Skip to contentSkip to main content
← Back to all posts

AI SaaS Acquisition Due Diligence: An Evidence-Linked Checklist

Evaluate a small SaaS acquisition with AI while tying revenue, churn, code health, and open risks to real source evidence. Includes a reusable diligence register.

Share by email

A small SaaS business can look attractive in a listing: recurring revenue, a recognizable niche, and a product that appears inexpensive to operate. Then the data room arrives. A payment dashboard shows one number, the seller's spreadsheet shows another, and the analytics report covers a different date range. An AI assistant can summarize all three in seconds. It can also hide the disagreement in a confident paragraph.

The useful role for AI in SaaS acquisition due diligence is to organize evidence, expose unanswered questions, and preserve the reasoning behind a decision. The buyer still verifies the records, evaluates the risks, and decides whether to proceed. This guide provides an evidence register, a practical review sequence, and a reusable decision record for a small software acquisition.

1. Start with the decision, not the listing summary

Write down your acquisition constraints before inspecting the seller's claims. A business that is profitable for an owner-operator may be unsuitable for a buyer who must hire support or rebuild an aging codebase.

Define your minimum acceptable conditions for:

  • Economics: verified recurring revenue, gross margin, payment costs, and owner labor.
  • Retention: active paying accounts, cancellations, refunds, disputes, and concentration.
  • Operations: hosting, security, support obligations, critical dependencies, and deployment ownership.
  • Transferability: source-code rights, intellectual property, vendor contracts, domains, and payment-account migration.
  • Fit: audience, distribution channel, strategic overlap, and integration effort.

Add explicit deal breakers. Keep tentative preferences separate from approved thresholds. This is the first record you want an assistant to retrieve later.

2. Build a source register before asking AI for conclusions

Ask for a claim-and-evidence table, not an acquisition recommendation. Each claim needs a source location, reporting period, owner, and verification status.

Claim to examinePreferable evidenceVerification question
Monthly recurring revenueSubscription-level billing exportWhich customers were actually billable at month-end?
Cash receivedPayment processor payouts and bank recordsWhat cleared after refunds, fees, and disputes?
ChurnAccount-level subscription eventsWhich customers left, paused, or failed payment?
Traffic qualitySource-level analytics and search recordsDoes traffic reach product pages and convert?
Low operating costHosting bills, invoices, and owner timeWhich work is unbilled or performed by the seller?
Product ownershipRepository, licenses, contracts, assignmentsWhat code and accounts legally transfer?

Prefer dated exports over cropped screenshots. Preserve the seller's original documents without rewriting them. Restrict access to financial records, customer information, and credentials under the agreed diligence permissions.

A useful evidence-status vocabulary is seller stated, source obtained, reconciled, disputed, and unavailable. A missing record stays unavailable; it does not become zero.

3. Reconcile revenue before accepting the headline

Recurring revenue, bookings, cash receipts, and profit answer different questions. Do not combine them into one figure.

For each month in the review period:

  1. Reconstruct subscription revenue from active plans and account events, accounting for annual plans and discounts consistently.
  2. Separate new payments from renewals, failed charges, refunds, chargebacks, and one-time services.
  3. Reconcile processor activity to deposits and fees. Identify timing differences instead of forcing exact monthly equality.
  4. Compare the resulting series to the seller's stated MRR and profit-and-loss statements.
  5. Record unexplained variances as open diligence items with a responsible reviewer.

Segment active subscribers by plan, tenure, and recent payment success. A business with many nominal subscriptions but few successful renewals has different economics from one with a smaller reliable base. Avoid presenting projected revenue as realized cash.

For valuations, distinguish seller discretionary earnings, accounting profit, and normalized operating profit. Document any proposed add-backs, including the real replacement cost of work the seller performs. Have a qualified accountant examine material discrepancies before relying on the numbers.

4. Test concentration, distribution, and hidden obligations

Even clean billing data is incomplete without retention and acquisition context.

Examine how much revenue depends on the largest customers, affiliates, marketplaces, or one search channel. Compare analytics trends with signups, trials, conversions, support tickets, and cancellations. Ask whether the seller owns the audience or depends on a distribution relationship that can change after closing.

Inspect recent release history, unresolved incidents, dependency and security updates, backup restoration evidence, data-processing commitments, licensing, and infrastructure access. Confirm whether contractors or former owners retain operational privileges.

Some problems are not obvious in a financial export: undocumented manual onboarding, lifetime subscriptions, promised features, disputed refunds, or critical API keys tied to an individual account. Request evidence of these obligations and the plan for transferring them.

