Do AI Business Valuation Tools Work? What They Miss for Brokers

AI business valuation tools can work as first-pass calculators and document checkers, but they cannot establish a defensible selling price from uploaded accounts alone. A broker must verify maintainable earnings, normalise owner adjustments, define what is included in the sale and test comparable deals. Use an automated range to spot questions, then apply professional judgement before quoting a seller.

“AI valuation” can mean a form that multiplies earnings, a spreadsheet with suggested assumptions, a document reader that extracts figures, or a model trained on transactions. Their outputs look similar, but the evidence behind them is not. Ask what the tool actually calculates before deciding where it belongs in your process.

Follow me on Instagram@sagnikteaches

Identify the model behind the number

Start with the tool’s input list. Does it need revenue, profit, seller’s discretionary earnings, adjusted EBITDA, debt, inventory, working capital and a sector? Which of those figures does it calculate from documents, and which does it simply accept from the seller? A polished result screen does not establish that a figure was audited or even read correctly.

Connect on LinkedInSagnik Bhattacharya

Request a method note: valuation basis, earnings measure, period, multiple source, assumed cash and debt, treatment of stock, and whether the result is an enterprise value, an equity value or an indicative asking range. Define those labels for a seller. Enterprise value describes the operating business before certain financing adjustments; the cash to a seller also depends on debt, working capital, deal expenses and agreed terms. Two identical displayed numbers can therefore mean different things.

Subscribe on YouTube@sagnikteaches

For a very small owner-run firm, seller’s discretionary earnings (SDE) is often discussed because owner pay and certain benefits may be recast to reflect the economic benefit of ownership. The International Business Brokers Association’s glossary treats SDE as another name for discretionary earnings: earnings before income tax, non-operating and non-recurring items, depreciation, amortisation, interest and one owner’s entire compensation, including benefits and personal expenses run through the business. A tool that accepts a single “profit” field without asking what the owner took out may be comparing unlike figures. The detailed recast belongs in the SDE and add-backs tutorial.

Some products combine these functions. Evaluate each stage separately. A tool might read statements accurately but apply a generic multiple. Another might use a credible database but not expose enough detail to adjust for lease, customer concentration or deal structure. The broker’s value lies in finding the assumptions that change the answer.

Reconcile earnings before trying any multiple

Ask for at least the period and source of each income figure. A seller may provide management accounts for the latest year, a filed return for the previous year and a current trading summary. They may treat an owner’s salary, one-off equipment purchase or family expense as an add-back. Record evidence for every adjustment and its rationale. The output should distinguish claimed from verified earnings.

An illustrative discrepancy: an extractor reads “revenue 1,240” from a report where the header says “figures in thousands”. Its summary states $1,240 rather than $1,240,000. A sensible check compares extracted figures with a manually entered control total and flags implausible units. Another frequent problem is a negative number shown in brackets that a model reads as positive. These errors can make an otherwise correct formula useless.

Before uploading seller records, check confidentiality and retention terms, access permissions and whether the data will be used beyond the task. If the tool cannot give a suitable answer, test with synthetic figures or keep the analysis in a controlled internal environment. Remove names and identifying customer lists from a first-pass calculation unless those details are genuinely necessary.

A buyer will also test the quality of earnings, not merely the arithmetic. If one customer accounts for 45% of revenue, the earnings may be less transferable. If the owner personally generates the majority of sales, a buyer may need a replacement salary or may expect a transition period. An AI model can flag those facts if supplied, but it cannot verify a relationship or negotiate the risk.

Run one illustrative deal all the way to a range

Illustrative numbers: a business broker reviews a small service company with $1.2 million of revenue, $140,000 of reported operating profit, $90,000 of owner compensation and $20,000 of claimed one-off expenses. A simple SDE candidate is $250,000: $140,000 plus $90,000 plus $20,000. The $20,000 is not automatically valid. The broker checks invoices and finds only $8,000 clearly non-recurring, so the reviewed candidate becomes $238,000.

Suppose a tool suggests a multiple range of 2.0 to 3.0 times SDE. Applied mechanically, that creates a $476,000 to $714,000 range on the reviewed figure. Treat these multiples as scenario inputs, not as a market claim. The broker then learns that two customers supply 50% of revenue, a key contract expires in six months and the premises lease may not be assignable. Those facts could materially reduce buyer demand or change terms. The model’s range is now a question list, not an asking price.

Assume the seller expects $700,000 because a competitor’s listing displays that number. A listing is an asking price, not evidence of a closed transaction. The broker asks what assets and working capital are included and whether the competitor’s owner remains after sale. A buyer might offer $600,000 with an earn-out tied to customer retention, which is not economically identical to $600,000 paid in cash at completion. The broker should present scenarios and uncertainty rather than use the model output to validate the seller’s anchor.

Even the lower part of the simple range needs a defined perimeter. If $35,000 of stock is included in one scenario and added separately in another, the two figures cannot be compared. If the seller keeps cash but the buyer assumes a payable, adjust accordingly. The broker’s worksheet should display “included”, “excluded” and “subject to diligence” beside the headline number.

