Do not approve another AI line without a baseline

On 6 August, Plug and Play released its 2026 Enterprise AI Strategy Pulse Survey. Amit Patel, Ventures Partner at the firm, summed up the finding in one sentence: "Enterprise AI has crossed the deployment threshold, but not the value threshold."

These are the world's largest enterprises. They deployed. The debate about whether to adopt AI is over in that cohort, and adoption is no longer the interesting question.

The uncomfortable part is what happened after deployment. A large share of these companies, with dedicated data teams and mature finance functions, went live and only then discovered they had no way to prove the spend returned anything. The proof was never captured. The one moment a baseline can be recorded is before a tool goes live, and that moment had passed.

Your renewal calendar does not care about company size. The decision this issue is about is whether any AI line item, new or renewing, gets approved this quarter without a baseline attached.

Half of the companies running AI in production cannot consistently measure whether it returned value. That is the Plug and Play 2026 Enterprise AI Strategy Pulse Survey, released 6 August and reported by Forbes the same day. In the same survey, 74 percent of respondents run at least one AI solution in production. A separate figure covers the narrowest deployments: among companies running AI in a single business function, the stage where attribution should be easiest, roughly three quarters say ROI is too early to measure or is not tracked at all. The two figures are close in size and describe different groups, so keep them apart when you repeat either one.

What the number means in plain language: most of this cohort has finished deploying and is still unable to show a return. The narrowest deployments are measured worst, which is backward, since a single function is the easiest place to isolate cause and effect.

The mechanism, as argued by Sandy Carter in the Forbes piece rather than measured by the survey: a use case that goes live without a pre-deployment metric has nothing left to compare against, so later analysis cannot recover the answer. The gap is created on the day the purchase order is signed.

Scope note: the sample skews to Fortune 500 and Forbes Global 2000 companies, so this describes the top of the market rather than the mid-market this newsletter is written for. The number of respondents is not stated in the reporting and I could not retrieve it from the publisher, so treat the figures as direction for companies below that size. Survey answers come from the leaders themselves.

The line to use in your next executive meeting, scope attached: at the top of the market, half of the companies running AI in production cannot measure whether it worked, per Plug and Play's Enterprise AI Strategy Pulse Survey of August 2026.

"These are Fortune 500 companies. Why would their measurement problem bind a 400-person firm?"

The study makes no claim about companies your size, and the scope note says so plainly. Two things survive the objection anyway. First, the timing constraint is arithmetic rather than survey evidence: a baseline can only be recorded before go-live at any headcount, so the cost of skipping it is identical at 400 employees and 40,000. Second, the size effect runs against you. The surveyed companies have measurement staff most mid-market firms lack, and they still ended up here. Mid-market data would settle the question properly. This dataset does not contain it, and no honest reading pretends otherwise.

Three steps, completable before next Thursday.

Step 1. List every AI line item that starts or renews before the end of Q3: the tool, the annual cost, and the person accountable for its result. This step needs nothing but your invoice list and 20 minutes.

Step 2. Run each item through the baseline gate at

For each one, write down the evidence itself: the metric the tool is supposed to move, that number today with its date, the one person who answers for it, and the review date. The tool reads every gap back to you with its consequence and a move for this week, totals the annual cost sitting on items that cannot prove their return, and turns each completed item into the exact sentence to say at the review. You set approve or hold yourself. Nothing you enter is stored or sent anywhere, and the finished list downloads as a CSV.

Step 3. Take every item you set to hold to its owner this week and have the baseline recorded before the renewal date. Hold is a statement about your evidence, and says nothing about the tool itself.

The final step is for the items where nobody can fill the fields. That is the work of the AI ROI, AEO and Data Activation Assessment: it produces evidence about your AI investments instead of a self-rating of them, with defined KPIs and a decision-ready roadmap, under the Clear-Scope and Implementation Clarity guarantees and milestone-based delivery. Book the fit conversation with John here: meetings.hubspot.com/john2956

Reply to this email with the number of items on your hold list. I read every reply.

Next Thursday: what enterprise buyers now rank above model performance, and what that changes when you pick a vendor.

The monthly AI Citation Report goes deeper than any weekly issue can, with the sources laid out page by page.

Elizabeta Kuzevska Co-Founder, Revenue Experts AI revenueexperts.ai

Sources