Ready data, not more pilots: what separates the 28% from the rest

Speed shows up first, revenue shows up later, and the companies that report revenue also report broader benefits. Eight points of movement in nine months says the payoff is real, and a 28% rate says most firms are still waiting. Plan the measurement window before the pilot, not after.

Short post or one-slide graphic: "Three clocks." Speed arrives in weeks, cost savings in quarters, revenue in years. Ask readers which clock their board is reading.

The asset-management numbers are the strongest of any sector KPMG surveyed, and also the most self-reported and the most skewed to very large firms. For a mid-sized wealth firm the useful read is the direction: agent deployment rose from 19% to 30% in one quarter while piloting fell from 52% to 34%. Firms are moving from trying to running.

Post: "Reading a $179 million average." Explain why averages from billion-dollar firms are a ceiling, not a forecast, and list the three questions a smaller firm should ask of any benchmark: who was surveyed, who paid for it, and what counts as value.

A total tells you what you spent. A unit cost tells you whether it was worth it. Until a workflow's cost can be set against the value it creates, a CFO cannot tell a good pilot from an expensive one, and neither can you.

Workshop exercise: pick one live AI workflow, write its cost per run and the value of one successful run on a single line. If either number is a guess, mark it as a guess and decide how to replace it.

Even the firms spending billions rarely put an AI return in a number that regulators and analysts can hold them to. Two banks have now built AI savings into formal multi-year targets. The practical lesson for a smaller firm: a number you are willing to be held to is a different thing from a number in a slide.

Post: "The number nobody says out loud." Contrast what firms say (hours saved, users) with what one bank said (cost and benefit), then ask: what is the one AI number you would put in front of your board?

Same models, same budgets, a 6.5-fold difference in the share reporting real returns, and the dividing line is data readiness. Companies are buying AI faster than they are preparing what it runs on. The cheapest ROI lever in this survey is not a new tool; it is knowing how ready your own inputs are.

LinkedIn post or carousel: "81 versus 12." Three slides: the gap, the one-hour data-readiness check (where it lives, who owns it, when it was last verified), and the question to ask before the next pilot starts.

Most firms bolted AI onto the old process and got a faster old process. The value shows up when roles, approvals and decision rights change with it. If the approval chain is unchanged, the time saved is just waiting somewhere else.

Workshop exercise: "Where does the saved hour go?" Trace one AI-assisted task end to end and mark every approval that still needs a person. Count how many were removed.

This is the data-readiness gap on a small scale. Nearly half the list would have failed before any AI touched it. We measure the input quality now and the outcome later: reply rate and meetings booked per approach are still being collected, so we are not claiming a return yet.

Build-in-public post: state the 55%, say what it would have cost to find out after sending, and commit to reporting reply rate and meetings booked once there is enough data.

The regulated firms are tightening while the broader market loosens, and that is the right instinct. Before an agent touches client data, three things need a named answer: who owns it, who monitors it, and who can switch it off. If a firm cannot answer all three, it has a pilot, not a production system.

Carousel: "Owner, monitor, off-switch." One slide per question, then a closing slide listing which AI uses in your firm currently have all three named.

The repeatable part is not the tool; it is organized data in a place the client already controls. That is also the data-readiness answer to this week's survey. No return to report yet, and we will not claim one until the client's own measures exist.

Short explainer: "Why your AI should start with your folders." Show a messy document library, the same library organized, and the first question an executive could ask of each.