74% of companies that put an ai agent in front of customers have already pulled it back out. the headcount they cut to pay for it returns in 2027 under a different title.

sinch surveyed 2,527 senior decision makers across ten countries this winter and found that 74% had already rolled back or shut down a live ai customer communications agent after deploying it, with the rate rising to 81% at the companies with the most mature guardrails. gartner, off its own survey of 321 service leaders, expects half the companies that credited ai for headcount reductions to rehire for the same functions by 2027, under new job titles. read those two together and the shape is obvious: deployment is the easy part, and the part nobody budgets for is the day the agent gets a case it was never scoped for. the cost is not the license. it is the cut you made in month one, the severance, the rollback, the rehire twelve months later at whatever the market charges by then, and the customers who spent the gap talking to something that could not help them and could not say so.

the fake fix is a better model or a bigger platform, on the theory that the last agent failed because it was not smart enough. the ones getting pulled are not failing at language. they are failing at governance: no rule about what the agent may promise, no line where it hands off to a person, nobody reading what it actually said.

do the version that holds: before it goes live, write the three things it may never do without a human, money out, commitments, anything you cannot reverse, then write the trigger that hands the conversation to a person, name who reads its transcripts every week, and keep headcount flat until it has run one full quarter with no rollback. 98% of those same companies are still increasing ai spend this year, so the market has already decided the tool is worth buying. it has not decided you get to cut first.