The AI Discount.

Diagnosing exposure before the multiple compresses further.

AI is repricing certain businesses. The owners of those businesses now face a decision — sell into the strategic window, pivot the model first, or hold and accept the rerating — and the decision is sharper in 2026 than in any year before it.

The repricing is uneven. Some businesses are unchanged. A subset are improving as AI lowers their cost base. A meaningful population is compressing — and the compression is showing up in transaction multiples that have moved several turns below the trailing-decade average. Owners of businesses in the third category have a narrowing window in which to act before the market repositions around the new reality.

This note offers a four-part diagnostic for assessing where a business sits, a sector-by-sector exposure map, the three paths available to an exposed seller with their comparative economics, the timing window behind all three, and a stakeholder-by-stakeholder action set for the next four quarters.

Market Context.

Before the diagnostic, four facts about the AI investment environment that frame the rest of this note.

The implication is straightforward. The capital exists to build AI-native competitors at scale. The productivity gains in pilot environments are large enough to support displacement of incumbent delivery models in specific task domains. And the strategic acquirers — both AI-natives and incumbents pivoting — are well-funded enough to underwrite premium consolidation while the rerating completes.

What follows is the diagnostic for assessing where a specific business sits inside this environment, the exposure map across sectors, and the decision framework for the three paths a board can take.

The Diagnostic.

1. The Deliverable Test

Begin with what the business sells. If the deliverable is a document — a memo, a report, a summary, a code module, an image, a transcript, a translation, a brief — and the deliverable is produced from inputs the customer already has, the work is AI-exposed. The exposure is not theoretical. The exposure is already priced into the next acquirer's underwriting.

Businesses that fail the deliverable test today include large segments of legal services, customer-support BPO, transcription, content marketing, low-end accounting, market research, paralegal work, translation, basic coding, and the long tail of professional services where the output is a document and the customer can articulate what they need.

Businesses that pass produce something that is not a document, or that requires inputs only the firm holds. Physical operations, regulated capital, networked distribution, deep proprietary data, decades-long client relationships that compound, and high-judgment work where the deliverable is a decision rather than a document — these survive the deliverable test.

2. The Moat Test

The moat test asks what protects the business if the deliverable test flashes red. The defensible moats in an AI-repricing market are narrow. Brand alone is not one of them. Pricing power alone is not one of them. The moats that hold:

• Regulated capital. Banking, insurance, registered investment advice, broker-dealer activity — the regulatory perimeter is a moat that AI cannot dissolve, only navigate.
• Network effects with switching cost. Marketplaces, platforms, payment rails, exchanges. The moat is the network, not the software.
• Physical operations at scale. Manufacturing, logistics, energy, real estate. AI optimizes them; it does not replace them.
• Proprietary data the customer cannot replicate. Decades of clinical outcomes, proprietary trading data, exclusive distribution rights, government contracts.
• Distribution embedded in customer workflow. ERP integrations, EHR integrations, regulatory filing systems — the switching cost is the moat.

A business that fails the deliverable test and lacks at least one of the above is exposed. The number of acquirers willing to pay the prior multiple shrinks; the underwriting tightens; the diligence lengthens; the price drifts.

3. The Customer Behavior Test

Diagnostics that depend on transaction multiples are lagging. The leading indicator is customer behavior. Three signals to watch:

• RFPs that require AI integration or AI-priced scope. When customers begin specifying AI-augmented delivery in procurement, the floor under the legacy model is moving.
• Renewal cycles compressing. Customers who used to commit to thirty-six-month contracts now commit to twelve or eighteen. The pricing balance is shifting toward the customer.
• AI-native competitors winning RFPs the firm used to win uncontested. The moment the compression becomes visible inside the company's own pipeline.

If two of three are present, the repricing is already underway in the firm's own market — regardless of what the public comparables show.

4. The Multiple-Watch Test

Public and disclosed-private comparables remain useful, but the trend matters more than the absolute level. Pull the trailing six quarters of disclosed transactions in the sector. Compare to the trailing five-year average. If the multiple at the median has compressed by more than 1.5 turns over six quarters, the market has already begun the rerating. The next acquirer will underwrite to the new normal, not the old one.

Sector Exposure Map.

The diagnostic applies at the business level. The map below applies at the sector level and reflects where the firm sees the densest population of red-posture businesses today. The exposure rating is directional, not predictive of any specific company's outcome.

The Three Paths.

For the business that passes the diagnostic into exposure, three paths exist. None is obvious; each carries its own cost.

Sell — into the strategic window

Strategic acquirers are paying premiums to consolidate exposed segments before the rerating completes. The acquirer's logic is: buy now, integrate the customer book onto an AI-native delivery model, capture the margin expansion. The seller's logic is: realize the premium while the strategic buyer is willing to underwrite to the old multiple. The window is real and it is narrowing.

Selling into this window requires three conditions: a credible buyer set that includes both strategics and AI-native consolidators, a clean financial diligence package, and a story the seller can tell about why the buyer's integration thesis works. The third condition is the one that often delays a process — sellers who cannot articulate the buyer's case lose the premium.

Pivot — and sell later

The pivot path is the harder one. It requires rebuilding the delivery model around AI rather than against it, demonstrating the new economics through at least two full pricing cycles, and then selling at the new multiple — which may be lower in absolute terms but higher relative to the post-rerating market. The pivot is real, but the seller needs to be honest about three things: the capital required, the timeline (twenty-four to thirty-six months in most cases), and the execution risk against incumbents that started the pivot earlier.

Hold — and accept the rerating

The hold decision is legitimate when the moat is real, the customer base is sticky, and the founder's time horizon is long. It is also legitimate when the alternative is a sale at a price the founder believes underwrites a transient compression rather than a structural one. The hold path requires intellectual honesty: the rerating may be permanent, and the option to sell at the old multiple may not return.

The Timing Question.

The question behind all three paths is the same: how long is the strategic window open?

Our reading of 2026 is that the window is open for sellers who can move within the next four to six quarters. Strategic acquirers have capital and consolidation theses. AI-native consolidators are well-funded. The public-market discount on AI-exposed businesses has not yet fully priced into the private comparables. Sellers who initiate a process in the first half of 2027 will likely transact, but to a different buyer set and at a different price.

This is not a prediction. It is a diagnosis of the current market posture and the trajectory of the most credible signals.

Implications by Stakeholder.

The diagnostic is the same; the action set is different depending on the stakeholder's position relative to the business.

What This Note Does Not Say.

This note does not say that AI destroys all businesses. It does not. A meaningful set of businesses are unaffected or improved. The framework above is a diagnostic — it tells the reader how to assess where their business sits, not what every business is worth.

It does not say that the right answer for every exposed business is to sell. The pivot path is real for owners with the conviction, the capital, and the time. The hold path is legitimate for owners with genuine moats and long horizons.

Nor does it stake a position on the long-run macro impact of AI on aggregate productivity, employment, or growth. Those are larger questions on which credible economists disagree. This note's scope is narrower: which businesses, today, face a transaction-market repricing that requires a decision in the next four to six quarters.

What it does say is that the question is no longer optional. The boards, founders, and sponsors who run businesses in AI-exposed segments need to have answered the four diagnostics by the end of this year. The answer points to one of three paths. The path determines the timing.

Cohen Capital Markets advises principals, boards, and management teams on the M&A decisions that arise from this diagnostic.