92% say their customer-facing AI meets expectations, but only 27% can prove it. Only 10% recorded a baseline

Belief in AI is high. Proof is rare, and the gap starts on the day before launch.

What happened

MarTech reported on a Metric Sherpa study produced with Sprinklr. Among organizations with customer-facing AI in regular use, 92% said it met or exceeded expectations, but only 27% could show proof. Almost everyone (99%) could name an intended outcome. Only 26% defined success metrics, and only 10% recorded a baseline before launch.

Cost claims had the same problem. Of those who said AI made work faster, 67% said it also cut costs, yet only 32% of that group tracked costs before launch. Just 17% name an owner for AI business outcomes, and organizations with a named owner quantified results 46% of the time, against 23% without one.

An earlier MarTech piece, citing The Martech Weekly's Enterprise Martech Outlook 2026, found 56% of teams that could prove value got budget increases, compared with 37.5% of those relying on faith.

My take

This matches what I see on client projects. Nobody skips measurement on purpose. It just is not part of the build, so it never happens.

I now treat three things as part of the automation itself, not a report for later:

1. Baseline before go-live. Two weeks of the current numbers: handling time, response time, error rate, cost per ticket or per lead.
2. One owner. A named person who decides to scale, change or stop the workflow.
3. Logging built in. Every run writes a row: started, finished, outcome, whether a human had to step in.

With that in place, the monthly review is a query, not a debate. And as the second study shows, the teams that can prove value are the ones that get more budget.