key-person risk and IP ownership are the most common price-moving findings on lower mid-market deals. Both are cheap to find and cheap to fix before completion, and expensive after. Make access transfer and confirmatory IP assignments conditions of completion, not 100-day actions.
- Open source and AI-generated code have changed the IP picture. Black Duck's 2026 audit of 947 commercial codebases found licence conflicts in 68%, up from 56% a year earlier, and linked part of the rise to AI tools reproducing code without its licence (Black Duck, 2026). A dependency scan is no longer optional.
- Test the AI story on real data. Ask for usage, cost per call and the model behind each feature. If the CIM says "AI-powered" and nobody can show last month's usage, discount the revenue that depends on it.
- Front-load the work. Send the P1 requests on day one, interview engineers in week one, and give the deal team a red-flag note with rough costs by the end of week one. Findings that move price take longest to negotiate.
- Scope to the question, not the fee budget. A five-day red-flag review answers "should we keep going, and at what price?" A full review gives lawyers, lenders and insurers something to rely on. A vendor review lets a seller fix the cheap problems before a buyer prices them.
2. How to use this checklist
Who it is for. Deal partners and associates who need to commission, scope or challenge technical due diligence. Operating partners who inherit the findings. Corporate development teams running acquisitions without a specialist adviser. Deal lawyers turning findings into SPA drafting. CTOs and founders preparing for a sale, who can run it on themselves first. VC investors can use the red-flag subset for Series A and later rounds; the VC technical due diligence page sets out the lighter version.
When to use it. Pre-LOI, use the executive summary and the P1 requests to frame questions for management. In exclusivity, use all of it. Sellers should run it three to six months before a process, which is long enough to fix missing IP assignments, open pen test findings and single-person admin access before a buyer finds them.
How long it takes. A red-flag review of a single-product business with up to 15 engineers takes five to seven working days. A full review takes two to three weeks. Larger, multi-product or carve-out targets take four to six weeks. The clock starts when code and people are available, not when the engagement letter is signed. Fees and published price points are compared in how much technical due diligence costs in the UK.
Inputs you need. The investment case (the growth, product and margin assumptions the technology must support), read-only repository access, the data room, four to seven interviews, and the target's salary bands so you can cost findings in its own money.
How it connects to the workbook. The Excel companion has five working tabs. The request list holds all 107 items from the data-room request list below, with the same IDs, plus dates, owner, status and a days-open counter. The findings register takes each finding with its area, severity, cost range, cost type and route. The dashboard totals both, by area, severity and route, and capitalises annual costs at the entry multiple you enter. The reference tab holds the severity and route definitions from the costing method and a calculator for engineer-time estimates. The example tab is the Northgate worked example, filled in.
A three-week plan
| When | What happens | Output |
|---|---|---|
| Day 1 | Send P1 requests; agree interview list and confidentiality circle; request read-only code access | Request list sent |
| Days 2 to 5 | First CTO session; engineer interviews; commit history and access review; dependency scan; cloud bill read | Working findings log |
| Day 5 to 7 | Red-flag note to the deal team, each item with a rough cost and proposed route | Two to five page note |
| Week 2 | P2 requests; security and data protection review; product and roadmap testing; AI claims tested on live data | Draft findings register |
| Week 3 | Second CTO session to raise findings; factual accuracy check with management; costing finalised | Final report and 100-day plan |
3. What changed in 2026
Four things move this checklist on from the versions most firms still circulate.
AI-written code. Black Duck's 2026 Open Source Security and Risk Analysis found that 17% of open source components entered codebases outside package managers, through copied snippets or AI generation, and that only 54% of organisations evaluate AI-generated code for IP and licensing risk. The same audit found high-risk vulnerabilities in 78% of codebases and components four or more years out of date in 92%. Google's 2025 DORA report put AI use among developers at 90%. The question is no longer whether a target uses AI to write code, but whether its review, testing and licence scanning kept pace.
UK data protection reform. Most of the Data (Use and Access) Act 2025 data protection changes came into force on 5 February 2026, and the duty to run a data protection complaints procedure (acknowledge within 30 days, investigate without undue delay) has applied since 19 June 2026, as Mishcon de Reya set out in 2026. Ask for the complaints log as well as the breach log.
EU product and AI rules. For targets selling into the EU, Cyber Resilience Act reporting duties for products with digital elements started on 11 September 2026, with the remaining obligations from 11 December 2027 (Skadden, 2026). Under the EU AI Act, the Digital Omnibus pushed high-risk obligations for Annex III systems to 2 December 2027 and for Annex I products to 2 August 2028 (Gibson Dunn, 2026). The EU Data Act's cloud switching rules have applied since 12 September 2025, and switching charges are largely prohibited from 12 January 2027 (Latham & Watkins, 2025), which matters for any SaaS target whose retention depends on high exit costs for EU customers.
Enforcement against software suppliers. In March 2025 the ICO fined Advanced Computer Software Group £3.07m, after a 2022 ransomware attack that came in through a customer account without multi-factor authentication (ICO, 2025). Software businesses that process data for customers carry their own exposure, and MFA coverage on customer-facing admin accounts is now a P1 request.
4. The checklist on one page
| Area | The question the investment committee needs answered | Most common route |
|---|---|---|
| Architecture and scalability | Can the platform carry the growth and product changes in the plan? | Price (known capex), 100-day plan |
| Code quality and technical debt | How much of next year's engineering capacity is already spent? | Price (debt-like item) |
| Team, delivery and key-person risk | Does the product survive any one person leaving? | Structure, condition |
| Security | Is there an unreported incident or an open door? | Condition, specific indemnity |
| Data protection and privacy | Is personal data handled lawfully, and is the paperwork real? | SPA protection, 100-day plan |
| IP, open source and AI-generated code | Does the target own what it sells? | Condition, specific indemnity |
| Infrastructure, cloud and cost to run | Is cost to serve under control and is the business recoverable? | Run-rate EBITDA, 100-day plan |
| Product and roadmap | Does the revenue plan depend on things that do not exist yet? | Price (model adjustment) |
| Third parties, vendors and contracts | Could one supplier or customer contract stop or reprice the business? | Condition, SPA protection |
| IT and internal systems | Do the back-office systems and devices hold up, and what does separation cost? | 100-day plan, TSA |
| AI and data | Is the AI real, used, affordable and governed? | Price (model adjustment), disclosure |
| Integration and the first 100 days | What has to happen between signing and day 100, and what will it cost? | 100-day plan |
Severity labels used throughout are critical, high, medium and low, defined in the severity and costing method. Each area's evidence column references the request IDs in the data-room request list.
5. Architecture and scalability
The question is whether the platform can carry the volume and the product changes in the investment case, and if not, what breaks first and what fixing it costs. Start with the plan, not the diagram: if the model assumes 2.5 times the customers and an enterprise tier, test for those two things.
Items to check
- The architecture diagram matches what engineers describe and what the cloud console shows
- The tenancy model, and the marginal cost and effort of onboarding one more customer
- The first bottleneck at two and five times current volume, as named by the team
- Customer-specific branches, forks or bespoke installs, and how many customers sit on them
- Availability against SLA for 12 months, and whether incidents have root-cause write-ups
- Integration surface: public APIs, partner integrations and how breaking changes are handled
- Whether the architecture supports what the plan needs next (enterprise SSO, data residency, multi-region, new product lines)
Evidence, good and red flags
| Evidence | What good looks like | Red flag |
|---|---|---|
| Team-drawn diagram and inventory (ARC-01, ARC-02) | Drawn recently by engineers; matches the console and interviews | Only the CIM version exists, or engineers contradict it |
| Tenancy model (ARC-03) | Shared infrastructure with logical isolation; onboarding is configuration | A new database, server or deployment per customer, done by hand |
| Incident log and availability (ARC-04) | SLA met; incidents written up with causes and fixes | Repeated incidents with the same cause; no write-ups |
| Bespoke code (ARC-05) | One codebase; customer differences in configuration | Forks per customer, each needing its own upgrades |
| Bottleneck view (ARC-07) | Named component, rough plan and cost, headroom matching the plan | "It scales" or "it's in the cloud" |
Severity and deal route
| Finding | Typical severity | Likely route |
|---|---|---|
| Platform cannot carry the plan's volume without re-architecture | Critical or high | Price as known capex; re-base the plan's timing |
| Single-tenant installs that make each customer a deployment | High or medium | Price or 100-day plan, depending on when growth bites |
| Customer forks requiring separate maintenance | Medium | 100-day plan; warranty on disclosed customisations |
| Missing incident write-ups, no capacity testing | Low | File, 100-day plan |
6. Code quality and technical debt
The question is how much of next year's engineering capacity is already committed to keeping the lights on, and what has to be rebuilt before the plan can happen. Technical debt should be costed, not described. Treat test coverage and static analysis scores as signals: modest coverage of billing and permissions beats high coverage of trivial code.
