Five ways to speed up with AI without losing control

Six months ago, most of your people were experimenting with AI on their own. Pasting a report into ChatGPT to summarize it faster, or keeping a private list of prompts nobody else knew about. That's the shadow AI stage. If you've read our piece on it, you know the fix: give people an approved tool and teach them to use it well.

Say you've done exactly that. The tool is approved and people use it in the open. In our experience, what happens next at most companies is surprisingly little.

Nobody's driving

The rollout gets treated as the finish line. IT hands out licenses, runs one training session and moves on to the next project. From that moment on, what people do with the tool is entirely up to them.

Ask ten people on the same team what "using AI responsibly" means and you'll get ten different answers. That may feel like flexibility, but it's really the absence of governance. Everyone is moving faster, and nobody is wearing a seatbelt.

AI takes the wheel, the job moves up

Self-driving cars are a useful way to picture what's happening. Every year they get better at the actual driving. Yet someone still decides where to go, owns the car and keeps it charged and maintained. The job shifts from operating the vehicle to directing and maintaining it.

Your people are going through the same shift. They spend less time drafting a first version or running a routine query by hand. Their job becomes setting the direction, catching the wrong turn and keeping the whole thing in good shape.

Moving faster without losing control means doing that job on purpose. These five tips make the difference.

1. Name one owner

On most teams, nobody owns what AI is for once the tool lands. IT owns the license, but no one owns the outcome. Pick one person close to the work and make them responsible for AI quality. They decide what "good" looks like, so ten people don't quietly decide it differently by default. This doesn't have to be a full-time role. It does have to be a clear one.

2. Make good practice visible

A good prompt or workflow tends to stay locked in one person's head, simply because there is nowhere for it to surface. Give it a place: a channel for "here's the prompt that finally worked", and a short monthly look at how the tool is actually being used. That's how a team copies what works and catches what doesn't, before it becomes the norm.

3. Keep the skill alive

Models drift and prompt libraries go stale. None of that fixes itself.

How someone phrases a prompt or which tool they reach for is craft rather than policy. Turn it into rules and you end up with a locked-down chatbot nobody uses. Leave people alone after one training session and the skill never builds either. So invest in it on purpose: regular practice, shared examples and someone to ask when you're stuck.

4. Check before it ships

A wrong number on a client dashboard gets caught in one way: someone other than the person who made it looks at it before it goes out.

Borrow the discipline finance and audit have relied on for decades and apply four-eyes. Anything with real consequences, such as a client email, a number in a report or anything touching a contract or a claim, gets checked by a second person before it leaves the building. Everything else can move at full speed.

5. Keep the rules to one page

Beyond that review step, resist the urge to write a policy handbook. One page is enough: which data can go into which tool, and how the output gets logged so you can trace it back later.

This is the seatbelt from the opening. You wear it on the assumption that something will eventually go wrong, and you want to be the one who knows exactly what happened when it does.

What this looks like on a real team

Take a BI team that used AI to speed up their report builds, with everyone doing it their own way. They named one owner and added a single rule: any number landing on a client dashboard gets checked against the source by someone other than its author before it's published.

How analysts drafted their DAX or iterated on a prompt stayed entirely up to them. The prompts that worked got shared, so the second analyst didn't reinvent what the first one had already cracked. Reports kept coming out at the same pace, and when a calculation did come out wrong, a second person caught it before a client ever saw it.

The bottom line

AI takes over more of the manual driving every year. The job moves up: pointing the direction, keeping the practice alive and checking the work before it ships.

Do those five things on purpose and you get a team that moves faster without gambling on which week someone happened to be careful.

We go through this exercise with our clients every day, and we know where it tends to get stuck. If you'd like support in streamlining it for your own team, get in touch. We're happy to think along.