Don't Hand AI a Broken Process: Why Operational Transformation Comes First

Every leadership team I meet right now is under pressure to "do something with AI." The usual response is to buy licenses, launch a few pilots, and ask every department to find use cases.

Six months later, the results are thin. People are using the tools, but the business isn't running any better. After 25+ years leading technology and operations, I can usually point to why: the AI was handed a broken process.

Automation doesn't fix a bad workflow. It speeds it up. If your order-to-cash process has five handoffs, three spreadsheets and a manual approval no one can explain, AI will help you run those same steps faster and repeat the same errors more often.

The Data Says the Same Thing

• •Employees are acting, not adopting. A Visier survey reported by CIO.com this week found that 45% of employees feel pressure to use AI tools they aren't confident with. 48% admit they overstate their AI skills, and 59% aren't sure their company has a clear AI plan.
• •Redesign is where the value is. BCG's 2026 research found that most organizations add AI without changing how work actually runs. Its early deployments produced productivity gains of only 10% to 20%. Companies that redesigned processes end to end saw cost reductions of up to 60%.
• •Accountability doesn't move to the machine. As one researcher put it in the CIO.com article, the person using the AI is responsible, not the AI. If no one clearly owns a process today, AI won't fix that.

Why Companies Skip This Step

I see the same three reasons again and again:

• •Process work isn't exciting. Mapping a workflow doesn't make a good board slide. An AI pilot does.
• •No one owns the process end to end. Sales owns one piece, finance owns another, and IT owns the systems in between. The gaps belong to no one.
• •The workarounds are invisible. Teams have spent years building spreadsheets and side channels to get the work done. Leadership often doesn't know they exist until someone tries to automate the "official" process.

What Operational Transformation Looks Like Before AI

You don't need a two-year program. You need a disciplined first pass on the processes that matter most.

• •Pick 2–3 value streams that move the business: order-to-cash, procure-to-pay, customer onboarding, service delivery or month-end close.
• •Map how the work really happens, including the workarounds, not how the process document says it happens.
• •Measure the baseline: cycle time, error and rework rates, handoffs, and cost per transaction. Without a baseline, you can't prove AI did anything.
• •Remove before you automate. Cut approvals that add no value, combine handoffs, and standardize the data. Many processes get 20–30% better before any technology is added.
• •Name one owner per process, with authority across departments and accountability for the result.
• •Only then bring in AI, aimed at the specific steps where it will move a measured number, with clear stop criteria if it doesn't.

Where AI Fits Once the Process Is Clean

With a redesigned process, AI stops being an experiment and becomes a tool for a specific job:

• •Intake and classification: reading emails, forms and documents, and routing them to the right place.
• •Exceptions: flagging the 10% of transactions that need a person, so people stop reviewing the other 90%.
• •Decision support: giving process owners real-time data instead of a monthly report.
• •Your ERP and CRM: turning on the AI features you already pay for in SAP, Oracle, Microsoft, Salesforce or NetSuite, now that they have clean, consistent data to work with.

What to Do Next: An Operations-First AI Checklist

• ☐List the 3 processes that most affect revenue, cost or customer experience
• ☐Assign one accountable owner to each
• ☐Map each process as it actually runs today, workarounds included
• ☐Capture baseline metrics (cycle time, errors, cost, handoffs)
• ☐Remove or simplify steps before adding any automation
• ☐Choose AI use cases tied to a specific step and a measured outcome
• ☐Give employees training and clear guidelines, not usage quotas
• ☐Review the results in 90 days and scale only what moved the numbers

Related Reading