Items to check
- Components on unsupported or end-of-life versions (languages, frameworks, databases, operating systems)
- The team's own debt list, and whether it is willing to write one down
- Share of engineering time on maintenance, support and bugs against new features
- Tests on the code that matters most: billing, permissions, the core data model, integrations
- Deployment pipeline: automated build, test and deploy, and how often it runs
- Code review practice, including for AI-assisted code
- Capitalised development costs, and whether capitalised work is still in use
Evidence, good and red flags
| Evidence | What good looks like | Red flag |
|---|---|---|
| Repository access (COD-01) | Full history; readable code; consistent structure | Access refused, or a curated copy without history |
| End-of-life list (COD-02) | Short, known and scheduled | Core product on a framework out of support with no plan |
| Debt list (COD-03) | Written, prioritised, roughly estimated | "We don't have much debt" alongside a feared, untouched module |
| Maintenance split (COD-06) | Well under half of engineering time on maintenance | Most capacity on support and fixes, so the roadmap is fiction |
| Deployment logs (COD-05) | Several deploys a week, automated, low failure rate | Monthly manual releases run by one person |
| Capitalised development (COD-07) | Policy applied consistently; assets in use | Capitalised spend on abandoned work, flattering EBITDA |
Severity and deal route
| Finding | Typical severity | Likely route |
|---|---|---|
| Rewrite or major upgrade needed before the plan's product changes | High | Price as a debt-like item; 100-day plan |
| Maintenance consumes most engineering capacity | High or medium | Price via slower roadmap in the model |
| Capitalised development on work no longer in use | Medium | Quality of earnings adjustment; flag to financial DD |
| Weak tests on billing or permissions | Medium | 100-day plan |
| No staging environment, manual releases | Medium or low | 100-day plan |
7. Team, delivery and key-person risk
The question is whether the product survives any one person leaving, and whether the team can deliver the roadmap at the speed the plan assumes. Research on 133 popular GitHub projects found 65% had a truck factor of two or less (Avelino and others, ICPC 2016), and MIT research on around 4,000 US startup acquisitions found 33% of acquired workers left within a year (MIT Sloan, Kim). A deal gives people the money and the moment to leave. The full method is in key person risk in due diligence.
Items to check
- Commit history by area (billing, permissions, core data model, integrations, infrastructure) over 12 to 24 months
- Who owns and administers the cloud root account, code hosting, domain and DNS, app stores, payments, identity and secrets
- Contractors and agencies who hold knowledge, with notice terms
- Who resolves serious incidents, and whether the on-call rota has at least three people
- Onboarding documentation that someone other than its author has used
- Attrition over two years and the reasons
- Retention, notice periods, restrictive covenants and post-deal incentives for key people
- Delivery metrics: deployment frequency, lead time, change fail rate and recovery time, as defined by DORA
Evidence, good and red flags
| Evidence | What good looks like | Red flag |
|---|---|---|
| Commit history (COD-01) | At least two active contributors on every critical area | One author behind most changes to billing or the core data model |
| Admin list (TEA-03) | Company-owned accounts, two or more admins, MFA | Personal email or card on the domain or cloud root account |
| Contractor list (TEA-02) | Contractors on defined work with assignments and handover | Long-serving contractor wrote the core, short notice, no assignment |
| Incident and on-call data (TEA-07) | Serious incidents resolved by different engineers | Same name on every incident, including weekends |
| Contracts and incentives (TEA-04, TEA-05) | Notice and covenants match the 100-day plan; rollover or retention in place | Large cash-out at completion, three months' notice, no reason to stay |
| Leavers (TEA-06) | Low churn; explainable departures | Several senior engineers left for the same reason |
Severity and deal route
| Finding | Typical severity | Likely route |
|---|---|---|
| Founder or CTO is the only person who can run the core platform, with no retention | High | Structure: rollover, retention escrow on handover milestones |
| Critical accounts held personally or by one person | High (cheap to fix) | Condition: transfer before completion |
| Core code written by an unretained contractor or agency | High | Condition (assignment); retention or handover contract |
| Delivery slower than the plan assumes | Medium | Price via the model's roadmap timing |
| Thin documentation, small on-call rota | Medium or low | 100-day plan |
8. Security
The question is whether there is an unreported incident or an open door, and what closing the open findings costs. SOC 2 and ISO 27001 are signals of process during an audit window, not proof of security. The pen test findings, and what happened to them, show what the team does when nobody is checking.
Items to check
- Pen tests: scope, date, tester, open findings and their age
- Incident history, including near misses and any ransomware or extortion contact
- MFA on staff accounts, admin consoles, production access and customer admin accounts
- Vulnerability scanning on code, dependencies and infrastructure, with ageing of high and critical items
- Secrets management, and whether secrets have ever been committed to repositories
- Certifications held and whether their scope covers the product being bought
- What the target has told customers in security questionnaires, against what is true
- Cyber insurance declarations, against what diligence finds
Evidence, good and red flags
| Evidence | What good looks like | Red flag |
|---|---|---|
| Pen tests and remediation (SEC-01) | Annual, reputable tester, findings closed on a documented timeline | Critical or high findings open for months; scope excludes the main app |
| Incident log (SEC-02) | Incidents recorded, responded to and learned from | "We've never had one"; or one that should have gone to the ICO and did not |
| MFA coverage (SEC-03) | Enforced everywhere, including customer admins and break-glass accounts | Optional MFA on customer accounts; shared admin logins |
| Certifications (SEC-04) | Scope covers the product; exceptions are minor | Certificate covers a head office, not the platform |
| Customer questionnaires (SEC-07) | Answers match the evidence | Answers that overstate controls, which is a contract and disclosure risk |
| Insurance declarations (SEC-08) | Consistent with findings | Declarations that diligence contradicts, putting cover at risk |
Severity and deal route
| Finding | Typical severity | Likely route |
|---|---|---|
| Evidence of an unreported or undisclosed incident | Critical | Specific indemnity; possibly a condition or renegotiation |
| Open critical or high pen test findings; MFA gaps on admin access | High | Condition: fix and retest before completion |
| Questionnaire answers or insurance declarations that overstate controls | High or medium | SPA protection; disclosure |
| Certification scope narrower than the product | Medium | 100-day plan |
| Policies exist but are not followed | Low | File |
9. Data protection and privacy
The question is whether personal data is handled lawfully under UK GDPR and the Data Protection Act 2018, as amended by the Data (Use and Access) Act 2025, and whether the paperwork describes reality. Most targets have a privacy notice. Fewer know where their personal data actually sits.
Items to check
- A record of processing activities that matches the data map
- Personal data outside the main database: logs, backups, analytics, support tools, the warehouse, laptops and spreadsheets
- Breach log and any ICO reports, and whether the 72-hour reporting rule was met
- Data processing agreements with every sub-processor, and the sub-processor list given to customers
- International transfer mechanisms and transfer risk assessments
- DPIAs for high-risk processing, including AI features and profiling
- Subject access requests and complaints: volumes, response times and the new complaints procedure
- Retention: a schedule, and evidence deletion actually runs
Evidence, good and red flags
| Evidence | What good looks like | Red flag |
|---|---|---|
| RoPA and data map (DAT-01, DAT-02) | Current, owned and consistent with the architecture | Personal data in places neither document mentions |
| Breach log and ICO contact (DAT-03) | Breaches logged, assessed and reported where required | A reportable breach that was not reported |
| Sub-processors (DAT-04) | DPAs signed; customer list kept current | Processors in use that customers were never told about |
| Transfers (DAT-05) | IDTA or UK Addendum in place; assessments on file | Data in non-adequate countries with no mechanism |
| DPIAs (DAT-06) | Done before launch for AI and profiling features | AI feature using customer personal data with no DPIA |
| Complaints and SARs (DAT-07) | Procedure in place since June 2026; responses on time | No complaints route; SAR backlog |
Severity and deal route
| Finding | Typical severity | Likely route |
|---|---|---|
| Unreported reportable breach | Critical or high | Specific indemnity; disclosure strategy |
| Personal data processed without lawful basis or customer knowledge | High or medium | Specific indemnity; 100-day remediation |
| Missing DPAs, stale RoPA, no transfer mechanism | Medium | Warranty; 100-day plan |
| No complaints procedure or retention schedule | Low | 100-day plan |
10. IP, open source and AI-generated code
The question is whether the target owns what it sells and whether any licence obligation threatens the commercial model. Under section 11 of the Copyright, Designs and Patents Act 1988 an employer owns copyright in code an employee writes in the course of employment. Contractors, agencies and founders writing before incorporation are not covered unless they assigned in writing.