What the tool misses in a real negotiation

Transferability: Can the buyer keep licences, contracts, supplier terms and the team? A repeatable subscription revenue figure may mask contracts with a change-of-control clause. The tool can prompt for the clause; a professional must read and assess it. A hypothetical design studio with impressive profit but one founder owning every key client relationship deserves a different risk discussion from a studio with documented account handovers.

Capital needs: A profit multiple may ignore equipment replacement, maintenance backlog and normal working capital. An illustrative company has $100,000 in annual profit but needs $70,000 of equipment spending next year. A buyer’s cash-flow view differs from a multiple of last year’s profit. The broker should ask for asset condition, not assume the accounts tell the whole story.

Timing: A tool may give the same value before and after a major customer gives notice if the uploaded accounts have not changed. The broker needs current management information, a customer conversation where appropriate and a dated pricing decision. A range should be tied to its valuation date and data cut-off.

Deal terms: Seller finance, deferred consideration, earn-outs, retained stock and transition support alter the seller’s proceeds and risk. An automated headline can obscure those differences. Write two side-by-side terms sheets rather than comparing only totals.

Evidence quality: A three-year average can conceal a decline or a temporary spike. A broker should inspect monthly revenue, gross margin, payroll, customer mix and cash conversion, then explain which period is representative. The confidential information memorandum workflow deals with presenting verified facts to buyers after this screening stage.

Challenge the tool with three closed deals

Choose three completed transactions the broker knows well: one ordinary, one with a major adjustment and one with unusual terms. Remove confidential identifiers and get permission before using any data in a third-party tool. Recreate only what was known at the time the broker first priced the business. If you feed the final sale price or later diligence findings into the model, the test is circular.

For each case, record the tool’s midpoint and range, the broker’s original range, the eventual consideration and the terms that explain differences. A sample comparison might read: “Deal A: tool $500,000–$650,000; closed at $590,000 cash. Deal B: tool $800,000–$950,000; closed at $720,000 plus a contingent payment after customer renewal. Deal C: tool $300,000–$420,000; closed at $405,000 including inventory.” These figures are illustrative. Deal B exposes a missing retention risk; Deal C shows why the inventory perimeter matters.

Three deals do not prove statistical accuracy. They do expose obvious errors and tell the broker whether the tool asks the right questions. Record how long it took to prepare clean inputs and check the output. A “five-minute valuation” that needs two hours of recasting is a workflow tool, not a five-minute professional result. If the model will be used in another sector, build a new challenge set.

Use a simple error log: wrong extracted number, unsupported adjustment, stale comparable, unclear transaction perimeter, missing risk and misleading wording. A tool that often lands near the closing number for the wrong reasons is dangerous. A tool that produces a broad range and clear questions may be more useful for triage.

Phrase the seller conversation as a screening exercise

Do not email a generated number under a heading that looks like a formal valuation. A safer message is: “Based on the information supplied so far, we have an indicative screening range. We have not verified the add-backs, customer concentration or lease transfer terms. We will revisit the range after reviewing those documents and current buyer evidence.” Tailor the wording to your engagement and professional obligations.

Ask the seller for the next three pieces of evidence that would change the answer. In the worked example they might be invoices supporting the $20,000 claimed add-back, the two largest customer contracts and the lease terms. This makes the model’s uncertainty actionable. It also prevents the seller from treating a provisional upper bound as a promise.

A short AI prompt can assist without making the decision:

Using only the anonymised figures and notes below, list:
1. inputs that are verified versus seller claims;
2. inconsistencies that need a source document;
3. questions likely to change maintainable earnings or deal terms.
Do not calculate a single valuation or invent market multiples.
Label every inference as an inference.
[paste controlled summary]

An illustrative output might flag that a claimed one-off legal expense appears in two consecutive years. That is worth investigating. If it states the expense “must be added back”, reject the conclusion: whether an item is non-recurring and acceptable to buyers depends on evidence and context. A broker can use AI to prioritise diligence, then retain the decision and supporting documents.

Know when a specialist valuation is needed

The International Valuation Standards Council’s July 2025 perspectives paper on AI in valuation says AI can support valuers but cannot replace them, and that professional judgement and scepticism remain essential for a standards-compliant valuation. A broker’s indicative asking-range discussion serves a different purpose from a valuation for litigation, tax, shareholder disputes or financing. Define the purpose before choosing the method or presenting a number. When the engagement requires a formal opinion, refer it to a qualified valuer with the relevant scope.

At the end of a pilot, the broker should be able to answer four questions: what did the tool calculate, which inputs were verified, which commercial facts changed the range and which person approved the seller-facing statement? If any answer is missing, the output belongs in an internal question list. That is a workable role for AI valuation tools: faster extraction and sharper challenges, with the price judgement still earned by evidence.

Ask for the comparable transaction behind each multiple

A model may state “similar companies sell for 2.5 times earnings” without showing a transaction. Ask whether its evidence comes from closed deals, asking prices, public-company data or user-entered estimates. Ask the observation period and whether deal terms are recorded. A listed business may stay unsold for months; a public company trades with liquidity and scale a small owner-run firm lacks. Neither is a direct match to a private local sale.

