Skip to main content
Damascode

Engineering perspectives

Technology Due Diligence for Acquisitions: How to Assess Risk, Integration Cost and Value Creation

A decision framework for operating partners and buyers: what to inspect, how to separate material risk from noise, and how to turn a technical finding into a cost range.

19 November 2026 · 43 min read

A target can be making money on software that one person understands, and it can also be carrying a tidy architecture that the deal never needed to change. Technology due diligence is the work of deciding which of those facts matters. It connects evidence about applications, data, security, infrastructure cost and engineering continuity to a decision a buyer can actually take: proceed, reprice, change the structure, fund an integration budget, or rewrite the first-year investment plan.

This article sets out a practical assessment framework for that decision. The framework was developed for this article. It is not an industry standard, a scoring product, or a substitute for a scoped review with access to the target. It is written for operating partners, investment professionals, corporate development teams, and executives who have to judge technology before or just after an acquisition.

How to commission and govern an IT consulting engagement, including reviews that are not tied to a transaction, is covered separately in IT consulting in Portugal: technical due diligence, architecture and modernisation. The rest of this article stays on the acquisition question.

Table of contents

  • Start from the deal thesis
  • What the published evidence supports
  • A six-part assessment framework
  • Application and architecture risk
  • Data and integration risk
  • Cybersecurity and regulatory exposure
  • Infrastructure and recurring technology costs
  • Engineering capability and delivery continuity
  • AI readiness without assumed savings
  • From a finding to a decision
  • Illustrative scenarios
  • Checklist by deal phase
  • How to estimate the cost
  • After the acquisition
  • Assessment and implementation
  • FAQ
  • References

Start from the deal thesis

The first question is not "is the technology good?". It is "what has to be true of the technology for this deal to work?".

A stand-alone buy of a stable business may only need the core systems to keep running, the key people or suppliers to remain available, and the security posture to be tolerable for the hold period. A buy-and-build thesis needs something stricter: a realistic path to add another company without rebuilding reporting, identity, and the operational systems that management will use to run the group. A cost-synergy thesis needs a named system that can actually be retired, and a cost to retire it. A growth thesis that depends on a new digital channel, a pricing engine, or an operational AI use case needs evidence that the data, access model, and engineering capacity can support that use case. If they cannot, the missing capability is an investment, and it belongs in the model before it is described as upside.

Bain described this standard in its 2022 Global Private Equity Report. Technology companies accounted for 31% of buyouts in 2021, and Bain reported that comprehensive technology due diligence was performed in closer to 9% of buyouts generally. Pure-play software deals, where Bain said that kind of diligence is almost always done, were less than 15% of the market. Bain also cited PitchBook data that buyouts of tech-enabled businesses had tripled over the previous five years. The public article does not publish the sample or method behind the 9% figure, so it should be read as Bain's research statement, not as a census of the buyout market. The practical point survives the caveat. A large share of deals now depend on technology, and a technology appendix that is disconnected from the deal thesis does not answer the investment question. Bain's own examples turn on that link: a healthcare-data combination looked complementary until diligence showed that the stronger customer-facing platform sat on duplicative, ageing systems, and the sensible direction of integration was the opposite of the original plan.

McKinsey's often repeated line, that 50 to 60 percent of initiatives intended to capture synergies are strongly related to IT, comes from a January 2011 McKinsey Quarterly article by Hugo Sarrazin and Andy West. They presented it as a finding from McKinsey's post-merger work. They did not publish a sample size, a period, or a geography. Treat it as a 2011 practitioner observation about how synergy plans depend on systems. Do not treat it as a measurement of the 2026 market, and do not treat "related to IT" as "caused by the IT department" or as a share of deal value.

What the published evidence supports

Four newer sources are useful if they are kept in their lane.

Integration is a budget, and it is not the same thing as IT. PwC's 2023 M&A Integration Survey reports interviews conducted in 2022 with 232 senior executives at Fortune 1000 companies that had completed a merger or acquisition in the previous three years. An independent firm conducted the telephone interviews. In that 2022 fieldwork, 59% of respondents said their company spent 6% or more of deal value on integration, compared with 38% in PwC's 2019 survey of 200 respondents. Among respondents PwC classified as "Successful M&A Organizations", 78% spent at least 6% of deal value, compared with 56% of the other respondents. The sample is large US corporates: 44% had revenue from $1 billion to under $5 billion, and 37% reported a purchase price under $1 billion, so it is not a mid-market private equity universe. The spend is self-reported, and "successful" is PwC's classification. The figure is total integration spend. It is not an IT budget, not a technology-failure rate, and not a rule that every buyer should reserve 6%.

