My AI workflow cut my governance and reporting workload nearly in half on a 465-application enterprise transition. It also missed the one risk that could have derailed a migration wave — because that risk never showed up as data. Here's what actually changed, and what still can't be automated.  

A few months into leading a large-scale enterprise transition — 465 applications, 8 technology towers, 50+ stakeholders — I hit a moment that's stuck with me. Heading into a monthly operational review, the RAID log was updated, the KPI dashboard was current, the governance pack was ready. Everything was "on track" on paper. What actually decided how the meeting went was a 10-minute conversation

beforehand with one quietly unhappy stakeholder, before he said it in front of the room. A plan tells you what needs to happen. It never tells you why it matters or how a room full of people with different agendas gets there. AI hasn't changed that. It's sharpened it — by taking over the work that used to eat the time I needed for the part it can't do.

Where AI Earned Its Place

On that same programme, delivered with zero critical service disruption:

• Governance documentation (RAID summaries, executive packs) — drafting time cut roughly in half using Microsoft Copilot; still reviewed by a human before it went out
• Meeting summaries & action items — captured and shared same-day instead of reconstructed 48 hours later, cutting "wait,
• did we agree on that?" churn
• Recurring operational reporting — automation reduced manual effort by ~25%, time redirected straight into stakeholder conversations

None of it changed what I was accountable for. It changed how much of my week was available for it.

Where AI Would Have Gotten It Wrong

A cross-tower dependency looked low-priority on paper — small scope, low stated impact, buried mid-way down the RAID log. Any scoring model trained on those fields would have ranked it accordingly. What the data couldn't see: over about three weeks, the vendor lead on the other side went from replying within the hour to taking two days, started sending one-line updates where he used to flag issues proactively, and stopped volunteering anything on status calls unless asked directly. None of that shows up as a metric. It shows up as a feeling in the room that something's off.

That pattern got escalated early through a direct conversation, not a dashboard flag — and it turned out the vendor was quietly short-staffed and heading toward missing a hard dependency date. Caught two weeks out, it was a resourcing conversation. Left to the data alone, it would have surfaced as a missed milestone in the middle of a migration wave — a far more expensive place to find out.

What AI Took Off My Plate — and What It Didn't

• Removing blockers — AI surfaces stalled dependencies fast. Resolving them still takes someone who can broker a decision between two teams that disagree whose problem it is.
• Aligning stakeholders — AI drafts the update. It doesn't hear the silence on a call, or know who needs a side conversation before the meeting.
• Managing uncertainty — AI is strong on historical pattern recognition. It's blind to what never makes it into structured data — tone, disengagement, politics.
• Creating clarity — AI turns messy updates into a clean summary in seconds. It can't decide what the story should be, or repeat it enough that a team believes it.

The shift that mattered most in my own approach: moving from "are we completing the tasks?" to "are we delivering the outcome that matters?" AI didn't make that shift for me — it bought back the time to actually ask the second question instead of drowning in the first.