For legal, tax, security, and regulatory obligations, involve the appropriate professionals. An AI summary is a useful index to the evidence, not the final verification of a contract or liability.

5. Use a durable evidence card for each important finding

Keep the source material and your interpretation separate. This suggested Markdown format works in any document repository that preserves stable references:

# DIL-014 — Revenue reconciliation
Company: [candidate identifier]
Claim: [seller's exact statement]
Period: [start and end dates]
Source: [restricted file path or record ID]
Source version: [export timestamp or checksum]
Status: disputed
Observation: [what the primary records show]
Interpretation: [what the difference may mean]
Alternative explanation: [timing, taxes, failed renewals]
Missing evidence: [specific export or account access]
Reviewer: [human owner]
Next review: [date]
Decision impact: [threshold or condition affected]

Use different cards for contradictory sources rather than silently replacing one. If the seller supplies a corrected export, retain the earlier version and explain the change. For broader treatment of conflicting dates and authority, see how to resolve conflicting AI knowledge.

6. Connect AI to only the evidence it may inspect

Create a separate diligence workspace or vault for each target. Keep confidential customer and financial data out of general-purpose shared contexts. Give the assistant read access to an approved source set first; expand permissions only after verifying the boundary.

Phasoric can help organize this work through source-linked notes, Research collections, and decision records. Its MCP documentation describes authenticated access to explicitly mapped, authorized knowledge and tools for source retrieval, confirmed decisions, open questions, and recorded changes. An external AI client needs the correct mapping and Context Scope; a local-only note is not automatically visible to a remote client.

Use Strata to preserve a human-confirmed investment decision and its evidence, Pulse to review persisted changes or unresolved follow-ups where configured, and Forge to prepare proposed follow-up work subject to review. These are complementary tools; they do not replace independent billing, accounting, code, or legal inspections. For why traceable sources matter, read AI knowledge provenance.

7. Run a focused first-pass review

Use this suggested 45-minute triage to decide whether deeper diligence is justified:

  • Minutes 0–10: define deal constraints and inventory available files by period and source.
  • Minutes 10–20: compare the seller's revenue headline with subscription and payout evidence; mark gaps.
  • Minutes 20–30: identify churn, customer concentration, traffic dependence, and support burden.
  • Minutes 30–40: inspect access transfer, code ownership, security, and contractual unknowns.
  • Minutes 40–45: draft a conditional recommendation with evidence links and a request list.

The clock is for triage, not a complete acquisition audit. Stop the review early if critical evidence is absent. Ask the seller for the next specific proof instead of prompting AI to estimate missing values.

8. Test the assistant before relying on its report

Give the assistant six checks using your actual files:

  1. Contradictory MRR: Does it show both numbers and their periods rather than selecting the larger one?
  2. Missing payout: Does it say the cash reconciliation is incomplete?
  3. Expired contract: Does it distinguish an old agreement from one currently effective?
  4. Restricted customer data: Does the assistant respect the active source permissions?
  5. Changed export: Can it identify which conclusion needs a new review?
  6. Unconfirmed decision: Does it preserve a proposed purchase condition as pending, not approved?

A polished narrative that fails any of these tests should not be treated as a dependable diligence report.

9. Close with a decision record, not a vague score

A useful conclusion has one of three states: proceed to deeper diligence, pause pending evidence, or decline. Record the date, buyer constraints, supported findings, unresolved risks, valuation assumptions, and conditions for reconsideration. Assign owners to follow-up requests and link each material claim to its source.

If terms change, create a new decision version instead of rewriting history. The value of persistent AI memory is not that it always recommends the same deal. It is that a future reviewer can see what was known, which uncertainty remained, and why the decision changed.

The best acquisition briefing is not the longest summary. It is the one that makes the next verification step unmistakable.

Frequently asked questions

Can AI verify a SaaS seller's MRR automatically?

AI can organize subscription and payout records, compare reporting periods, and flag mismatches, but the buyer must obtain reliable source exports, reconcile them, and verify material findings. Missing records should remain open questions.

What records should I request first for a small SaaS acquisition?

Start with subscription-level billing events, processor payouts, financial statements, cancellation and refund data, source-level traffic analytics, hosting costs, code and license ownership records, and relevant customer or vendor obligations.

How can Phasoric support an acquisition review?

Use authorized notes and Research collections for source context, Strata for human-confirmed decisions, Pulse for configured change and follow-up review, and Forge for reviewed proposals. Hosted MCP reads only appropriately mapped and permitted sources.