Technical debt estimates from large CIOs do not transfer to a smaller target. In July 2020 McKinsey surveyed 50 CIOs at financial-services and technology companies with revenue above $1 billion, and published the results on 6 October 2020. Those CIOs reported that 10 to 20 percent of the technology budget dedicated to new products was diverted to resolving tech debt. They estimated that tech debt amounted to 20 to 40 percent of the value of their technology estate before depreciation. Sixty percent said their organisation's tech debt had risen perceptibly over the previous three years. These are self-estimates by CIOs of large firms, not audited balances, and "value of the estate" is not a line in the financial statements. A buyer of a €40 million industrial company cannot quote the 20 to 40 percent range as that company's hidden liability.

Breach statistics describe investigated incidents, not the expected loss on a target. Verizon's 2025 Data Breach Investigations Report, released on 23 April 2025, analysed 22,052 incidents, of which 12,195 were confirmed breaches, across victims in 139 countries. Ransomware was present in 44% of the breaches reviewed, up from 32% in the prior report. In that dataset, ransomware was a component of 39% of breaches at larger organisations and 88% of breaches at SMBs. Verizon also warned that ransomware is easier to document than many other breaches, because the actor often discloses the victim, and said it sampled ransomware cases rather than ingesting every one, to limit that bias. The 44% figure is a share of reviewed breaches. It is not the probability that a given company will be hit this year. IBM's 2025 Cost of a Data Breach Report, researched by the Ponemon Institute and published by IBM on 30 July 2025, put the global average cost at USD 4.44 million, down from USD 4.88 million in the 2024 report. The United States average was USD 10.22 million. Ponemon studied 600 organisations that had a breach between March 2024 and February 2025, across 17 industries and 16 countries or regions, and interviewed 3,470 leaders. Breaches in the study ranged from 2,960 to 113,620 compromised records; very small and very large breaches were excluded. Cost is an activity-based estimate covering detection and escalation, notification, post-breach response, and lost business. It is an average among organisations in the study that already suffered a breach of that size. It is not an expected-loss figure to drop into a mid-market model because a backup test is missing.

Cloud cost discipline is still unfinished even among large spenders. The FinOps Foundation's 2025 State of FinOps survey collected 861 responses from practitioners whose organisations together represent about USD 69 billion of annual public cloud spend. Thirty-one percent of respondents were at organisations spending more than USD 50 million a year on public cloud, 41% were at organisations with more than 20,000 employees, and more than 20 respondents were at organisations spending USD 1 billion a year. Workload optimisation and waste reduction was the top current priority, named by 50% of practitioner respondents. Full allocation of cloud spend, and accurate forecasting, were the next priorities. This is a community survey of people who already practise FinOps, skewed to large cloud estates. It does not say that 50% of cloud spend is waste. It does say that allocation and waste remain open problems where cloud spend is large and professionally managed. A target with no owner for the invoice, and no split by environment or product, is behind that standard. The absence of an allocation is the finding. A savings percentage is not.

Use these sources to frame judgement. Build the number that goes into the investment memo from the target's own evidence.

A six-part assessment framework

The framework below is a way to organise an acquisition review. Score nothing until the deal thesis says why the dimension matters. A missing monitoring stack is a different fact in a regulated payments business and in a single-site manufacturer whose downtime is visible on the shop floor within minutes.

DimensionQuestion the buyer is answeringEvidence that is actually usefulBusiness consequence if the answer is weak
Application and architectureWhich systems the business cannot operate without, and whether they can survive the hold period and the thesisInventory tied to revenue or operations, owners, runtime versions, failure domains, supplier contractsUnplanned replacement, halted releases, or a growth plan the architecture cannot carry
Data and integrationWhether management can see the group, and whether another company can be added without a new reporting projectSystem map, reconciliation samples, manual bridges, a trial consolidated packSlow closes, bad add-on decisions, integration cost discovered after signing
Cybersecurity and regulatory exposureWhich failures are plausible, and which legal regimes might applyIdentity model, privileged access, restore tests, vulnerability status, incident contacts, a counsel view on scopeOperational disruption, notification duties, remediation cost, deal friction
Infrastructure and recurring costWhat technology costs every month, and what breaks if volume growsInvoices by service and environment, utilisation, licences, support contracts, capacity limitsMargin leakage, a step-change in spend, or a false synergy
Engineering continuityWhether anyone other than the current authors can change the system safelyRepository control, a second person who can release, test evidence, backlog, documentationDependency on one person or vendor, stalled product change, key-person risk
AI readinessWhether a proposed automation has data, controls, and a measurable outcomeSource systems, access rules, process fitness, operating cost, human oversight, baseline metricA pilot that consumes budget and does not move a business number

A red flag in the report is a prompt to investigate, not a conclusion that value has already been lost.

Application and architecture risk

Start with the applications that take orders, schedule work, invoice, pay people, or keep a regulated record. Ask who owns each one, who last changed it, and what happens to revenue or operations if it stops for a day. An inventory that lists every laptop and none of those processes is the wrong inventory.

