Microsoft Copilot Won't Predict Who's Leaving. It Will Show You What You're Already Missing.

Most articles about AI and staff turnover start with a prediction engine. Feed in enough data, so the theory goes, and a model will tell you who's about to hand in their notice.

For most UK HR teams, that's the wrong starting point.

Not because the theory is wrong, but because it assumes infrastructure most functions don't have: solid data or a data team, a clean historical dataset.

Typilically, I find that what HR teams actually have, in most organisations with a Microsoft 365 licence, is Copilot. It needs the exhausted survey data, exit interview notes and one to one summaries already sitting in your SharePoint.

The problem was never the data. It was the time to look at it.

Depending on how it's measured, UK turnover sits at around 35% (CIPD's analysis of ONS data, covering everyone who changes employer or leaves work within a year) or closer to 15% (CIPD's employer-reported annual survey). The gap between the two is a reminder that "turnover" means different things depending on who's counting and how.

Regardless, behind those averages sits a pattern most HR professionals will recognise: the signals of someone leaving are usually there in the data long before the resignation letter. Declining engagement scores. A quiet drop-off in one-to-one notes. Exit interview themes that repeat, quarter after quarter, without anyone joining the dots.

The reason nobody joins the dots isn't a lack of insight. It's a lack of time. Reading twelve months of exit interviews properly takes hours most HR Managers don't have between the meetings, the ER cases and the policy queries already filling the diary.

This is where Copilot can change that - not necessarily by predicting anything, but by reducing the cognitive load of finding a pattern by reading through reams up reams of data.

What this actually looks like on a Tuesday

Not a dashboard. A conversation with the documents you already have.

Finding themes in exit interviews Instead of re-reading a folder of exit interview notes from the last two quarters, you could ask Copilot in Word or SharePoint:

"Review the exit interview notes in this folder. Identify the three most common reasons given for leaving, and note which team or department each theme is concentrated in."

You're not asking it to predict anything. You're asking it to do the reading you don't have time for, and hand you back a starting point for a real conversation with your leadership team.

Preparing for a retention conversation Before a one-to-one with someone you're concerned about, Microsoft 365 Copilot can pull together everything already sitting across your inbox, Teams messages and shared documents:

"Summarise recent one-to-one notes, engagement survey comments and any relevant Teams messages relating to [employee/team], and flag anything that suggests a change in engagement over the last three months."

That summary isn't a verdict. It's twenty minutes of preparation done in two minutes, so the conversation itself can be about listening rather than remembering.

Surfacing engagement survey themes Open-text survey comments are usually the richest data HR holds and the least analysed, because nobody has a free afternoon to code four hundred free-text responses by hand.

"Read the attached engagement survey comments. Group them into themes, note the tone of each theme (positive, neutral, concerned), and identify which themes appear most frequently in [specific team/department]."

Briefing leadership without the all-nighter When a director asks "what's going on with retention in operations," Copilot can turn a scattered set of documents into a first draft brief in minutes:

"Draft a one-page briefing for our HR Director summarising retention themes in the Operations team over the last two quarters, based on the attached exit interview summary and engagement data. Keep it factual and avoid overstating conclusions."

That last instruction matters. Ask it to stay factual, because it will confidently overstate a pattern if you let it.

The theory behind why this works

Augmented intelligence, not artificial intelligence. The distinction is more than semantics. Copilot doesn't replace the judgement of an HR professional who has read a room and knows the context Copilot can't see. It reduces the volume of raw material a person has to process before they can apply that judgement. The decision stays human. The reading gets faster.

Cognitive load reduction. HR professionals aren't short of insight, they're short of the mental bandwidth to extract it from documents while also running the rest of the job. Every hour Copilot saves on synthesis is an hour returned to the parts of the role that actually require a person: the difficult conversation, the judgement call, the read of what someone isn't saying.

Knowledge synthesis across silos. The reason patterns get missed isn't usually that the data doesn't exist. It's that it's scattered across an inbox, a Teams channel, a shared drive and someone's personal notes, an excel report and nobody has connected all the information together.

Microsoft 365 Copilot's real advantage over a standalone chatbot is that it can work across those silos, inside your existing Microsoft environment, without you needing to copy and paste anything anywhere.

Remember to use the forward slash / to add documents to your prompts to connect various pieces of information together.

Human in the loop, by design, not by hope. Every prompt above ends with a person making a decision. Copilot surfaces a pattern but a human - the expert in the room - decides whether it's meaningful, whether it warrants action, and what that action should be. That sequencing isn't optional. It's the entire safeguard against the tool being wrong.

The misconception worth noting

The most common misconception isn't that Copilot is useless. It's that it's a crystal ball. Framed as prediction, it invites exactly the wrong kind of trust: HR teams treating a theme summary as a verdict, or worse, acting on it without the conversation that should follow. Framed as synthesis, it invites the right kind: a faster route to the question that was always the actual job, which is "what's happening here, and what do we do about it."

Where caution really matters

Copilot is working with whatever documents it has access to, and the quality of any summary is only as good as the quality and completeness of what's fed in. A theme drawn from three exit interviews is not evidence of a pattern; it's three data points wearing a pattern's clothing.

There's also a governance question that has to be settled before any of this starts, not after. What employee data is appropriate to put through Copilot, who has access to what, and what the organisation's position is on using it for anything that touches individual performance or wellbeing. Under UK data protection law, that means being able to explain clearly how the data is used, avoiding anything that could produce a discriminatory outcome, and being transparent with staff about it. None of that is a reason to avoid the tool. It's the groundwork that makes using it safely possible.

And the human element doesn't move. AI-generated themes are a starting point for a conversation, never a substitute for having it. The moment a summary becomes the whole intervention instead of the trigger for one, the value has been lost.

Where human judgement stays non-negotiable

Reading a room is so important and not something AI can do for you. Knowing that someone's engagement score dropped because their mother is unwell, not because they're disengaged. Deciding that a pattern, however consistent, doesn't yet warrant raising with the individual concerned. Copilot can tell you where to look. It cannot tell you what to do when you get there, and it shouldn't be asked to.

Four things to try this week

1. Pick one exit interview theme you suspect exists but haven't proven. Ask Copilot to check it against your last two quarters of notes.

2. Before your next retention-focused one-to-one, ask Copilot to summarise the relevant history first. Notice how much of the meeting that frees up for actually listening.

3. Take your last engagement survey's free-text comments and ask Copilot to theme them by department. Compare it against your own gut sense of where the problems are.

4. Agree, in writing, with your HR Director what data is and isn't appropriate to put through Copilot before anyone in the team starts using it for anything people-related.

The bottom line

The future of retention in UK HR doesn't belong to whoever builds the most sophisticated model. It belongs to whoever stops letting the data they already have go unread. Copilot won't tell you who's leaving. It will give you back the time to notice for yourself, and that's the more useful promise of the two.

If your HR team has access to Copilot but isn't using it for anything beyond drafting emails, that gap between what you have and what you're using is exactly what we work through in a Rethink:Team day.