Build a comparison grid rather than importing a multiple unchanged. For each deal, note sector, revenue, earnings definition, growth, customer concentration, owner role, recurring revenue, assets included, working capital, financing terms and sale date. Mark unknown items. If most columns are unknown, the “comparable” is a clue at best. A broker might use it to frame a conversation about value drivers, while declining to present it as firm pricing evidence.

Consider an illustrative pair of service businesses, each reporting $240,000 of SDE. Business A has documented processes, five managers and no client above 8% of revenue. Business B relies on the founder for the largest three contracts, which together supply 62% of revenue. A model that sees only SDE and sector may return similar ranges. A buyer may not. The broker should identify the evidence needed to establish transferability rather than force a cosmetic adjustment such as “subtract 0.3 from the multiple” without market support.

Ask the tool vendor how often it updates its data and whether a user can inspect the underlying transactions. If a source is confidential or cannot be displayed, request enough methodological detail to judge fit. A model’s prediction might be useful even when individual deals cannot be shown, but the broker should state the limitation and avoid claiming transaction-level support it does not have.

Separate pricing from the seller's desired net proceeds

Many sellers ask, “What will I take home?” An automated enterprise or asset value does not answer that question. Build a separate net-proceeds bridge with the seller and their qualified advisers. Show the indicative consideration, debt repayment, transaction expenses, working-capital adjustment, deferred amounts and any tax estimate supplied by the seller’s adviser. Do not let an AI tool invent tax treatment or silently assume all consideration arrives at completion.

For an illustrative $600,000 headline deal, suppose $100,000 is deferred for a year, $40,000 of debt is repaid at completion and $25,000 of transaction costs are expected. The illustrative cash before tax at completion is $435,000, assuming no other adjustment: $600,000 less $100,000, $40,000 and $25,000. That is not the same as a $600,000 all-cash offer. If the deferred amount depends on future sales, it may never be paid in full. Keep the assumptions visible in each scenario.

AI can draft the bridge table and check arithmetic, but the broker must verify the terms and the seller should obtain tax and legal advice. A frequent error is to subtract debt twice: once because the model reports equity value after debt and again in the proceeds bridge. Label the valuation basis at the top of the worksheet and have a second person review the calculation before a seller meeting.

Put a gate between a lead form and a seller pitch

A free valuation form can attract enquiries. It can also create a false certainty if it immediately displays “your business is worth $1.4 million”. Use an honest first screen: request the year, earnings definition and whether the owner works in the business. Return a wide indicative range only if the data meets minimum quality criteria, and name the three assumptions that dominate it. If the inputs are inconsistent, show a request for clarification instead of a number.

A sample lead arrives with revenue of $900,000, profit of $120,000 and owner pay of $150,000, but says the owner works “zero hours”. The model might add all owner pay to profit and produce an impressive SDE. The broker should ask whether a paid manager already runs the company or whether the form answer is mistaken. The tool should flag this contradiction. It should never treat the seller’s self-description as verified diligence.

Where the lead proceeds, provide a document checklist and a secure route to share records. Do not request full customer lists in the first form. The broker due-diligence checklist explains how to organise the later evidence request. Track how many model-led leads require a material correction after the first call; if the rate is high, tighten the form rather than polishing the result page.

Questions brokers ask about automated estimates

Is a model's estimated value the same as an asking price?

No. An automated range can screen a lead or start an internal discussion. An asking price requires verified financials, a defined transaction perimeter, market evidence, seller objectives and a broker's judgement about buyer demand. If a formal valuation is required, instruct a qualified valuer for the specified purpose and standard.

How many past deals should a broker use to test a tool?

Three closed deals make a useful initial challenge set, not a statistical validation. Choose different sizes and risk profiles, use only information available at the original pricing date, and compare the model's range with the actual terms. Add more deals before relying on a tool for a wider mandate or a different sector.

Should a seller enter accounts into a free valuation site?

Ask what the site stores, shares and uses the information for before uploading confidential accounts. A broker can first test with synthetic figures or a stripped-down summary. The seller should understand whether the output is marketing for a lead-generation service, a simple multiple calculation or a professionally reviewed valuation.

Further reads

• Business Broker CRM With AI: What to Compare Before You Buy — Keep valuation leads and deal records organised.
• Business Broker Marketing: AI Buyer Outreach and Anonymous Listings — Separate price testing from buyer outreach.
• Business Broker NDA Process: Qualify Buyers With AI Screening — Qualify buyers before releasing sensitive material.
• Business Teaser Template With AI: Attract Buyers, Hide the Seller — Create a teaser without exposing the seller.
• Can AI Tell You What to Charge? The Limits of AI Pricing Research — Understand why AI pricing research needs evidence.
• Dental Practice Valuation Explained: Prepare Your Figures With AI — How buyers value a dental practice, and how to use AI to tidy your accounts, add-backs and patient data before a broker or accountant takes over.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: International Valuation Standards Council, Navigating the Rise of AI in Valuation: Opportunities, Risks, and Standards (23 July 2025); International Business Brokers Association, Glossary of Terms Used by Business Brokers and Advisors. Both checked 4 November 2026.