Then test whether those systems are maintained. Maintenance is visible: a repository the buyer can be shown, a release in the last two quarters, a named person who can deploy, a runtime and framework that the vendor still supports, and a dependency list that is not frozen on abandoned libraries. Unsupported runtimes matter because security patches stop, and because the next integration has to be built on a platform engineers can still hire for. They do not, by themselves, mean the system must be replaced in year one. Replacement is justified when the thesis needs a change the current stack cannot absorb, or when the vendor end-of-support date falls inside the hold period and no supported path exists.

Single points of failure sit in more places than the server room. One database with no replica, one payment provider with no degraded mode, one VPN that every warehouse uses, and one engineer who is the only person with production access are the same class of fact. The business consequence is different in each case. Loss of the database can stop invoicing. Loss of the engineer can stop change, while daily operations continue, until the day a certificate expires or a schema has to move.

Architecture for growth is a thesis question. A monolith that comfortably serves one brand in one country is not a liability because a conference slide prefers services. It becomes a constraint when the buyer plans a second brand, a second country, or a second acquired company on the same codebase and the data model, tenancy, or release process cannot isolate them. The financial consequence is either a longer integration, a period of dual running, or a decision to keep the companies operationally separate and give up the synergy that was used to justify the price.

Supplier concentration belongs in this section as well as in the engineering section. If the critical application is owned by a vendor, read the contract for assignment, change of control, escrow, audit rights, price rises, and what happens if the vendor stops trading. If the intellectual property was never assigned to the target, the buyer may be purchasing a business that does not own the software it runs. That is a legal conclusion for counsel, reached from a technical and contractual fact the diligence team should surface early.

A deeper method for reading debt, scalability, and maintainability outside a live deal is in the software architecture audit. In a transaction, the audit's findings still have to be sorted into "fund now", "fund if the thesis requires it", and "accept for the hold period".

Data and integration risk

Group reporting fails in ordinary ways. The ERP, the CRM, the billing tool, and the warehouse system each have a customer or a product, and the identifiers do not match. Finance exports to a spreadsheet, an analyst corrects the exceptions from memory, and the board pack is a file that only one person can rebuild. Month-end then depends on that person, and a buyer cannot see whether the acquired revenue is the same revenue the commercial team describes.

In a buy-and-build, the problem compounds. Two companies can both be profitable and still be unable to produce a consolidated view of margin by customer, stock by location, or backlog by week. The delay is not aesthetic. Pricing, working-capital, and add-on decisions wait on a report that does not exist. Integrating the companies "at the system level" is sometimes the wrong ambition. A narrower path, a shared chart of accounts for the lines the thesis cares about, plus a reconciled feed, can be enough for management control while the operational systems stay separate. That path still has a cost, and it still needs an owner. Pretending that a data warehouse will appear in the first quarter without a source-system map is how integration budgets slip.

Ask for a sample. A map of systems is weaker evidence than one month of revenue, or one week of orders, traced from the operational system to the general ledger with the reconciling items written down. Manual exports, CSV drops, and inboxes used as integration middleware are findings. They become material when the thesis needs the number they produce to be faster, more frequent, or trustworthy enough for a covenant, a pricing change, or the next acquisition.

Data engineering is the implementation discipline behind that path: pipelines, models, and reconciliation, scoped to the decisions management will take. Diligence should name those decisions before anyone proposes a platform.

Cybersecurity and regulatory exposure

Separate three layers, and do not let them collapse into a single "compliance" rating.

The technical layer is observable. Identity and access: who can reach production, whether joiners and leavers are actually removed, whether administrators use shared accounts. Privileged access: standing global admin rights versus time-bound elevation. Backup and recovery: whether a restore has been completed, not whether a backup job is green. CISA's #StopRansomware Guide tells organisations to keep offline, encrypted backups of critical data and to test that those backups can be restored in a disaster-recovery scenario. CISA's small-business backup guidance makes the same point in operational language: test full and partial restores, and write down the procedure. An untested backup is an unproven recovery. It is not yet a measured loss.

Vulnerability management and exposed APIs are the same kind of evidence. Ask for the last assessment, the date, the scope, and the items still open. Compare internet-facing products with CISA's Known Exploited Vulnerabilities catalog. The OWASP API Security Top 10 (2023) is a useful list of failure classes for customer and partner APIs. It is not a prevalence statistic. Third-party dependencies belong here too: the payroll provider, the hosted ERP, the agency with production credentials, the open-source libraries in the order service.

The incident layer is whether anyone would notice and know whom to call. Logs that exist and are never reviewed, and an incident contact that is the same single developer who holds production access, are continuity findings as much as security findings.

The legal layer is not established by the technical review. The NIST Cybersecurity Framework 2.0, published on 26 February 2024 as NIST CSWP 29, is a taxonomy of outcomes an organisation can use to describe and prioritise cybersecurity. NIST is explicit that the Framework does not prescribe how outcomes must be achieved. Using it as a discussion structure does not certify the target, and it does not prove legal compliance.