Items to check
- A written assignment or employment term for every person and firm that has written code, including the agency that built version one
- Founder code written before incorporation, and any code brought from a previous employer
- A dependency scan or SBOM for each shipped product, with licences identified
- Copyleft components (GPL, AGPL, LGPL) and how they are used against the distribution model
- Code entering outside package managers: copied snippets, vendored libraries, AI-generated blocks
- AI coding tools: which, under what licence, with what review rules and whether the vendor offers IP indemnity
- Customer escrow agreements and their release triggers
- Trade marks and domains in the company's name
Evidence, good and red flags
| Evidence | What good looks like | Red flag |
|---|---|---|
| Assignments (IPO-01) | Signed by everyone who wrote code, founders included | Early contractor or agency with a licence, not an assignment |
| SBOM or scan (IPO-02) | Mostly permissive licences; copyleft absent or isolated | AGPL in a hosted product; GPL in distributed or on-premise software |
| Distribution model (IPO-04) | Licence analysis matches how each product ships | On-premise or mobile builds never scanned |
| AI coding tools (IPO-06) | Business licences, review gates, licence scanning in the pipeline | Personal accounts, no review change, no scanning |
| IP claims (IPO-03) | None, or resolved | Open licence compliance letters |
| Escrow (IPO-07) | Triggers limited to insolvency and failure to support | Change of control as a release trigger |
Severity and deal route
| Finding | Typical severity | Likely route |
|---|---|---|
| Core code not owned (missing assignment from founder, agency or contractor) | High or critical | Condition: confirmatory assignment before completion; specific indemnity |
| Copyleft licence obligation that conflicts with the commercial model | High or medium | Specific indemnity; replacement cost in the 100-day plan |
| No dependency scanning; unmanaged AI-generated code | Medium | 100-day plan |
| Trade marks or domains held personally | Medium (cheap) | Condition |
11. Infrastructure, cloud and cost to run
The question is whether cost to serve is under control, whether the business can recover from losing its main environment, and whether the hosting line in the model is right. Hosting cost growing faster than revenue with nobody able to explain it is a run-rate EBITDA question, not just a 100-day saving.
Items to check
- Twelve months of cloud bills by account and service, against revenue growth
- Cost allocation by customer or product, and gross margin by product
- Committed spend, reserved capacity and their terms
- How much of production is defined in code, and whether it could be rebuilt from scratch
- Backup coverage and the last full restore test, with the time it took
- Disaster recovery targets (RPO and RTO) and the last test against them
- Monitoring, alerting and who gets woken up
- Engineering tooling costs and renewals
Evidence, good and red flags
| Evidence | What good looks like | Red flag |
|---|---|---|
| Cloud bills (INF-01) | Growth slower than revenue; main lines explained | Cost up faster than revenue; nobody can explain the top three lines |
| Cost allocation (INF-08) | Margin known by product or customer tier | Large customers served at a loss without anyone knowing |
| Restore test (INF-02) | Full restore tested in the last six months, timed | Backups taken but never restored |
| DR plan (INF-05) | Targets set and tested | Single region, no tested failover, targets unknown |
| Infrastructure as code (INF-04) | Most of production defined in code | Hand-built servers nobody dares to touch |
| Commitments (INF-06) | Matched to steady-state usage | Multi-year commits above current usage |
Severity and deal route
| Finding | Typical severity | Likely route |
|---|---|---|
| Cost to serve rising faster than revenue, unexplained | High or medium | Run-rate EBITDA adjustment; 100-day plan |
| Backups never restored; no DR capability | Medium (high if regulated) | Pre-completion undertaking to test; 100-day plan |
| Over-commitment on cloud contracts | Medium | Debt-like item for unused commitment |
| Hand-built infrastructure | Medium or low | 100-day plan |
12. Product and roadmap
The question is whether the revenue plan depends on features that do not exist yet, and how far to trust the team's estimates. The best evidence is last year's roadmap set against what shipped.
Items to check
- Roadmaps from 12 and 24 months ago, against delivery
- Revenue in the plan that depends on unbuilt features or products
- Features promised to named customers, in contracts or sales, with dates
- Cohort retention by product and feature-level usage
- Support volumes and categories, as a read on product quality
- How usage is metered and billed, and whether billing matches contracts
- Accessibility conformance where the target sells to the public sector
Evidence, good and red flags
| Evidence | What good looks like | Red flag |
|---|---|---|
| Past roadmaps (PRD-02) | Most commitments delivered roughly on time | A third delivered; the rest quietly dropped |
| Plan dependencies (PRD-05) | New-product revenue modest or already in build | Year-one revenue from a product with no team assigned |
| Customer commitments (PRD-04) | Few, dated and resourced | Contractual feature promises with penalties, unresourced |
| Cohorts and usage (PRD-03, PRD-06) | Retention holds by cohort; a few features do most of the work | Headline ARR hides decaying older cohorts |
| Support data (PRD-07) | Stable or falling tickets per customer | Rising tickets on core workflows |
Severity and deal route
| Finding | Typical severity | Likely route |
|---|---|---|
| Plan revenue depends on unbuilt features delivered by a team with a weak record | High or medium | Price: discount or delay that revenue in the model |
| Unresourced contractual feature commitments | Medium | SPA disclosure; 100-day plan |
| Billing and contract mismatch | Medium | Financial DD; warranty |
| Accessibility gaps for public sector customers | Low or medium | 100-day plan |
13. Third parties, vendors and contracts
The question is whether one supplier or one customer contract could stop, reprice or reshape the business. Technology diligence reads these contracts for what lawyers can miss: a 30-day termination clause on the API that runs the core feature, or an SLA the platform cannot meet.
Items to check
- Every third-party service the product depends on, with termination, change of control and price change terms
- Top customer contracts: SLAs, service credits, liability caps, uncapped indemnities, escrow and change of control
- Service credits actually paid against SLA performance
- Outsourcing and agency contracts, including exit and handover
- Platform dependencies: app stores, cloud marketplaces, a single payment or messaging provider
- The target's own supplier security reviews
Evidence, good and red flags
| Evidence | What good looks like | Red flag |
|---|---|---|
| Supplier list (VEN-01) | Alternatives known for critical suppliers | Core feature on an API with short-notice termination or uncapped price rises |
| Customer contracts (VEN-02) | Caps on liability; no change-of-control termination | Uncapped data protection indemnity; termination on change of control from a top customer |
| SLA history (VEN-05) | Met, with few credits | Credits paid routinely, or SLAs the architecture cannot support |
| Outsourcing (VEN-03) | Exit assistance and handover terms | Agency holds code, keys or knowledge with no exit terms |
Severity and deal route
| Finding | Typical severity | Likely route |
|---|---|---|
| Major customer can terminate on change of control | High | Condition: consent before completion; price if refused |
| Critical supplier with short-notice termination or uncapped pricing | High or medium | 100-day plan to dual-source; warranty |
| Uncapped indemnities in customer contracts | Medium | Specific indemnity; renegotiate at renewal |
| Agency with no exit terms | Medium | Condition or 100-day plan |
14. IT and internal systems
The question is whether the systems the business runs on (finance, CRM, HR, billing, identity, devices) hold up, and what separating or integrating them will cost. For a tech-enabled services business this area often matters more than the code.
