AI Rollbacks, SAP U-Turn.
74% of enterprises rolled back AI customer agents. What that means for accounting firms deploying AI in AP, close, and client workflows.
74% of Enterprises Rolled Back Live AI Agents — Here's Why That Number Should Slow You Down
Sinch research out this week found that nearly three-quarters of enterprises have already shut down or pulled back a live AI customer communications agent after deployment. The failure mode wasn't the AI itself — it was the gap between demo performance and production reality: messy data, edge cases, and zero tolerance for errors in customer-facing workflows. Accounting is not immune to this. AP automation and financial close agents look clean in sandboxes and fall apart on your actual chart of accounts.
A rolled-back AI deployment is worse than no deployment — it burns budget, erodes team trust, and sets your AI timeline back 12 months.The Accountants Who Survive AI Rollbacks Are the Ones Who Designed the Guardrails
When an AI agent fails in a financial workflow, someone has to diagnose it — and that person is rarely an engineer. Firms that trained even one staff accountant to map exception handling, define escalation rules, and audit AI outputs before go-live are recovering faster than firms that handed the whole project to IT. Think of it like internal controls: you wouldn't let a system go live without a control owner, so don't let an AI agent go live without one either. Assign someone on your accounting team to own the failure modes, not just the success metrics.
Designate an AI control owner on your team before you deploy anything — someone who understands the accounting, not just the software.SAP Quietly Reversed Its Cloud-Only AI Stance — On-Prem S/4HANA Firms Just Got Back in the Game
SAP's CEO confirmed this week that AI features previously withheld from ECC and on-premise S/4HANA will now be made available — a direct reversal of the cloud-only innovation strategy that had been pushing firms toward expensive migrations. If your clients are running on-prem SAP and stalled on AI adoption because of that roadblock, that block is gone. This matters most for mid-market manufacturers and distributors who run complex close processes on legacy SAP and couldn't justify a cloud lift just to access AI-assisted reconciliation or journal entry automation.
If you have clients on ECC or on-prem S/4HANA who paused AI conversations because of the cloud mandate, reopen that conversation now.AI Agents in Accounting Won't Be Stopped by Bad Technology — They'll Be Stopped by Bad Change Management
The Sinch rollback data points at something the accounting software vendors won't say out loud: the technology is good enough, but most firms aren't operationally ready to deploy it. The firms treating AI agents like a software install — flip it on, train the team later — are the ones generating the rollback statistics. The firms seeing durable results are treating each AI deployment like a process redesign with a software component, not the other way around. One regional CPA firm I'm aware of spent 60 days remapping their AP exception workflow before touching a single automation tool — their error rate on vendor payments dropped 40% inside 90 days of go-live.
Your AI deployment isn't failing because the model is wrong. It's failing because your process was never clean enough to automate in the first place.