Where European law may matter, record the fact pattern and hand the scope question to counsel. Regulation (EU) 2016/679 sets, in Article 83(5), an upper tier of administrative fines of €20 million or 4% of worldwide annual turnover for the preceding financial year, whichever is higher, for defined classes of infringement. That is a statutory ceiling. It is not an expected fine, and a missing access review does not establish an infringement. Directive (EU) 2022/2555 (NIS2) requires, in Article 20, that management bodies of essential and important entities approve and oversee cybersecurity risk-management measures, and it provides for management liability. Whether a target is in scope is a legal question. In Portugal the transposition is Decreto-Lei n.º 125/2025, covered from the engineering side in NIS2 in Portugal: cybersecurity requirements and technical readiness. Regulation (EU) 2022/2554 (DORA) matters when the target is a financial entity or an ICT provider to one. A technical team can collect evidence of ICT risk management. It cannot, by itself, determine DORA scope.

Application and API security reviews, identity work, and recovery tests sit in this technical layer. They are not a penetration-test certificate, a SOC retainer, or a legal opinion unless that is what was contracted.

Infrastructure and recurring technology costs

Recurring technology cost hits the P&L every month: hosting, support, licences, and the people required to keep the estate running. It affects margin when it is cash, ongoing, and not already in the seller's run-rate in a way the buyer has understood. A one-off migration does not become EBITDA by being large. A cloud optimisation does not become EBITDA either, unless the saving is realised, recurring, and not spent again on growth, a reserved commitment, or the migration itself.

Ask for invoices, not a rounded IT budget. Split cloud and hosting by account, service, and environment. A bill that has grown for four quarters with no owner, no tagging, and no distinction between production and forgotten projects is a control finding. It might also be a cost-reduction opportunity. Those are different conclusions, and the second one needs a utilisation view: idle capacity, oversized databases, log retention, and non-production estates left running. The FinOps survey cited above is the reason to insist on allocation before announcing a saving. The engineering detail of rightsizing, commitments, and unit cost is in Enterprise FinOps. Cloud and DevOps engineering is relevant when the buyer wants that work done, not when the memo only needs the invoice understood.

Reliability belongs next to cost. An architecture that is cheap because it has no monitoring, no on-call, and no tested failover will show a flattering run-rate until the first serious incident. Availability claims should be checked against incident records and against what the business does when the system is down, including the manual workaround and who performs it.

Licences and vendor contracts are a separate recurring commitment. Note term, renewal, price-increase clauses, audit rights, and whether the contract can move to the buyer. A licence that cannot be assigned, or a support contract that expires inside the first year with a reinstatement fee, is a cash item. Count it when the contract says so. Do not invent a penalty the contract does not contain.

Scalability constraints are capacity facts. A database approaching storage or connection limits, a licence cap on seats or transactions, or a single-region deployment with no workable path to a second site, each implies a future spend or a growth ceiling. The implication should be dated: what breaks at the volume in the investment case, not at an abstract "scale".

Engineering capability and delivery continuity

Owning software and being able to change it are different assets. A repository the target does not control, a release process that is a person remoting into a server, and a test suite that nobody runs, mean the buyer has a system in production and does not yet have a delivery capability. The operational consequence is that the first urgent change, a tax rule, a customer integration, a security patch, waits on whoever still remembers the system.

Internal ownership is a name: a person on the target's side who can explain the architecture, accept a release, and stay after completion. Supplier concentration is the absence of that name, or a contract that puts design authority entirely outside the company with no exit assistance. Documentation is evidence when it matches the system a reviewer can see. A wiki last edited three years ago is a clue, not proof of the current design.

Release, deployment, and testing show whether change is repeatable. A buyer does not need a particular methodology. The buyer needs to know whether a small change can be built, checked, deployed, and reversed without heroics. A maintenance backlog is material when it contains defects or obligations that affect customers, cash, or security, and someone can point to the items. A long backlog of wishes is not, by itself, a valuation item.

The forward test is the thesis again. If the 100-day plan needs an integration, a new price list, or a customer portal, ask who would build it and what they would have to learn first. If the answer is "the same external developer, and only if they are available", the finding is a continuity risk with a capacity constraint behind it. Software development outsourcing can supply that capacity after closing. It does not remove the need to secure source, access, and a handover while the current author is still in reach.

AI readiness without assumed savings

An AI pilot is a demonstration that a model can produce text or a prediction on a sample. Operational value is a change in a business outcome, at a known cost, with a person accountable for the errors. Diligence should refuse to bridge those two with an assumption.