Items to check
- Internal systems inventory with owners, licences and renewal dates
- Identity and access: SSO, joiners and leavers, shared admin accounts
- Device management and disk encryption on laptops
- Spreadsheets and manual steps that produce month-end, billing or board numbers
- Shared services provided by a parent or seller group, and transitional services needs
- Licence compliance for major internal software
Evidence, good and red flags
| Evidence | What good looks like | Red flag |
|---|---|---|
| Shared services (ITS-01) | Listed and costed, with a TSA outline | Carve-out where the parent's systems run billing and nobody has priced separation |
| Identity and devices (ITS-03) | SSO, MDM and encryption on every device | Shared global admin; unmanaged laptops with customer data |
| Manual processes (ITS-04) | Few, documented | Board numbers built by one person in a spreadsheet |
| Licences (ITS-07) | Counts match usage | Under-licensing that an audit would catch |
Severity and deal route
| Finding | Typical severity | Likely route |
|---|---|---|
| Unpriced separation from a parent's systems | High | Price; TSA terms in the SPA |
| Shared admin accounts, unmanaged devices | Medium or low | 100-day plan |
| Key reporting dependent on one spreadsheet and one person | Medium | 100-day plan |
| Software under-licensing | Medium or low | Warranty; true-up cost |
15. AI and data
The question has three parts. Is the AI described in the CIM real and used? What does it cost per use, and is that cost in the margins? Is it governed well enough to survive a customer audit and the EU AI Act? The fuller method is in AI due diligence for private equity and AI washing in due diligence, and the full question bank is our AI due diligence questions.
Items to check
- Every AI feature, the model or vendor behind it and when it launched
- Usage per feature last month and the trend
- Inference and AI vendor cost per month and per unit, at current and planned adoption
- A live demonstration on real data, not a recording
- Evaluation sets, accuracy measures and what happens when the model is wrong
- Training data provenance and rights, including customer data
- AI vendor terms: data use, training rights, retention, rate limits and indemnities
- Governance: policy, risk register, EU AI Act classification, DPIAs
- Who built the prompts, evaluation sets and model integrations (a newer form of key-person risk)
- Exposure: whether AI agents threaten the target's own pricing or customers' need for it
Evidence, good and red flags
| Evidence | What good looks like | Red flag |
|---|---|---|
| Inventory and usage (AID-01, AID-02) | Features named, used and growing | "AI-powered" in the CIM; no usage data |
| Unit cost (AID-03) | Known cost per call; in the margin model | Plan assumes wide adoption with no inference cost |
| Evaluation (AID-05) | Test sets and measured accuracy; fallbacks | Nobody can say how often it is wrong |
| Training data (AID-04) | Rights documented; customer data use permitted by contract | Customer data used to train models without contractual right |
| Vendor terms (AID-06) | Enterprise terms; no training on target's data | Consumer terms; data retained and used by the vendor |
| Governance (AID-07) | Classified against the AI Act where relevant; DPIAs done | Possible high-risk use with no assessment |
Severity and deal route
| Finding | Typical severity | Likely route |
|---|---|---|
| AI claims in the CIM not supported by a working, used feature | High or medium | Price: discount the AI-dependent revenue; disclosure warranty |
| Inference cost missing from the plan's margins | Medium | Price: run-rate margin adjustment at the multiple |
| Customer data used for training without rights | High | Specific indemnity; stop and remediate |
| AI Act high-risk use without a compliance plan | Medium | 100-day plan with budget |
16. Integration and the first 100 days
The question is what has to happen between signing and day 100, who owns it and what it costs. The 100-day plan should contain only what the findings justify, each item with an owner, a budget and a date. If the fund has a wider value creation plan, link the two, as set out in AI due diligence and the value creation plan.
Items to check
- Accounts, domains and licences in founders' or the seller's names that must transfer
- Interim covenants between signing and completion: no major re-platform, hires or vendor changes without consent
- In-flight projects with budgets and status
- Technology budget for this year and next
- Decisions due in the next six months: renewals, hires, vendor changes
- National Security and Investment Act exposure: the mandatory regime covers 17 sectors, including artificial intelligence and data infrastructure, and the government confirmed changes in March 2026 (Herbert Smith Freehills Kramer, 2026)
- For corporate acquirers: which systems are merged, kept or retired, and the cost of each
A 100-day plan template
| Window | Typical actions | Owner |
|---|---|---|
| Before completion | Admin access transferred; confirmatory IP assignments; critical security fixes retested; customer consents | Seller, with buyer sign-off |
| Days 1 to 30 | MFA everywhere; second admin on every critical account; backup restore tested; hiring started for key-person cover | CTO |
| Days 31 to 60 | Documentation sprint tested by a new joiner; dependency scanning in the pipeline; cloud cost review; DPAs and RoPA fixed | CTO, CFO, privacy lead |
| Days 61 to 100 | Debt remediation started with a budget; on-call rota of three or more; AI cost and governance in board reporting | CTO, operating partner |
Severity and deal route
| Finding | Typical severity | Likely route |
|---|---|---|
| NSI Act notification needed but not planned | Critical (void transaction risk) | Condition: clearance before completion |
| Assets held outside the company | High (cheap) | Condition |
| Large in-flight migration with uncertain budget | Medium | Price or interim covenant |
| Integration cost not in the buyer's model | Medium | Price for a corporate buyer; 100-day plan |
17. Data-room request list
All 107 requests, numbered by area so they can be pasted straight into a request tracker or the Excel workbook, which uses the same IDs. P1 items could move the price and go out on day one (40 items). P2 items are needed for a full report. P3 items are useful context and can be skipped on a red-flag review. Where an item does not exist, record "not available": the absence is usually a finding.
Architecture and scalability
| ID | Request | Priority |
|---|---|---|
| ARC-01 | Current architecture diagram drawn by the engineering team, dated (not the CIM version) | P1 |
| ARC-02 | Inventory of products, services, repositories and mobile apps, with languages, frameworks and versions | P1 |
| ARC-03 | Tenancy model: how customer data is separated (shared schema, schema or database per tenant, single-tenant installs) | P1 |
| ARC-04 | 12 months of availability against SLA, and the incident log with root-cause write-ups | P1 |
| ARC-05 | Customer-specific branches, forks or bespoke deployments, and the customers on them | P1 |
| ARC-06 | Load or performance test results, and peak utilisation of the main database and services | P2 |
| ARC-07 | The team's written view of the first bottleneck at two and five times current volume, with a rough cost | P2 |
| ARC-08 | Integration inventory: public APIs, webhooks and partner integrations, with volumes and versioning policy | P2 |
| ARC-09 | Data model documentation and the main database schema | P3 |
| ARC-10 | Design documents or decision records for the last two major architectural changes | P3 |
Code quality and technical debt
| ID | Request | Priority |
|---|---|---|
| COD-01 | Read-only access to all source repositories with full history | P1 |
| COD-02 | List of languages, frameworks, databases and operating systems on unsupported or end-of-life versions | P1 |
| COD-03 | The team's own technical debt list or backlog, with any estimates | P1 |
| COD-04 | Test strategy and recent automated test results, with coverage for billing, permissions and the core data model | P2 |
| COD-05 | CI/CD pipeline configuration and six months of deployment logs | P2 |
| COD-06 | Split of engineering time between new features, maintenance and support, for 12 months | P2 |
| COD-07 | Capitalised development costs for three years and the capitalisation policy | P2 |
| COD-08 | Static analysis or code quality reports, if the team runs them | P3 |
| COD-09 | Code review policy and branch protection settings | P3 |
Team, delivery and key-person risk
| ID | Request | Priority |
|---|---|---|
| TEA-01 | Org chart for engineering, product, data and IT, with role, tenure, location and employment type | P1 |
| TEA-02 | Contractor, freelancer and agency list, with spend, scope, contract dates and notice terms | P1 |
