Implementation

A 90-day AI automation roadmap for an operations team.

How to identify the first workflow, ship a controlled pilot, measure it, and build an automation backlog from real evidence.

October 12, 2026 · 8 min read

The short version

One measurable workflow in production creates more strategic clarity than a company-wide list of AI ideas.

Days 1–20: find the operational constraint.

Interview the people doing the work and observe the process directly. Look for repeated handoffs, duplicate entry, long queues, fragmented information, and decisions supported by stable evidence. Record volume, handling time, error rate, wait time, and the system of record.

Prioritize a workflow with meaningful frequency, a visible outcome, manageable access requirements, and a reversible first deployment. Avoid choosing only because a demo is easy.

Days 21–60: ship the smallest reliable system.

Map the happy path and the exception path. Keep deterministic steps deterministic, add model judgment only where it creates value, and design logging and review before launch. Test with historical cases and edge conditions supplied by the team.

—Define success and failure in business terms.
—Connect the minimum data and tools needed for the job.
—Add approvals before consequential actions.
—Launch to a narrow group with a visible fallback process.

Days 61–90: measure, harden, and decide what scales.

Compare the pilot against the baseline. Review exceptions, corrections, latency, adoption, and cost per completed outcome. Improve the process or tool contract before reaching for a larger model or more autonomy.

The next automation should emerge from what the first system revealed: poor source data, a downstream queue, or another repeated handoff. That creates a coherent operating layer instead of a collection of disconnected experiments.

Related analysis

All blogs →

AI strategy

AI agents or workflow automation? Choose the simpler system first.

A practical decision framework for choosing between deterministic automation, AI-assisted workflows, and autonomous agents.

Infrastructure

What MCP changes for business software—and what it does not.

Model Context Protocol can standardize how AI systems connect to tools and data. Here is the business case without the protocol hype.