Data quality and ownership come first. If the records the model would use are duplicated, undefined, or trapped in a spreadsheet that changes every month, the model will reproduce that confusion at higher speed. Access control matters because a useful assistant is connected to systems of record. A tool that can read payroll, tickets, or customer contracts needs the same identity rules as the systems it reads, plus a log of what it did. Process suitability is the unglamorous filter: repetitive, bounded, already measured work is a better candidate than a judgement the company has never defined. Integration cost is usually larger than the model fee. Operating cost includes inference, storage, review time, and the vendor minimum. Human oversight is a design choice, not a slogan: who checks the output before it hits a customer or a ledger, and what happens when that person disagrees.

The measurable outcome has to exist before the pilot is called a synergy. Cycle time, error rate, containment of a support queue, or days to close are acceptable baselines. "Productivity" without a denominator is not. Nothing in this section assumes that adopting a model reduces headcount or cost. Some uses will. The diligence product is a short list of uses that have a baseline, a data source, a control, and a cost, and a longer list of uses that do not.

The EU AI Act, Regulation (EU) 2024/1689, adds duties that depend on role and risk class. A technical review can inventory systems and describe logging, access, and human review. It does not classify legal risk by itself. For the data foundations, see Building AI-ready enterprise data. Implementation advisory on use cases and integration sits with AI consulting when the buyer wants it built, not merely described.

From a finding to a decision

FindingTechnical evidencePotential business consequenceQuestions for the investorPossible action
One external developer maintains the system that takes ordersRepository location, last 20 releases, who can deploy, contract and IP assignmentChange stops when that person is unavailable; ownership of the code may be unclearHas anyone else deployed in the last year? Does the target control the repository and the IP?Knowledge transfer and access handover before close if the deal depends on the system; hold back a structural protection if IP is unresolved
Two acquired companies run ERPs that cannot consolidate marginSample month traced to the ledger, list of manual journals, identifier mismatchesGroup decisions on price, stock, and the next add-on wait on a pack one person assemblesWhich lines of the thesis need a common number, and how fast?Fund a reconciled management view for those lines; defer full ERP merger until that view proves the case
Cloud spend is rising and nobody owns the invoiceTwelve months of invoices, tags, environment split, utilisation of the top servicesRun-rate margin is harder to trust; a "saving" may be idle waste or it may be growth the model wantsIs the growth in the volume case, or in forgotten environments?Allocation and a 90-day waste review; put only evidenced, recurring savings into the model
Backups have never been restored in a testBackup configuration, last successful job, absence of a restore record, RPO the business expectsA ransomware or deletion event may be unrecoverable inside the recovery time the business assumesWhich systems, if lost, stop revenue this week? When was the last real restore?A restore test on a copy before or immediately after close; treat an untested backup as contingent, not as a booked loss
Operational reporting is a spreadsheetNamed file, author, sources, reconciling items, time to rebuildBoard and lender packs depend on one person; add-on integration repeats the same gapWhat decision this month was taken on that file?Document the pack, then replace the manual steps that affect cash or the thesis
A core platform loses vendor support inside the hold periodVendor lifecycle notice, version in production, workaround, volume assumptionsForced migration, dual running, or a security exception the buyer may not acceptWhat still works on the current version, and what does the thesis need to change?Date the migration, estimate one-off and dual-run cost, and decide whether that spend is in the price or in the post-close plan

Illustrative scenarios

The situations below are illustrations of how to investigate. They are not case studies, and they are not Damascode engagements. No monetary loss is asserted, because the loss depends on facts the illustration deliberately leaves open.

A single external developer on the order system. The alarming version of this finding is "key-person risk, therefore value down". The investigation comes first. Confirm whether a second person has deployed, whether source and infrastructure logins sit in a company-controlled store, whether the contract assigns IP and survives a change of control, and how often the system actually changes. A stable system, fully assigned, with a rehearsed handover, is a manageable transition. A system that changes every week, with source only on the developer's laptop, is a different risk: the buyer may be unable to patch or to integrate, and may not own what it thinks it bought. Materiality follows from that second pattern plus the revenue process that stops if the system stops. It does not follow from the headcount of one.

Incompatible ERPs after a buy-and-build. Two ledgers and two product masters do not prove that the thesis has failed. They prove that a consolidated number does not exist until someone builds it. Investigate which decisions require the consolidated number, whether a mapping already lives in a spreadsheet, and whether either ERP is the one the group should keep. A full replacement is one option. A reporting layer over both, with a reconciliation the CFO will sign, is often the option that matches a hold period. The finding becomes material when the price assumed a synergy, such as shared purchasing or a single stock pool, that this landscape cannot deliver without a project nobody has scoped.

Cloud cost without an owner. Growth in the invoice is not waste. Investigate the top services, the split between production and everything else, commitments that cannot be cancelled, and whether unit volume in the commercial case explains the curve. Only the residual, the estate nobody can tie to a product or an environment, is a candidate for a near-term reduction. Even then, the reduction is a hypothesis until a change is made and the next invoice confirms it. Booking that hypothesis as EBITDA at signing overstates what diligence knows.