| TEA-03 | Owner and admin list for cloud, code hosting, domain registrar and DNS, app stores, payments, identity provider and secrets vault | P1 |
| TEA-04 | Employment contracts for key technical staff: notice, restrictive covenants and IP clauses | P1 |
| TEA-05 | Current and proposed retention, bonus, option and incentive arrangements for key technical staff | P1 |
| TEA-06 | Leavers in the last 24 months, with role, tenure and reason | P2 |
| TEA-07 | On-call rota, and who resolved each serious incident in the last 12 months | P2 |
| TEA-08 | Delivery metrics if measured: deployment frequency, lead time, change fail rate, recovery time | P2 |
| TEA-09 | Salary bands and loaded cost per engineer by level (for costing findings) | P2 |
| TEA-10 | Onboarding documentation and runbooks, with authors and last-edited dates | P3 |
Security
| ID | Request | Priority |
|---|---|---|
| SEC-01 | Last two penetration test reports in full, with remediation evidence for each finding | P1 |
| SEC-02 | Security incident log for 36 months, including near misses, ransomware and extortion contacts | P1 |
| SEC-03 | MFA and SSO coverage: staff, admin consoles, production access and customer admin accounts | P1 |
| SEC-04 | SOC 2 Type II report, ISO 27001 certificate and statement of applicability, Cyber Essentials status | P2 |
| SEC-05 | Vulnerability scanning set-up and the open high and critical findings, with age | P2 |
| SEC-06 | Secrets management approach, and any history of secrets committed to repositories | P2 |
| SEC-07 | Customer security questionnaires completed in 12 months, and any customer audit findings | P2 |
| SEC-08 | Cyber insurance policy, claims history and the declarations made at the last renewal | P2 |
| SEC-09 | Security policies in use (access, joiners and leavers, incident response), with last review dates | P3 |
| SEC-10 | Named security owner, reporting line and security budget | P3 |
Data protection and privacy
| ID | Request | Priority |
|---|---|---|
| DAT-01 | Record of processing activities (RoPA) | P1 |
| DAT-02 | Data map: where personal data sits, including backups, logs, analytics, support tools and the warehouse | P1 |
| DAT-03 | Personal data breach log and all ICO reports or correspondence | P1 |
| DAT-04 | Sub-processor list with signed data processing agreements, and the customer-facing sub-processor list | P2 |
| DAT-05 | International transfer mechanisms (IDTA, UK Addendum, adequacy) and transfer risk assessments | P2 |
| DAT-06 | DPIAs for high-risk processing, including AI features | P2 |
| DAT-07 | Subject access request log with response times, and the data protection complaints log | P3 |
| DAT-08 | Privacy notices, cookie consent set-up and marketing consent records (PECR) | P3 |
| DAT-09 | Retention schedule and evidence that deletion runs | P3 |
| DAT-10 | DPO or privacy lead, and the last internal privacy review | P3 |
IP, open source and AI-generated code
| ID | Request | Priority |
|---|---|---|
| IPO-01 | IP assignment or employment terms for everyone who has written code, including founders before incorporation and agencies | P1 |
| IPO-02 | Software bill of materials or dependency scan for each shipped product, with licences | P1 |
| IPO-03 | Any IP claims, takedown notices or licence compliance correspondence | P1 |
| IPO-04 | Distribution model per product (SaaS, on-premise, mobile, embedded), for licence analysis | P2 |
| IPO-05 | Commercial code licensed in (SDKs, libraries, white-labelled components), with terms | P2 |
| IPO-06 | AI coding tools in use, licences held, usage policy, review rules and vendor IP indemnity terms | P2 |
| IPO-07 | Source code escrow agreements granted to customers, with release triggers | P2 |
| IPO-08 | Registered IP: trade marks, domains and patents, with owner of record | P2 |
| IPO-09 | Open source policy and any approved licence exceptions | P3 |
Infrastructure, cloud and cost to run
| ID | Request | Priority |
|---|---|---|
| INF-01 | 12 months of cloud bills by account and service | P1 |
| INF-02 | Backup policy, and the last restore test with date and time taken | P1 |
| INF-03 | Hosting diagram: accounts, regions and environments | P2 |
| INF-04 | Infrastructure-as-code repositories, and the share of production defined in code | P2 |
| INF-05 | Disaster recovery plan, RPO and RTO targets, and the last DR test | P2 |
| INF-06 | Reserved instances, savings plans and committed spend, with terms | P2 |
| INF-07 | Engineering tooling subscriptions, with cost and renewal dates | P2 |
| INF-08 | Cost allocation by customer or product, and the gross margin build-up | P2 |
| INF-09 | Monitoring, alerting and on-call tooling | P3 |
| INF-10 | Any data centre, colocation or on-premise hardware, with lease and refresh dates | P3 |
Product and roadmap
| ID | Request | Priority |
|---|---|---|
| PRD-01 | Product roadmap for the next 18 months, with resourcing | P1 |
| PRD-02 | Roadmaps from 12 and 24 months ago, and what shipped against them | P1 |
| PRD-03 | Cohort retention, logo and revenue, by product | P1 |
| PRD-04 | Features promised to customers in contracts or sales, with dates | P1 |
| PRD-05 | Revenue plan items that depend on features not yet built | P1 |
| PRD-06 | Feature-level usage data | P2 |
| PRD-07 | Support ticket volumes, categories and resolution times for 12 months | P2 |
| PRD-08 | Pricing and packaging, and how usage is metered and billed | P2 |
| PRD-09 | Accessibility conformance (WCAG) where selling to the public sector | P3 |
Third parties, vendors and contracts
| ID | Request | Priority |
|---|---|---|
| VEN-01 | Third-party services the product depends on, with spend, term, termination and change-of-control clauses | P1 |
| VEN-02 | Top 20 customer contracts: SLAs, service credits, liability caps, uncapped indemnities, escrow and change of control | P1 |
| VEN-03 | Outsourcing and agency agreements, statements of work and exit provisions | P1 |
| VEN-04 | Standard customer terms and any material deviations | P2 |
| VEN-05 | SLA performance and service credits paid over 24 months | P2 |
| VEN-06 | The target's own supplier security reviews | P3 |
| VEN-07 | Reseller, marketplace and platform terms (app stores, cloud marketplaces) | P3 |
| VEN-08 | Public sector frameworks and contracts with security or data clauses | P3 |
IT and internal systems
| ID | Request | Priority |
|---|---|---|
| ITS-01 | Shared services provided by a parent or seller group, and expected TSA needs | P1 |
| ITS-02 | Internal systems inventory (finance, CRM, HR, billing, support, warehouse), with owner and licence | P2 |
| ITS-03 | Identity and device management: SSO, MDM, disk encryption, joiners and leavers process | P2 |
| ITS-04 | Spreadsheets and manual steps that run month-end, billing or board reporting | P2 |
| ITS-05 | Microsoft 365 or Google Workspace admin set-up and licensing | P3 |
| ITS-06 | IT support model and any outsourced IT contract | P3 |
| ITS-07 | Licence position for major internal software (Microsoft, Oracle, Adobe and similar) | P3 |
AI and data
| ID | Request | Priority |
|---|---|---|
| AID-01 | Inventory of every AI feature in the product and in operations, with the model or vendor behind each | P1 |
| AID-02 | Usage data per AI feature: customers, calls and trend | P1 |
| AID-03 | Inference and AI vendor cost, per month and per unit of use | P1 |
| AID-04 | Training data provenance and rights, including any customer data used for training | P1 |
| AID-05 | Evaluation sets, accuracy measures and failure handling for each AI feature | P2 |
| AID-06 | AI vendor contracts: data use, training rights, retention, indemnities and rate limits | P2 |
| AID-07 | AI governance: policy, risk register, EU AI Act classification and DPIAs | P2 |
| AID-08 | Customer contract terms on AI and customer data | P2 |
| AID-09 | Data platform: warehouse, pipelines, data quality measures and ownership | P3 |
Integration and the first 100 days
| ID | Request | Priority |
|---|---|---|
| INT-01 | Accounts, domains and licences held in founders' or seller group names that must transfer | P1 |
| INT-02 | In-flight major projects (migrations, re-platforms), with budget, status and owner | P1 |
| INT-03 | Self-assessment against the NSI Act mandatory notification sectors | P1 |
| INT-04 | Technology budget for this year and next, opex and capex | P2 |
| INT-05 | Technology decisions due in the next six months (renewals, hires, vendor changes) | P2 |
| INT-06 | Systems the buyer expects to merge, keep or retire, mapped to the target's estate (buyer to complete) | P2 |
18. Interview plan and questions by role
Documents tell you what the target wants you to know. Interviews tell you what the documents leave out. The best interview question in technical due diligence is a request to be shown something, on a shared screen, while you watch. The rationale for each core question, and what good and worrying answers sound like, is in technical due diligence questions.
