make ai fit the team.
Your team's licensed for Copilot, and so far it hasn't quite paid off - some of it sitting idle, some of it writing DAX nobody can fully vouch for. What pays off is getting genuinely good with AI, without handing over your judgement - so a stretched team does more, and better. If that's roughly where you are, I help you work out what to hand over, what to keep, and who checks what: where it earns its place, and where it costs you more than it saves.
who's checking the dax?
you're not behind
I won't walk in assuming your team's fallen behind on AI, or that the people are the problem. The teams I sit with usually know their own data and models inside out. What they're short of is an unhurried hour to work out which of these tools earns its place, and which is a demo that won't survive contact with their real reports.
how it works
It's a short piece of work, spread over two or three weeks and fitted around how the team already works. It starts with a conversation, and ends with a map you can act on.
-
a proper conversation with you
It starts with what the team is for. What does it exist to do, and what are the signs - your signs - that the job is being done right? And honestly: how close are you to the nitty-gritty of how the work happens now?
-
conversations with the team
Then I sit with the team and get the real picture of how the work happens - what's frustrating, what's easy, what tools they're drawn to, what they want to be able to do. If AI's already in use, we open up what it's producing - the DAX it wrote, the model it built - and go through it together, where it holds up and where it doesn't.
-
the fit map
What you get is a map: how the team actually works today, the skills in the room, and a destination that holds up regardless of the tools.
what's in the fit map
a good chunk of it is knowing what to stop worrying about.
- faster / better
what the team could be building faster, and what it could be building better
- can't yet / shouldn't touch
what AI can't improve for you yet, and what it shouldn't touch at all
- the skills brief
what to hire for, what to develop, what to stop worrying about
- the 90-day decision list
do now · don't do yet · watch
- the watch list - specific to your stack
if Microsoft brings a Fabric or Copilot feature out of beta, you'll already know what it makes feasible
the questions teams bring me
- We're paying for Copilot - what are we actually getting for it?
- It always sounds sure - so how do we know when it's right?
- What's safe to hand over, and what do we keep hold of ourselves?
- When we point it at our data, where does that data go?
- The ground keeps shifting - what do we even train for, or hire for?
afterwards
The difference is in the decisions: you make them knowing where AI helps your team, where it hinders them - and how you'd know.
When the next Copilot upgrade or data agent lands, the team doesn't panic - they know what their work actually is, and it isn't the clicks. The aim, longer term, is a team that's resilient to the tools changing, more engaged with the work, and prouder of what they put out.
already rolled out copilot?
Good - that's a start, but it's only licences so far. What changed once it landed: is anyone faster, are the outputs any better, and is someone reading the DAX back before it ships? The catch with Copilot is that it sounds just as sure about a model it's misread as one it's nailed. If those answers come back crisp - outputs demonstrably better, someone clearly checking the DAX - you're in good shape and you probably don't need me. If any of them made you hesitate, that's where I'd start.
what it comes down to
Every piece of work is a stack of decisions. Hand a problem to AI without naming those decisions, and it makes them for you - and a tool can't take responsibility for a decision.