No successful restore test. The existence of a backup job is not recovery. Investigate scope (which systems, which data), where copies are stored, whether they are offline or merely in the same account ransomware can reach, and whether anyone has restored them onto a clean system. CISA's guidance, cited above, is the reason to ask for the test, not a reason to assume a breach. The Verizon and IBM figures explain why recovery is worth testing. They do not price this target's incident. Until a restore is attempted, the right label is contingent uncertainty. After a failed test, the right label is a remediation project with a scope, plus a residual operational risk if the business cannot tolerate the restore time that was just measured.

Fragmented operational data. A group that cannot produce one view of orders, stock, or margin will be slow to manage the asset and slower to buy the next one. Investigate whether the gap is a missing warehouse, a missing identifier, or a missing habit of reconciliation. The first two are projects. The third is a management routine that can start before any new system, and it tells the buyer how much of the pain is tooling. Treat the finding as material when a decision in the 100-day plan, a price rise, a plant closure, a further acquisition, needs the missing number.

A legacy platform that needs replacement sooner than the model assumed. Vendor end of support, a framework with no security releases, or a capacity limit inside the volume case can pull spend forward. Investigate the vendor's published lifecycle, the version actually installed, whether an upgrade exists short of replacement, and which business process would stop if the platform were frozen. "Legacy" is not a cost. A dated customer portal that will not change during the hold can be accepted. A warehouse control system that cannot take the volume in the base case cannot. The cost section below is how to turn the second situation into a range.

Checklist by deal phase

This checklist is a set of evidence requests. It does not replace a scoped technical due diligence engagement, a legal review, or the buyer's own investment judgement. Items marked for signing are the ones that change price, structure, or the decision to proceed. Confirmatory work is what needs access that sellers often grant only after exclusivity. Post-close work is what ownership makes possible and signing could not finish.

Evidence requestBefore signingConfirmatory diligenceAfter acquisition
Inventory of business-critical applications, tied to the process they supportRequest it and compare it with the revenue walk-throughSample three systems: owner, version, where it runsMake it the operating list, with a named owner
Application owners and maintenance providers, internal or externalNames and contracts for anything on the critical pathInterview the owner; read assignment, change-of-control, and exit helpMove access to company-controlled identities
Architecture and integration documentationAsk what exists; note gaps against the thesisCompare documents with repositories and runtimeKeep only documents that match production
Critical dependencies and end-of-life componentsVendor lifecycle for core platforms and internet-facing softwareDependency list from the repository or bill of materialsPatch, upgrade, or accept with a dated exception
Last vulnerability assessment and open remediationDate, scope, and unresolved critical itemsRe-test the items that face the internet or hold sensitive dataClose or formally accept each open item
Evidence a backup was restored, not only takenAsk for the last restore record on critical systemsRun a restore onto an isolated copyPut restore tests on a calendar, with a measured recovery time
Cloud and hosting spend by service and environmentTwelve months of invoicesTag or export a split; identify unowned spendMonthly review against the investment case
Recurring licences and contractual commitmentsSchedule of contracts, renewal dates, assignment limitsCounsel reads assignment, audit, and price clausesCalendar the renewals that fall inside 12 months
Systems required for consolidated group reportingList the numbers the thesis needsTrace one period from source to ledgerCFO signs the pack, including what is still manual
Manual processes that depend on spreadsheets or one employeeName the packs and the authorsTime a rebuild; list the sourcesDecide which manual steps stay and which get a feed
Development, deployment, and incident routinesWho can release, who is called at nightObserve one low-risk change, or a rehearsalWrite the path down and remove shared privileged accounts

Before signing, refuse to let absence of access become silence. A seller who cannot show a repository, a contract, or an invoice has given you a finding: the fact is not available, and the memo should say what decision is being taken without it. During confirmatory diligence, prefer a small number of deep samples over a wide questionnaire that nobody checks. After acquisition, the list becomes an operating routine. The sections below say what to do first.

How to estimate the cost of a technology risk

A finding becomes a number only after it is split. Mixing these categories is how memos invent precision.

  1. Immediate remediation. The work required to remove an unacceptable operational or security exposure on the current system: access handover, a restore test, patching a known exploited product, recovering source into a company repository.
  2. Required modernisation. Change the thesis needs from this system during the hold, which the current system cannot provide without structural work. If the thesis does not need it, it is not required.
  3. Integration expenditure. The cost of making two or more companies report, identify customers, or operate together to the degree the thesis demands. Full system merger is a choice, not the definition of integration.
  4. Recurring operating cost. People, support, licences, and hosting that continue after the project ends, including a retainer that exists because the capability was not transferred.
  5. Contingent risk. A loss that depends on an event that has not happened and may not: a failed recovery during an incident, a vendor who disappears, a contract that fails assignment. Describe the trigger and the range of outcomes. Do not book the downside as if it were incurred.
  6. Longer-term investment. Work that could improve the asset and is not required for the base case. Keep it out of the price adjustment. Put it in the value-creation plan if someone will actually fund it.