Who to interview, and in what order
| Session | Length | Purpose |
|---|---|---|
| CTO, first session | 90 minutes | The CTO's own account of the system, strengths and problems |
| Most senior engineer other than the CTO | 45 to 60 minutes | How the system works in practice; who knows what |
| Engineer hired in the last year | 45 minutes | What was confusing; what the documentation missed |
| Whoever handles production incidents | 45 minutes | Reliability, recovery and the 2am reality |
| Product lead | 45 minutes | Roadmap credibility, usage and customer commitments |
| Security or privacy lead, if one exists | 45 minutes | Incidents, open findings, data handling |
| Data or AI lead, if AI is in the investment case | 45 minutes | What the AI does, costs and gets wrong |
| CFO or finance lead | 30 minutes | Cost to run, capitalisation, contracts, tooling spend |
| Customer support or success lead | 30 minutes | What customers complain about and why they leave |
| Agency or main contractor, if they write meaningful code | 45 minutes | Knowledge held outside the company; exit terms |
| CTO, second session | 60 minutes | Reconcile the first account with the evidence; raise findings |
Not every target has all of these roles. For a team of 5 to 30 engineers, four to seven interviews are usually enough: the CTO twice, two or three engineers, the incident handler and the product lead. Interview engineers without the CTO or founders present, and say at the start that the aim is to understand the system, not to assess individuals. Leave the CEO and CFO out of technical sessions. Ask several people the same core questions (how does a change reach production, what would you fix first, who would you call at 2am) and compare. Agree the confidentiality circle with the seller in writing, and record in the report's scope section anyone you could not speak to.
CTO, first session
- If customer volume doubled next year, what breaks first, and what would fixing it cost?
- Walk me through what happens when a customer does the one thing they do most.
- What would you rebuild with six months and no feature requests?
- Show me the last change that went to production, from ticket to deploy.
- Which part of the codebase is everyone afraid to touch?
- Who has admin access to the cloud account, code hosting, domain registrar and app stores?
- Who wrote the first version of the product, and were they employees?
- Which outside service, if it disappeared or doubled its price, would hurt most?
- Why did the cloud bill change last year?
- Show me the AI feature running on real data, and tell me how many customers used it last month.
- What do you want to be doing in two years? (Ask it plainly. The answer prices the retention package.)
Senior engineer and recent joiner
- How does a change get from your laptop to production? Show me.
- If something broke at 2am on a Sunday, who would you call?
- What did you need to know in your first fortnight that was not written down?
- Which part of the system do you avoid, and why?
- How do you use AI coding tools, and does AI-written code get reviewed differently?
- What would you fix first if you were given a month?
- What is the slowest part of shipping a feature here?
Incident handler or platform engineer
- Tell me about the last serious incident: how it was found, who fixed it, what changed.
- When did you last restore from backup, and how long did it take?
- What alerts wake you up, and how many of them are noise?
- Could you rebuild production from what is in the repositories?
- What is still open from the last pen test?
Product lead
- Take me through last year's roadmap and what shipped.
- Which features in the plan have no team on them yet?
- What have sales promised customers that is not built?
- Which features do customers use most, and which did you build that nobody uses?
- Why do customers leave?
Security and privacy lead
- Tell me about your last security incident and your last personal data breach.
- Where does personal data live, including the places it should not?
- Which sub-processors were added in the last year, and were customers told?
- Is MFA enforced for customer admin accounts?
- How do you handle a data protection complaint now that the complaints duty applies?
Data or AI lead
- Which model or vendor sits behind each AI feature, and what does a call cost?
- How do you know when the model is wrong, and what happens then?
- What data trained or tunes it, and do customer contracts allow that use?
- If the model vendor changed its price or terms, what would you do?
- Who else could maintain the prompts, evaluation sets and integrations if you left?
CFO, support lead and agency
- CFO: what is hosting and tooling as a share of revenue, and what is capitalised as development?
- CFO: which technology contracts renew or step up in price in the next 12 months?
- Support lead: what are the top three ticket categories, and are they rising?
- Support lead: which customers have threatened to leave over product problems?
- Agency: what code, accounts or keys do you hold, and what happens if the contract ends?
What to listen for
Who gets named as the person to ask: if it is always the same person, you have found the key-person risk. Where answers go vague: people are precise about what they are proud of and general about what worries them. Rehearsal: identical polished answers from three people about security suggest a briefing, so ask the follow-up nobody prepared for. And the newest engineer's account of deployment, which is usually the accurate one.
19. Severity and costing method
Severity
| Severity | Definition | Default treatment |
|---|---|---|
| Critical | Could stop the deal, or cause material loss or regulatory exposure: an unreported breach, core code not owned, a platform that cannot carry the plan, an unplanned NSI Act notification | Walk away, renegotiate, or condition plus specific indemnity |
| High | Should change the price or terms: a costed re-platform, key-person dependency with no retention, missing assignments, a major customer's change-of-control right | Price, condition, specific indemnity or structure |
| Medium | Real cost or risk that belongs in the first 100 days with an owner and budget | 100-day plan, sometimes a warranty |
| Low | Hygiene | File |
Severity describes consequence, not how easy the fix is. A personal email on the domain registrar is high severity and a £500 fix, which is why it should be a condition of completion.
Costing rules
- Cost in the target's money. Use its salary bands and loaded costs (TEA-09). Where they are not available, state the assumption. This guide uses £85k a year loaded for a mid-level engineer and £110k for a senior hire, which are illustrative assumptions, not market data.
- Use engineer-months. Cost equals engineers times months times monthly loaded cost. At £85k a year, one engineer-month is about £7.1k. Three engineers for six to nine months is £128k to £191k.
- Give a range and its basis. A single number implies precision nobody has. Write the basis next to the number, so the deal team can argue with the assumption rather than the conclusion.
- Add contingency for rewrites and migrations. Twenty to fifty per cent is normal for work the team has not done before; state which you used.
- Count opportunity cost separately. Engineers rebuilding a framework are not building the plan's new module. Show the delay in months and let the model team price the revenue effect, rather than adding it to the remediation cost and counting it twice.
- Separate one-off from annual. One-off remediation is roughly worth its cost. An annual cost the plan misses (inference, hosting, licences, an extra hire) is worth its annual value times the entry multiple, because it reduces run-rate EBITDA. At 12 times, £60k a year is £720k of enterprise value.
- Weight contingent items for indemnity sizing, not for price. A possible regulatory fine or licence claim should not be added to the price as if certain. Estimate the plausible range, and use it to size a specific indemnity, escrow or cap.
- Say when something cannot be costed yet, and what it would take. "Cost unknown until the agency confirms whether it will assign; legal estimate £5k to £15k if it does" is a finding. "May have IP risk" is not.
From severity and cost to route
| If the finding is | Route it to |
|---|---|
| Known, one-off, needed for the plan to happen | Price, as a debt-like item or known capex |
| A recurring cost missing from the plan | Price, through run-rate EBITDA at the multiple |
| Cheap to fix and binary (access, assignment, consent, critical fix) | Condition of completion |
| A known issue with uncertain cost (a licence claim, a possible breach) | Specific indemnity, possibly with escrow; known issues are usually excluded from W&I cover |
| A dependency on a person | Structure: rollover, retention, escrow released on handover milestones |
| A real cost with an obvious fix and an owner | 100-day plan with budget |
| Hygiene | File |
The scope section of the report matters for insurance as well as for the findings. Paul Hastings' 2026 buyer's guide to W&I notes that the scope of protection depends heavily on the quality of diligence and that cyber cover depends on a review of the target's IT and security. State exactly which repositories, commits, environments and people were reviewed. How each finding then appears in the deliverable is set out in what a technical due diligence report should contain.
20. Red-flag, full and vendor scoping
| Type | When to use it | Typical time and fee | What it covers |
|---|---|---|---|
| Red-flag review | Pre-LOI or early exclusivity; uncertain deals; VC rounds | Five to seven working days; £8k to £12k at our rates | P1 requests only; CTO and two engineer interviews; commit and access review; dependency scan; rough cost and route per finding |
| Full buy-side review | Exclusivity on a deal heading to completion; lenders or W&I insurer involved | Two to three weeks; £15k to £30k at our rates | All twelve areas; P1 and P2 requests; full interview plan; costed register; scope and reliance; 100-day plan |
| Vendor review | Three to six months before a sale process | Usually longer, with a fix phase | All areas, written for several unknown buyers; reliance letter for the winner; remediation before launch |
| Top-up review | Buyer relying on a vendor report | Two to five days | Tests the vendor report's key findings and anything that changed since |
Published fees from other UK and US providers are compared in technical due diligence cost UK. Two scoping rules matter more than the fee. First, insist on read-only code access: documents-only diligence is cheaper and reliably misses the findings that matter. Second, scope by the investment committee's questions. "Can the platform carry the growth?" is a different job from "is the AI real?" or "can we separate this from the parent in six months?"
