Deterministic Valuation: Why the Same Inputs Must Give the Same Number

The question an owner will eventually ask

Every managing director has had this conversation. The bank pitched a range in the engagement meeting. Six weeks later, after discovery, the range moved. The owner wants to know why. If the answer is "the adjusted EBITDA came in $300,000 lower because the add-back for the warehouse lease did not hold up, and here is the lease," the owner may be unhappy but will accept it. If the answer is "the model gave us a different number this time," the mandate is at risk.

The same thing happens with buyers. A sponsor's associate rebuilds the bank's numbers from the data room. If the bank's valuation cannot be reproduced from the same inputs, the buyer discounts everything else in the CIM. Credibility in a sell-side process is cumulative, and the valuation is the first place it is tested.

Determinism is what makes that defense possible. Given the normalized financials, the chosen comparable set, the selected multiple range, and the stated adjustments, the output should be fixed. Change one input and the output changes in a way that can be explained. Change nothing and it does not move. Professional valuation guidance likewise treats clear assumptions, internal consistency, and enough information for a reader to critique the analysis as core qualities of effective valuation work [1].

Why a language model cannot be the engine

Large language models are probabilistic. Ask one to "value this company" with the same prompt twice and you can get two different answers, both plausible, both unsupported. This is not a flaw to be fixed with better prompting. It is a reason to separate probabilistic assistance from the calculation that produces a bank's signed-off number. A model can be useful for drafting prose, summarizing a contract, or suggesting which line items look like add-backs. It is not a substitute for a defined valuation method and controlled inputs.

The mistake some banks make is to treat the model's fluency as rigor. A model will produce a confident paragraph explaining a 6.2x multiple with references to industry norms and growth premiums. The paragraph reads well. It may also be wrong, and more importantly, it is not reproducible. The associate who runs it again the next morning to update one figure may get 5.8x and a different paragraph.

The right architecture separates the two. A deterministic engine takes structured inputs and applies defined logic: comparable multiples by vertical, size and growth adjustments, the normalization schedule, and whatever methodology the bank has chosen. The language model sits beside it, helping the associate assemble inputs and draft the narrative that explains the output. The number comes from the engine. The words come from the model. The judgment comes from the banker. NIST guidance similarly calls for AI models and outputs to be validated, documented, and interpreted in context [2].

What traceability looks like in practice

A deterministic valuation is only useful if someone can follow the trail. Traceability means that every figure in the output has a visible origin: the adjusted EBITDA links to the normalization schedule, each add-back links to the ledger entry or the owner's questionnaire answer that supports it, and the multiple links to the comparable set and the adjustments applied to it.

In practice this changes how a bank reviews valuation work. Today a partner opens an associate's Excel model, clicks through tabs, and hopes the links are not broken. With traceability built in, the partner can start from the headline enterprise value and click backward through every input in a few minutes. Review becomes a check on judgment, did the associate choose the right comparable set, is this add-back defensible, rather than a check on arithmetic. The CFA Institute describes comparable valuation as a relative method that depends on a benchmark multiple and notes that the reader should be able to critique the valuation and its assumptions [1] .

It also changes what happens when the owner disputes a figure. Instead of "let me look into it," the associate can show the owner the exact source: "This $85,000 add-back for your salary is the difference between your W-2 and a market rate for a general manager in your region. Here is the questionnaire answer where you gave us your compensation and here is the market figure we used." The conversation moves from trust to evidence.

Consistency across mandates and across the firm

A boutique bank with three partners often has three ways of valuing a company. Each partner has a preferred methodology, a mental list of comparables, and habits about how aggressively to normalize. That variance is invisible until a buyer who has seen two of the bank's CIMs in a year notices that the same kind of company was valued two different ways.

A deterministic engine applied across the firm's mandates produces consistency. The same vertical uses the same comparable logic. The same type of adjustment is treated the same way. Partners can still override, but the override is explicit and documented rather than baked into one person's spreadsheet. Over time this can build an institutional valuation method that does not depend on any single partner's spreadsheet or memory.

For a managing director, this matters at the pitch stage too. When the bank can say "here is how we value companies in your vertical, here are the inputs we will need from you, and here is why our range is defensible," the pitch sounds different from a competitor's gut-feel range.

What to do this week

Pick the last valuation your bank delivered. Give it to an associate who did not build it, along with the inputs, and ask them to reproduce the headline number without talking to the original author. Time how long it takes and note every place they had to guess. Each guess is a point where the valuation is not deterministic and where an owner or buyer could reasonably challenge it.

Then ask a second question: if one input changed, say the trailing-twelve-month revenue was updated for the latest month, how many cells in how many tabs would need to be touched by hand, and how confident is the bank that none would be missed? The answer tells you how much of the valuation phase is currently held together by one person's attention rather than by a method.

How Acquivo Pro fits

Acquivo Pro's valuation engine is deterministic: the same inputs always produce the same output, and every figure traces to its source in the client questionnaire and data room. The firm runs valuations across 21 industry verticals on one login, which gives partners a consistent method across mandates rather than three personal spreadsheets. Because the same intake data feeds the teaser and CIM, the numbers in the marketing materials are the numbers in the valuation, not a re-keyed copy. Client data is isolated and not used to train models for other customers. To see it on a live mandate, request a demo through the Contact page.