Keep the money types separate as well. One-off implementation cost is a project. Recurring operating expenditure repeats. Capital expenditure is an accounting classification under the buyer's policy, not a label a technical reviewer can assign because the work is "large". Internal staff time is a capacity constraint even when it is not invoiced. External supplier cost is the invoice. Disruption during migration is lost throughput, dual running, and the period in which the old and new systems can both be wrong. It is often the largest unknown, and it should stay visible rather than being hidden inside a day-rate estimate.

An illustrative build-up

The figures in this example are assumptions chosen to show the arithmetic. They are not market rates, not a quote, and not a Damascode result.

Assume an industrial distributor. The system that takes and invoices orders is maintained by one external developer. Diligence has already established three facts: the target does not hold the repository, no second person has deployed in the past year, and the developer's contract is silent on assignment of intellectual property. Counsel has been asked to resolve the IP point. The technical estimate covers the engineering work only.

Immediate remediation, to regain control. Read-only review, transfer of source and deployment secrets into a company-controlled store, a written deploy and rollback path, and one supervised release. Assumption: 15 to 25 person-days of a senior engineer who can work in the stack. Assumption used only for this arithmetic: an external cost of €800 to €1,200 per person-day. The product is €12,000 to €30,000. Internal time for the operations manager who must learn the release: 5 to 10 days, left as time, not converted into a second currency figure. Confidence is low until the repository is actually transferred and a release is rehearsed. If the code on the laptop and the code in production differ, the top of the range moves.

Required modernisation, only if the thesis needs it. Suppose the investment case adds a second sales channel that the current design cannot isolate, and an upgrade path short of structural change has been ruled out by reading the code, not by taste. Assumption: 80 to 140 person-days to separate the order API and put automated checks around pricing and invoicing. At the same illustrative day-rate, that is €64,000 to €168,000 of external delivery cost. If the second channel is upside rather than the base case, this band stays out of the price and inside the value-creation options.

Integration expenditure. Not estimated. A second company has not been opened. Inventing an ERP integration cost here would hide the missing input, which is the other system's data model and contract.

Recurring operating cost. If, after handover, the business still needs the external engineer for 3 to 6 days a month, that is 36 to 72 days a year, or €28,800 to €86,400 a year at the same illustrative rate. Hosting is excluded until invoices are read. This recurring band is the piece that can affect a multi-year earnings view. The one-off handover cannot, unless the buyer capitalises nothing and also spreads the project cost in a way that misstates cash.

Contingent risk. If the developer becomes unavailable before handover, the buyer may be unable to change or even run the system, depending on what the restore test and the secret store show. No euro amount is assigned. The commercial team still has to say what a week of manual order-taking would do to revenue. That input is not a technical output.

Longer-term investment. A replacement of the order system might be attractive in year three. It is not in this estimate.

The honest presentation in the memo is a band, a confidence label, and the assumptions on one page. A single figure such as "€250,000 of technical debt" would mix a handover, a hypothetical rebuild, a retainer, and an unpriced outage into a number nobody could audit. Ranges are wider before access and narrower after a rehearsal. False precision is a way of pretending the rehearsal already happened.

When a finding may influence the deal

A technical finding does not automatically change enterprise value or EBITDA. It can influence the deal in specific circumstances.

  • Valuation. When the buyer can connect the finding to a cash flow the model is missing: a recurring cost that continues through the hold, or a one-off spend inside the hold that the seller's forecast does not contain. The adjustment should use the same confidence label as the estimate. A contingent event that has no measured probability is a reason to refuse a precise NPV, not a reason to invent one.
  • Deal structure. When the fact cannot be tested before signing. Untested recovery of the system that takes cash, or IP that counsel cannot clear, can support an escrow, a specific indemnity, or a condition. The technical work defines the trigger. Counsel drafts the protection.
  • Working capital and completion accounts. When prepaid licences, deferred hosting, or annual commitments change cash at closing. Read the contracts and the balance sheet together. Do not assume a technology issue is a working-capital issue.
  • Integration budget. The one-off bands for remediation and for the integration the thesis actually requires.
  • Post-acquisition investment plan. Modernisation and optional investment, with an owner and a date. Work that has no owner is not a plan.

PwC's integration-spend figures are a reminder that experienced buyers fund integration as a share of the deal, not as a rounding error. They are not a formula for the technology line. The technology line is the sum of the bands you can evidence.

After the acquisition

The post-close plan should be a continuation of the findings, or it will be replaced by a generic programme that ignores them.