A vendor review is worth doing when there is time to act on it. Fixing missing assignments, transferring admin access and closing pen test findings costs a seller little; a buyer finding them prices them at its own estimate. When to commission one is covered in vendor technical due diligence, and what a seller should fix, and when, is in our exit readiness checklist.
For carve-outs, add a separation workstream: shared code, shared infrastructure, shared licences and shared people, each with a transitional services cost and an exit date. For VC rounds, cut to the red-flag scope and weight team, IP and architecture-for-scale; most of the rest has not had time to matter.
21. Sector variants
The twelve areas apply everywhere, but the weight shifts. Use these as additions, not replacements. Our sector guides cover the deal questions for each industry in more depth.
SaaS
Weight up architecture (tenancy), cost to run and product (the software and SaaS guide goes further). Add:
- Gross margin by product and customer tier, with hosting, support and third-party costs allocated
- Net revenue retention by cohort, reconciled to product usage
- Multi-tenancy isolation testing: can one customer see another's data?
- Enterprise readiness the plan assumes: SSO, audit logs, data residency, uptime commitments
- Exposure to EU Data Act switching rules if retention relies on high exit costs for EU customers
- Seat-based pricing exposed to AI agents doing the users' work
Marketplace
Weight up data, security, third parties and product (the parking and mobility guide covers parking apps and marketplaces). Add:
- Liquidity metrics from raw data: listings, searches, match or fill rate, repeat use, by region
- Payments flow: who holds funds, and whether it relies on a regulated payment provider's permissions or needs its own
- Fraud, fake listings and reviews: detection, losses and chargeback rates
- Trust and safety tooling, and whether user-to-user features bring the service within the Online Safety Act 2023
- Dependence on one acquisition channel or platform (search, app stores, a single partner)
Tech-enabled services
Weight up IT and internal systems, data and AI exposure; weight down code. Add:
- Which delivery steps are automated, which are manual, and the cost per unit of service
- Data quality in the operational systems that the plan's efficiency case depends on
- Off-the-shelf systems and their renewal, pricing and integration limits
- How exposed the service is to AI doing the same work for less, using the AI disruption risk assessment
- Spreadsheet and person dependencies in billing and reporting
Hardware and IoT
Weight up security, IP, third parties and product. Add:
- Firmware update mechanism: signed, over the air, with rollback, and the share of the fleet on current firmware
- UK PSTI compliance for consumer connectable products, which has applied since 29 April 2024 (GOV.UK)
- EU Cyber Resilience Act readiness, including the vulnerability reporting process that has applied since September 2026
- Bill of materials, single-source components, end-of-life notices and the cost of a forced redesign
- Certification (UKCA, CE, radio) and the cost of recertifying after a hardware change
- Device fleet telemetry, field failure rates and warranty returns
- Long-term support obligations to customers for devices already in the field
22. Worked example
Illustrative example. Northgate Software Ltd is a fictional target. The buyer, the figures and the findings are invented to show the method and are not drawn from any real deal. The same register is filled in on the Example tab of the workbook.
The target and the question
Northgate Software Ltd, a fictional UK business, sells scheduling software to facilities management firms: field engineers, jobs, routes and customer reporting. It has £9.2m of ARR, EBITDA of £3.5m and an agreed enterprise value of £42m, 12 times EBITDA. There are 22 people in engineering and product: 15 employees and 7 through a nearshore agency. The co-founder CTO wrote much of the core.
The investment case rests on three assumptions: 2.5 times the customers in three years; an enterprise tier in year two; and an AI tier, sold as "AI route optimisation", reaching 60% of customers.
The buyer commissioned a full review over 15 working days. The red-flag note went to the deal team on day six with seven items and rough costs.
The findings register
| # | Finding | Severity | Cost (£k) | Route |
|---|---|---|---|---|
| N01 | Co-founder CTO wrote 64% of commits to the scheduling engine in 18 months, is sole owner of the cloud root account and domain, and plans to step back within a year | High | 95 to 140 | Structure |
| N02 | Nearshore agency that built the mobile app (2019 to 2022) worked under a licence, not an assignment | High | 5 to 15 | Condition |
| N03 | Customer web app built on AngularJS, out of support since the end of 2021; must be rebuilt before the enterprise tier | High | 255 to 340 | Price |
| N04 | Database per customer on self-managed servers; each onboarding takes three engineer-days; plan assumes 2.5 times the customers | Medium | 128 to 191 | 100-day plan |
| N05 | Two high pen test findings open for seven months; MFA not enforced on customer admin accounts | High | 10 to 20 | Condition |
| N06 | Personal data in application logs and the product analytics tool, not in the RoPA; DPAs missing for 3 of 14 sub-processors | Medium | 12 to 25 | SPA protection |
| N07 | AGPL-licensed PDF library inside the hosted reporting service, with no commercial licence | Medium | 15 to 35 | SPA protection |
| N08 | Cloud bill up 41% against revenue up 19%; no cost allocation; £70k to £110k a year judged avoidable | Medium | 15 to 25 | 100-day plan |
| N09 | "AI route optimisation" is a heuristic solver; the one LLM feature is used by 9% of customers; plan assumes 60% uptake with no inference cost | Medium | 60 to 95 a year | Price |
| N10 | Backups run nightly but a full restore has never been tested; RPO and RTO undocumented | Medium | 8 to 15 | 100-day plan |
| N11 | Shared Microsoft 365 global admin, 40% of laptops outside device management, Cyber Essentials lapsed in 2025 | Low | 6 to 12 | 100-day plan |
| N12 | Largest customer (11% of ARR) can terminate on change of control and has an uncapped data protection indemnity | High | None (about £1.0m ARR at risk) | Condition |
| N13 | Team delivered 38% of last year's roadmap on time; the plan books revenue from a new module in year one | Medium | None (timing) | Price |
The arithmetic behind N03 is four engineers for nine to twelve months at £85k loaded: 4 × £85k × 0.75 = £255k, and 4 × £85k × 1.0 = £340k. N04 is three engineers for six to nine months: £128k to £191k. N01 is a senior hire at £110k loaded with a £20k fee, six months' overlap and a documentation sprint. The ranges exclude contingency, which the deal team added in the model.
The totals
Five findings are high and seven medium; none is critical. One-off remediation totals £549k to £818k. The single annual item, N09's inference cost at the plan's adoption rate, is £60k to £95k a year. By route, one-off costs split as follows: price £255k to £340k, structure £95k to £140k, conditions £15k to £35k, SPA protection £27k to £60k and the 100-day plan £157k to £243k.
Capitalised at 12 times, N09 is worth £720k to £1.14m of enterprise value, more than the largest one-off item. This is the finding most deal teams would have filed as "AI claims somewhat overstated".
What the deal team did with it
- Verdict: proceed with conditions, at a lower price.
- Price: N03 treated as a debt-like item at the midpoint, about £300k. N09 taken through the model as a margin adjustment at the midpoint, about £0.9m of enterprise value. N13 handled by moving the new module's revenue back six months in the plan, which the model team priced.
- Conditions of completion: cloud root account and domain moved to company ownership with two administrators (N01); confirmatory assignment from the agency (N02); pen test findings fixed and retested, MFA enforced (N05); written consent from the largest customer waiving its change-of-control right (N12).
- SPA protection: specific indemnities for pre-completion data protection breaches (N06) and the AGPL exposure (N07), and for any claim arising from the agency's pre-assignment work (N02).
- Structure: the CTO rolls over equity, and part of the CTO's consideration sits in escrow, released on handover milestones at six and twelve months: a second engineer leading releases, documentation tested by a new joiner, incidents resolved without the CTO.
- 100-day plan: budget of £157k to £243k for N04, N08, N10 and N11, plus the senior hire under N01. N08's £70k to £110k a year saving is upside for the value creation plan, not a price chip.
