Six months ago, the easiest way to start an AI conversation across a portfolio was to ask every company for its AI spend. The appeal is obvious. One question, one column, a quick comparison. The problem is that the answer can combine internal experimentation, employee tools, and customer-facing product workloads, even though each represents a different decision and owner.

In 2027, investment firms should standardize the questions and resist standardizing away the exposure. Internal AI spend is an operating expense and governance question. AI embedded in a product is COGS and a margin question. One is not a proxy for the other.

Consistent questions, different answers

Every company can be asked to define its AI workloads, name accountable owners, explain the boundary between experimentation and production, and show how cost is tied to provider invoices. Each should also state the unit that matters to its business and the decision the measurement informs.

What follows will differ. A services firm may care about cost per client engagement. A software company may care about cost per account or feature. A company using internal assistants may need to understand operating expense and governance before it has a customer-facing unit at all. Consistency belongs in the method and disclosure, not in a forced common denominator.

Portfolio prompt: Does this figure describe internal use, customer-facing product cost, or both? If it combines them, what decision can it actually support?

Do not confuse exposure with performance

A high AI cost is not automatically a problem. It can reflect deliberate research, valuable product usage, or a cost structure that is already priced into customer contracts. A low cost is not proof of control. It can reflect low adoption, incomplete capture, or missing infrastructure and human-review costs.

Six months ago, a portfolio ranking might have felt like progress. For 2027, a ranking without context risks making teams optimize for the score instead of the business. Ask what the number includes, what is still unknown, and what action follows. A company that has found a gap and assigned an owner may be in a stronger position than one with a neat total and no reconciliation.

Make the standard useful to the board, VC, and PE

Directors need a clear management owner, decision cadence, and escalation path. Venture teams need to understand whether unit economics improve as the company scales and whether the product can support its financing story. Private-equity teams need a baseline that can enter operating reviews and survive transaction diligence.

These audiences can share a core evidence record without sharing the same conclusion. Separate the two cost pools, state the measurement boundary, and make the company explain how the number will change a decision. That gives each audience a way into the evidence that fits its mandate.

The portfolio question for next year

Ask each company to show not only the current figure, but what changed since the last review and why. Ask what evidence would falsify the explanation. Then ask who owns the next step.

That conversation will be less tidy than a single AI maturity score. It will also be more valuable. In 2027, the useful portfolio standard is not one number applied everywhere. It is a transparent method that keeps separate exposures separate and gives each company a defensible path from measurement to action.