The first 30 days, tied to control. For every critical system flagged in diligence, confirm that the buyer can reach source, infrastructure, identity administration, and the vendor contract. Remove shared administrator accounts where the business can tolerate it, and record the ones that must wait. If recovery was untested, run the restore on an isolated copy in this window, before any celebration of the architecture. If a single supplier holds production, start the handover while that supplier is still contracted. Freeze changes that are not required for operations or for the security items already rated urgent. Name one person on the buyer's side who accepts the list.

The first 100 days, tied to the thesis. Decide keep, integrate, or replace for each critical application, in writing, against the deal thesis rather than against a preference for new tools. If reporting was manual, produce one management pack from the sources diligence mapped, with the reconciliations visible, even if the pack is still partly manual. That pack tells you whether the 12-month data work is a feed, a model, or a new system. If cloud spend had no owner, finish the allocation far enough to separate production, non-production, and unowned cost, and only then decide whether a reduction is real. If a vulnerability assessment left critical items open on internet-facing systems, close or formally accept them in this window. Integration work that does not serve a decision the thesis named waits.

The 12-month technology roadmap, tied to what the first 100 days confirmed. Fund replacement of a legacy platform only if the 100-day evidence still shows that the base case needs it. Sequence security remediation that has an owner and a residual risk the business accepts in the meantime. Set architecture rules before the next acquisition: which system is the system of record for customers, products, and orders, and what an add-on must be able to exchange on day one. A buy-and-build that skips this step collects another ERP and another spreadsheet. The 12-month plan should also say which recurring costs are now understood well enough to sit in the budget, and which estimates remain low confidence.

Future acquisitions reuse the same evidence standard. A group that has one reconciled pack and one identity model can diligence the next target against a real destination. A group that has not done that work will repeat the first integration, with less patience from the investment committee.

Assessment and implementation

The assessment answers what is true and what it costs under stated assumptions. Implementation is a different contract: access handover, a restore test, an integration feed, a security fix, or a software change in the systems the thesis depends on. Buyers get into trouble when the report recommends a programme and nobody is retained to do the first month of it, and also when the same unsigned assumptions are used both to cut the price and to commission a rebuild.

Damascode's published work is IT consulting, including technical due diligence, architecture assessment, security and access reviews, and engineering-process advice, together with software development, data engineering, cloud and DevOps engineering, cybersecurity, and AI implementation advisory. The useful distinction is that assessment and hands-on engineering can be done in one place when the client wants execution after the finding, with the scope of each written separately. Damascode was established in 2019 and builds and operates its own software, including Syndara. That is evidence of running a production product. It is not a private equity track record, and this article does not imply one. Procurement material on delivery is on the delivery evidence page, and the company background is on About.

FAQ

What should technology due diligence decide in an acquisition?

It should decide whether the target's systems, data, security, costs, and engineering continuity support the deal thesis, and what spend or contractual protection is required if they do not. A list of technical observations without that link is not yet diligence.

How is this different from a software architecture audit?

An architecture audit inspects debt, scalability, security, and maintainability in depth. Acquisition diligence uses that kind of evidence only where it changes price, structure, the integration budget, or the first-year plan. The audit method is described in the software architecture audit.

What belongs before signing, and what belongs in confirmatory diligence?

Before signing, obtain the facts that would change the decision to proceed or the protections you need: critical-system ownership, IP and assignment, end-of-life of core platforms, open critical vulnerabilities, and whether a restore has ever been proven. Confirmatory work is the sampling that needs broader access: a traced month of revenue, a supervised look at deployment, a restore onto an isolated copy.

Does a technical finding change enterprise value or EBITDA?

Only when it changes a cash flow you can describe. A recurring cost through the hold, or a one-off spend the forecast omitted, can support an adjustment. An untested backup, a disliked framework, or a pilot AI project does not, until the cash effect is evidenced. Contingent events belong in structure or in a risk note, not in a false NPV.

How should a buyer estimate the cost of a technology issue?

Split immediate remediation, required modernisation, integration, recurring cost, contingent risk, and optional investment. Use ranges, name the assumptions, and lower the confidence label until a repository, a contract, or a rehearsal has been seen. Keep one-off project cost separate from recurring spend.

Which security evidence matters most before signing?

Who has privileged access, whether a restore of the systems that take cash or hold sensitive data has succeeded, and whether known exploited vulnerabilities are present on internet-facing products and still open. A technical review does not establish legal compliance or a certification.

Can the same firm assess the technology and later do the engineering?

Yes, if the assessment scope and the implementation scope are written separately and the buyer can take the findings elsewhere. The assessment should be usable without hiring the same firm to build.

References

Contact

For a technical discussion about an acquisition, technology risk or the engineering challenges of a portfolio company, you can contact Felipe Damasceno directly at fdamasceno@damascode.pt.

Discuss technology risk on a transaction

Describe the deal thesis, the systems in scope and the decision you need. We will suggest a bounded technical assessment. Implementation stays a separate scope.