AI transformation is often framed as a technology-selection problem. In practice, the first design problem is usually the process itself.
If a process contains duplicate work, avoidable hand-offs, ambiguous ownership, inconsistent knowledge or approvals that no longer add value, adding AI can create an illusion of progress while preserving the underlying operating problem.
Why broken processes become more dangerous with AI
Human friction can sometimes hide process defects because people compensate informally: they ask a colleague, correct a bad input, bypass an unnecessary step or recognize an exception. AI may remove that friction without removing the defect.
The result can be faster propagation of bad rules, faster escalation of exceptions and more expensive rework.
The 7-step simplify-before-automate method
- Define the outcome. What customer, employee, financial or risk outcome should improve?
- Map the current journey. Include waiting, rework, hand-offs and decision points—not just system clicks.
- Remove non-value work. Challenge approvals, reports, duplicate data capture and policy relics.
- Standardize what remains. Fix conflicting rules, knowledge and exception paths.
- Clarify ownership. Every decision and hand-off should have an accountable owner.
- Choose the right intervention. AI may be appropriate, but so may policy change, workflow redesign, self-service, analytics or no technology at all.
- Design benefits before implementation. Decide how time saved will become better service, growth, cost reduction or risk reduction.
Where AI is genuinely powerful after redesign
Once unnecessary work is removed and decision logic is clearer, AI can add disproportionate value through knowledge retrieval, prediction, guided decision support, natural-language interaction, intelligent orchestration and autonomous execution.
The point is not to slow down AI adoption. It is to ensure that AI accelerates a process worth accelerating.
The 10-question pre-AI process test
- What business/customer outcome should this process produce?
- Which steps directly contribute to that outcome?
- Which steps exist because of legacy policy, system or organizational boundaries?
- Where does waiting exceed actual work time?
- Where does rework occur and why?
- Which decisions are rule-based, judgment-based or unnecessary?
- Is knowledge consistent enough for AI to use safely?
- Are exceptions understood and measurable?
- Who owns the end-to-end process after automation?
- How will released capacity become real value?
Research and further reading
- Gartner — Why You’re Not Getting AI ROI — And How to Build Compounding AI Value — 13 May 2026. Public abstract: deploying AI into existing workflows often creates isolated projects and marginal gains; core processes and value streams need fundamental redesign.
- BCG — Making AI Productivity Pay Off — 5 May 2026. BCG argues that productivity gains do not automatically become lower cost or better performance; capacity must be deliberately redirected and work redesigned.
- Deloitte — State of AI in the Enterprise 2026 — 2026. Deloitte reports widespread productivity gains but only 34% of organizations truly reimagining the business; only 30% are redesigning key processes around AI.
- PwC — 2026 AI Performance Study — 13 Apr 2026. PwC reports that 75% of AI economic gains are being captured by 20% of companies; leaders are twice as likely to redesign workflows around AI and are more focused on growth, not just productivity.