AI and Business Valuation: Underwrite the Cash, Rights, and Risk
How a buyer should distinguish a transferable AI capability from a persuasive story—and decide what belongs in earnings, the forecast, or buyer-created value.
AI changes what a buyer should pay when it changes sustainable cash generation, required investment, competitive durability, or the ability to transfer and operate the business. A demonstration or an “AI-powered” label does not by itself establish an acquisition premium.
The useful unit of analysis is the revenue stream and the workflow behind it. Ask what customers pay for, what work produces the result, what AI changes, who keeps the economic benefit, and whether this buyer can reproduce it after close.
AI Business Valuation Underwriting Map
Decision: does an AI capability or exposure change standalone value, buyer-created value, or the willingness to acquire? Record the economic mechanism, source evidence, rights, costs, downside, and receiving decision for every material claim.
| Claim | Evidence that matters | Valuation treatment |
|---|---|---|
| Higher revenue quality | Paid adoption, renewal cohorts, conversion, service quality, and incremental contribution after selling and support costs. | Model supported revenue and margin. Avoid counting the same improvement in retention and new sales. |
| Lower delivery cost | Before/after cost per accepted output; volume and mix; review, rework, failures, model and vendor expense. | Include net recurring savings once. Hours saved become cash only when staffing, outsourcing, overtime, or profitable capacity changes. |
| More scalable operations | Peak-load execution, exception rates, customer acceptance, remaining bottlenecks, and funded implementation capacity. | Change the relevant operating drivers and investment needs. More throughput has value only with demand and acceptable economics. |
| A proprietary data or AI asset | Ownership, lawful permitted use, customer permissions, exclusivity, quality, portability, contracts, and commercial demand. | Value the usable economic capability. Data possession and data volume do not establish transferable rights or scarcity. |
| AI erodes the business | Observed pricing pressure, churn, lost bids, substitution, changing customer behavior, and the cost of adapting. | Build downside revenue, margin, reinvestment, and terminal durability scenarios. Sector labels alone do not establish obsolescence. |
| Buyer can unlock AI synergies | Specific buyer assets, delivery owner, cost, timing, constraints, integration plan, and failure case. | Keep it in buyer-created value until delivered. Use the Synergy Underwriting & Value Bridge to make seller sharing explicit. |
AI capability versus AI narrative
A useful capability has a repeatable result in an identifiable workflow, at an observable cost, with acceptable failure rates and usable rights. Ask someone other than the founder to operate it, including an exception and an outage. A polished demonstration, uncosted pilot, or dependence on one person’s private account belongs in the evidence gaps.
A transferable third-party tool can improve a business without creating a proprietary moat. Conversely, a specialized dataset or workflow may be valuable only if contracts, permissions, staff, and customer demand survive the acquisition. Signal vs. Narrative tests the claimed result; Dependency vs. Leverage tests who controls the conditions required to keep producing it.
AI Earnings and Investment Bridge
Separate current earnings, future improvement, and buyer synergy
Scenario recurring EBITDA = normalized EBITDA + incremental cash savings + incremental revenue contribution − incremental recurring AI and oversight costs − contribution lost from substitution or repricing.
Exclude improvements already in normalized earnings from the incremental lines. Show adoption timing, one-time implementation, maintenance investment, and working capital separately in the cash forecast. If an improvement requires this buyer’s distribution, data, systems, or capital, identify it as buyer-created value.
| Input or result | Evidence case | Downside case |
|---|---|---|
| Normalized EBITDA | $1,000,000 | $1,000,000 |
| Incremental realizable cash savings | +$120,000 | +$60,000 |
| Incremental recurring AI/vendor expense | −$45,000 | −$55,000 |
| Incremental recurring oversight and rework | −$25,000 | −$35,000 |
| Contribution lost through repricing/substitution | −$60,000 | −$150,000 |
| Scenario recurring EBITDA | $990,000 | $820,000 |
| Gross value at assumed 5.0× EBITDA | $4,950,000 | $4,100,000 |
| Separate immediate implementation cash | −$100,000 | −$150,000 |
| Value less implementation cash | $4,850,000 | $3,950,000 |
Both cases hold the hypothetical multiple constant to isolate the economics. The starting value would be $5,000,000 on $1,000,000 at 5.0×. Even the evidence case produces no automatic AI premium: cost savings coexist with vendor costs, oversight, and pricing pressure. The downside is a separate scenario, not a forecast probability or an additional deduction from the evidence case.
This simplified bridge assumes immediate recurring run-rate effects and immediate implementation spending. For an actual acquisition, phase adoption and disruption in the forecast, bridge EBITDA to after-tax cash flow, include capital and working-capital needs, and reconcile debt, cash, and deal adjustments. A separate change to the multiple, discount rate, or terminal value must reflect remaining risk rather than repeat a loss already modeled.
Who owns the improvement?
- Already operating and in the financials: validate sustainability and transfer. It is already part of earnings; do not add it again as an AI synergy.
- Seller capability, not yet economically demonstrated: model adoption, cost, and failure scenarios. Put unsupported upside outside the base case.
- Improvement that requires the buyer’s platform: underwrite it separately as buyer-created value. Affordability and willingness to share it with the seller are distinct decisions.
- Capability that cannot be used after close: a rights or operating gap may prevent value capture altogether. A price concession does not fix missing permission.
Diligence questions that can change price or stop the deal
- Can the result be reproduced? Observe representative live work, cost it, and compare it with the prior workflow using consistent volume, mix, and acceptance standards.
- Can the buyer use it? Check model/vendor contracts, data and IP ownership, permitted uses, customer terms, credentials, and change-of-control provisions with the relevant advisers.
- What happens when it fails? Test errors, outages, vendor replacement, security and data-handling constraints, review capacity, and manual fallback cost.
- Who keeps the savings? Examine contract pricing, renewal negotiations, competitive response, and whether customers can capture the benefit.
- What still requires the founder? Map prompts, exception judgment, client assurance, and undocumented integrations. AI may concentrate founder dependency instead of removing it.
- What would defeat the thesis? Name the result—such as unacceptable error economics, rights that do not transfer, or faster pricing erosion—that changes the bid, the operating plan, or the decision to acquire.
Classify material gaps as Investigate, Price, Protect, or Walk. Record an owner, required proof, and deadline. Do not advance a critical gate solely because the investment presentation calls the target an “AI company.”
Bring the analysis into the buyer’s decision
Use Follow the Work for workflow and rights evidence. Review peer economics with the Comps Workbench; do not import a technology multiple simply because a target uses AI.
Carry the scenarios into Deal Affordability, separate synergies in the Value Bridge, and state the disconfirmers and price conditions in the IC memo. Send the operating commitments to After the Deal. The Scarcity vs. Growth test helps distinguish a purchasable tool from a hard-to-replicate capability the buyer can actually control.