The red-flag note on day six already contained N01, N02, N03, N05 and N09 with rough costs. That is what gave the deal team a week to negotiate rather than a day.
23. Red flags that should stop or reprice a deal
- Refusal of read-only code access without a credible reason and an alternative (a clean room, a supervised session).
- Evidence of an incident that was not reported to the ICO, customers or the insurer when it should have been.
- Core code the target does not own: a founder, agency or contractor who never assigned, and will not.
- One person who holds the platform up and is about to receive a large cheque with no reason to stay.
- A plan that needs volume or features the platform cannot carry without a re-architecture nobody has costed.
- Copyleft code in a product whose distribution triggers its obligations, discovered by you rather than disclosed.
- AI revenue with no usage and no unit cost. If the feature cannot be shown working on real data, the revenue line depending on it is a hope.
- Hosting cost outgrowing revenue with nobody able to explain the main lines.
- Customer contracts the technology cannot honour: SLAs, data residency or security commitments that the evidence contradicts.
- An unplanned regulatory gate: an NSI Act notification that was not scoped, or a high-risk AI use with no compliance plan.
24. Common mistakes
Starting technical diligence after financial and legal. The findings most likely to move price (missing assignments, key-person dependency, an uncosted rewrite) take longest to fix and negotiate. Scope tech DD first, or at least at the same time.
Reporting maturity scores instead of money. A score of 2.8 out of 5 for "engineering practices" gives the lawyers nothing to draft and the model nothing to price.
Adding opportunity cost to remediation cost. Count the engineers' cost once, and give the model team the delay in months so they can price the revenue effect.
Treating annual and one-off costs alike. A missing £80k a year belongs in run-rate EBITDA and is worth the multiple.
Making cheap fixes 100-day actions. Access transfer, confirmatory assignments and critical security fixes cost little before completion and much more after, when the seller has the money and no reason to cooperate.
Taking certificates as conclusions. SOC 2 and ISO 27001 tell you a process existed during an audit window and within a stated scope. Check the scope covers the product you are buying, then read the pen test.
Interviewing only the CTO. The second engineer and the newest hire are where the accurate account of deployment, documentation and key-person risk comes from.
Asking questions with predictable answers. "Do you follow agile?" and "Is your code documented?" get a yes from everyone. Ask to see the tests for the billing code and the document the last new joiner used.
Scanning only the main web app. Mobile apps, on-premise builds, firmware and internal tools have their own licence and security profiles, and are often where the AGPL or the hard-coded secret sits.
Forgetting what the report is for. Lenders and insurers read the scope section. Lawyers read the register. The investment committee reads one page. Write each part for its reader.
25. Glossary
Bus factor (truck factor). The smallest number of people whose departure would stall a project.
Condition (condition precedent). Something that must happen before completion, such as an IP assignment or a customer consent.
Copyleft. Open source licences (GPL, AGPL) that require derivative work to be released under the same terms in some circumstances. AGPL extends this to software offered over a network.
Debt-like item. A known future cost treated as a reduction in the price, in the same way as debt.
DORA metrics. Software delivery measures from the DORA research programme: deployment frequency, change lead time, change fail rate, recovery time and rework rate.
DPA. Data processing agreement between a controller and a processor under UK GDPR.
Loaded cost. Salary plus employer costs, benefits, equipment and overhead for one person for a year.
RoPA. Record of processing activities required under UK GDPR.
RPO and RTO. Recovery point objective (how much data you can afford to lose) and recovery time objective (how long you can afford to be down).
SBOM. Software bill of materials: a list of every component in a piece of software, with versions and licences.
SPA. Share purchase agreement: the contract that carries warranties, indemnities, conditions and the price mechanism.
TSA. Transitional services agreement, under which a seller keeps providing systems or services to a carved-out business for a period after completion.
W&I insurance. Warranty and indemnity insurance, covering loss from breach of warranty. Known issues found in diligence are usually excluded.
26. Sources and further reading
Statistics and regulation cited
- Black Duck, 2026 Open Source Security and Risk Analysis: 947 commercial codebases; licence conflicts, vulnerabilities, outdated components and AI-generated code.
- Department for Science, Innovation and Technology, Cyber Security Breaches Survey 2025/26, published April 2026.
- IBM, Cost of a Data Breach 2025, UK findings, July 2025.
- Information Commissioner's Office, Advanced Computer Software Group fine, March 2025.
- Google, 2025 DORA report, and DORA's software delivery metrics.
- Avelino, Passos, Hora and Valente, A novel approach for estimating truck factors, ICPC 2016.
- MIT Sloan, Your acquired hires are leaving: here's why, on Daniel Kim's research.
- Copyright, Designs and Patents Act 1988, section 11.
- Mishcon de Reya, data protection complaints duty from 19 June 2026, 2026.
- Skadden, Cyber Resilience Act reporting FAQs and checklist, September 2026.
- Gibson Dunn, EU AI Act omnibus agreement, 2026.
- Latham & Watkins, EU Data Act switching requirements, August 2025.
- Herbert Smith Freehills Kramer, government response on NSI Act mandatory sectors, 2026.
- GOV.UK, UK product security regime (PSTI).
- Paul Hastings, Insuring the deal: the role of W&I insurance, 2026.
Other good free checklists, credited
- AKF Partners frames diligence around the investment thesis and the cost of closing critical gaps, across people, process and technology. Its preparation guide is useful for targets.
- Crosslake sets out a broad scope including SDLC, business continuity, customer support and enterprise systems, and offers a pre-LOI review.
- Thomas Prommer's technology due diligence checklist covers ten domains with evidence lists and deal levers.
- Peony's technical due diligence guide has a clear data-room folder structure and ranked red flags.
- TechCrunch's prep checklist for startups is a good founder-side starting point.
- madewithlove on what is in a technical due diligence report is a candid practitioner view of the deliverable.
Most of these stop at the inventory: areas and documents, sometimes red flags. Few cost findings, route them into the SPA or the model, separate one-off from annual costs, or give a numbered request list and a working tracker.
Related Anyway Ventures articles: technology due diligence checklist, technical due diligence questions, what a technical due diligence report should contain, technical due diligence cost UK and key person risk in due diligence. The service is described on technical due diligence.
27. About this resource
Written by Anyway Ventures, the York technology and AI practice led by Brannan Coady, who has been on both sides of technical due diligence as a founder, CEO and investor. Version 1.0, October 2026. Free to use, adapt and share with attribution. Not legal, tax or investment advice.
Questions
What should a tech due diligence checklist include?
At least twelve areas: architecture and scalability, code quality and technical debt, team and key-person risk, security, data protection, IP and open source, infrastructure and cost to run, product and roadmap, third-party contracts, internal IT, AI and data, and integration. For each one a useful checklist names the evidence to request, what good looks like, the red flags, how severe each is and where it goes in the deal: price, SPA protection, a condition of completion or the 100-day plan.
What questions are asked in technical due diligence?
The useful ones ask to be shown something rather than told. What breaks first if volume doubles? Show me the last change that went to production. Who holds admin access to the cloud account and the domain? What is still open from the last pen test? Who wrote version one, and did they assign the IP? Show me the AI feature on real data and its usage last month. Ask engineers separately from the CTO, then compare the answers.
What is private equity technology due diligence?
It is the review a fund commissions before buying or investing in a business to test whether the technology can deliver the investment case. It differs from a general IT audit because every finding is costed in pounds and routed: some move the price, some become conditions or specific indemnities in the SPA, some shape retention and earn-out terms, and the rest go into the first 100 days with an owner and a budget.
What does a tech due diligence report include?
A one-page verdict, a scope and limitations section that lenders and W&I insurers rely on, a findings register in which each finding has a severity, a cost range with its basis and a route, a short section per technical area ending in a plain conclusion, and a 100-day plan. Raw tool output, generic best practice and maturity scores with no cost attached belong in appendices or nowhere.
What is the difference between IT due diligence and technical due diligence?
IT due diligence looks at the internal systems a business runs on: infrastructure, licences, devices, identity, support contracts and security. Technical due diligence looks at the product the business sells: code, architecture, engineering team, delivery and roadmap. A software company needs both, with most weight on the product. A tech-enabled services business usually needs more of the IT and data review, and less of the code.