AI training for operations, finance and managers

In last week’s newsletter we set out some light-touch training examples for sales, marketing, and customer support. This week we look at ways to activate and elevate AI for operations, finance, and managers of all stripes.

Operations

Why this lands differently

Operations is where much of the actual workflow redesign happens, and it is often the function that gets trained on the wrong thing. Give an ops team AI training and they will come out using it for day-to-day efficiency gains like meeting summaries. That has real value, but it is like using Excel only for sorting.

AI also exposes a documentation problem. Most mid-market processes are half-documented, and the ‘truth’ lives in the heads of people who have run them for years. AI can only work with the written half, which means the gaps become visible fast.

Practices

The handoff test. Have each person document one process they own using AI, then hand it to a colleague who does not do that job and watch them try to follow it. Everywhere they stall is tacit knowledge that was never written down. This is the single highest-value exercise in the whole curriculum, because it produces both a trained employee and an asset.

Redesign one workflow assuming AI does the first draft of every step. 90 minutes, one process, and a whiteboard. Most teams discover the bottleneck was never drafting. It was waiting for approvals or chasing an input from another department.

Measurement

Cycle time on one named process, with a baseline established before training starts. Pick one process, measure it for two weeks, then train.

Finance

Why this lands differently

Finance is the only function where the failure mode is a plausible number. A fabricated figure in a variance commentary (the explanation of why actuals differed from budget) looks correct and can travel a long way before anyone catches it.

That inverts the emphasis of training. Everywhere else you are teaching how to use AI for generation with a review attached. Here you are teaching review as the primary skill, with generation as the easy part.

The second issue is data. Finance holds the material the company can least afford to leak.

Practices

The planted-error drill. Give the team an AI-generated variance commentary containing one fabricated figure and one correct-but-misleading framing. Score on catching both. Run it quarterly with fresh errors, because the skill and the urgency decay with every day that things run smoothly.

Set review tiers explicitly. Work collectively to sort outputs into three buckets. Bucket one is internal working analysis. Bucket two is anything going to a department head. Bucket three is anything leaving the building or reaching the board. Write it down and name owners. Without tiers, either everything gets reviewed and the AI is not useful, or nothing does.

Reconcile-then-generate, never the reverse. Train people to verify the underlying figures first and use AI only for the narrative on top. The temptation is to hand it raw data and ask for the story, which is exactly where fabrication enters.

Write the prohibition list as a team. What data never goes into which tool. Ledger extracts, unreleased results, anything with customer names or contract terms. Teams that write their own list follow it. Teams handed one from above do not.

Measurement

Catch rate on the planted-error drill, tracked quarter on quarter. If it is not improving, the review tiers are not being applied.

Managers, across all departments

Why this group matters most

This is the audience most curricula skip. And most managers skip the training that does exist. But they are the ones who determine whether any of the above sticks.

The mechanism is simple. If a manager cannot tell whether a report they received was reviewed or rubber-stamped, they cannot enforce review. Their team learns this in a couple of weeks. Every quality standard you set at department level is enforced, or not, by one person reading the output.

There is also a workload point managers consistently miss. BCG’s 2026 AI at Work survey of 11,749 workers found 47% now spend more time managing and directing AI than doing the work themselves. If managers plan capacity as though AI removed work rather than changing its shape, they will set expectations their teams cannot meet and will not understand why.

Practices

The blind-review exercise. Give each manager six pieces of work from their own team’s recent output: some AI-assisted and carefully reviewed, some waved through. Ask them to sort. Most wildly overestimate their ability to spot AI. Discovering that is the point. It turns ‘how should I know?’ into a practical question rather than an abstract one.

Redesign one of your team’s workflows. 90 minutes, the same exercise as operations, but scoped to work they personally own. Managers who have never done this default to approving AI use case by case, which produces tool adoption without process change.

Rehearse the replacement conversation. Every manager will get some version of ‘am I being automated out of this job?’ within the year. Have them practise it on each other, out loud, with a colleague pushing back. Unrehearsed, most managers reach for reassurance they cannot guarantee, which costs them credibility exactly when they need it.

Measurement

Proportion of managers who have redesigned at least one workflow they own. It is a blunt number, but it separates tool adoption from process change.