More managers now ask an AI assistant before they ask HR. At our latest webinar we looked at what that advice is built on, what it costs when it is generic, and what changes when the AI knows the person.
In short: 6 in 10 managers rely on AI for decisions about their direct reports (Resume Builder, 2025), using tools that know nothing about the person. AI for managers only works when it is grounded in validated insight, there in the moment, governed by role, and offers options instead of a script.
We opened the session with a question. When a manager has a difficult conversation or a 1:1 coming up, where do they go first: their instinct, HR, an AI assistant, or do you honestly not know? Instinct led the vote, with Ask HR and Ask AI close behind.
AI is further ahead than most HR teams think. In a 2025 Resume Builder survey of 1,342 US managers, 6 in 10 said they rely on AI for decisions about their direct reports.
Your managers already have a coach. It’s there any time. It sounds confident and it’s generic. It knows nothing about the person in front of them.
Sarah Kelsey, Product Marketing Manager, Saville Assessment
Talent leaders tell us managers are already pasting validated assessment reports into general AI tools. Nobody waited for a policy. Short of time and short of answers, people go to whatever answers fastest.
The chain is short. A manager gets generic advice, the conversation goes badly, and it ends in a resignation or a claim. One talent leader put the root cause simply: “Managers aren’t enabled to have the conversation and give employees the tools.”
In our research with 300 senior HR leaders (Sapio Research for Saville Assessment, 2025), 55% find it hard to balance data with human judgment, 47% struggle to keep people decisions objective, and 36% say managers often or always bypass formal processes on hires and promotions. Generic AI sits on top of all three.
The biggest cost lands at the last step. 71% of the same leaders say people leave because of poor management, rather than role or pay. Now set that next to the 6 in 10 managers asking AI about the people they manage.
The advice shaping your biggest retention risk is coming from a tool that has never met your people, and that nobody here chose or controls.
Sarah Kelsey, Product Marketing Manager, Saville Assessment
The first cases are already here. In Australia, a ChatGPT-drafted redundancy letter showed the decision had been made before consultation, and the dismissal was ruled unfair (Lord v Millet Hospitality Geelong [2025] FWC 2740). As Alex Bailey, CEO of Bailey and French, told us on The Deep Dive, the risk is sharpest where teams are stretched and there is no HR person to sense-check the advice.
AI isn’t the problem. Generic advice is.
Sarah Kelsey, Product Marketing Manager, Saville Assessment
Most L&D teams plan around 70/20/10: roughly 10% of development from formal learning, 20% from other people, mostly the manager, and 70% from the job itself. The 10% happens once, and has faded by the time a manager prepares for a hard conversation.
The 20% is where development turns into behavior: the feedback, the 1:1, the coaching moment. It rests entirely on the manager, and 82% of people who step into management have never had formal management training (Chartered Management Institute, 2023).
The money follows the 10%. Developing one first-line manager costs roughly £1.5k (CMI), and manager development takes about 12% of training budgets (Training Magazine Industry Report). Yet in our survey of 500 assessment users (Saville Assessment, 2026), only 41% of organizations say line managers use assessment reports often, against 61% for talent acquisition.
One HR leader described the usual ending: “They got a report and they said, ‘oh yeah, it was nice,’ and I just put it away.” When we asked attendees where most of their manager development budget goes, around 46% said on the job and 33% formal training, with almost nothing on coaching.
Regardless of how plain and transparent an assessment output is, busy managers want the headlines, and they’re likely to need support in understanding what practically they should do with the information.
Lauren Jeffery-Smith, Principal People Scientist, Saville Assessment
Most organizations have tried at least one fix already. Each one misses something.
So we hold any AI for managers to four standards. Miss one and the advice is generic again.
The last one matters more than it looks. When Lauren tried a standalone AI tool to prepare for progress check-ins, it handed her a script full of very definite interpretations. She rewrote it as talking points. Most managers will not have the time to do that.
If someone is reading from an AI script in a conversation, you can tell, and it undermines what they are saying.
Lauren Jeffery-Smith, Principal People Scientist, Saville Assessment
We asked one question twice: how do I tell Sam his promotion isn’t happening this year? A general assistant said to be honest and direct, acknowledge his feelings, focus on the facts and agree next steps. It sounds right. It fits anyone, so it fits no one.
Then Lauren ran live scenarios through our AI plugin, which connects Wave Connect Develop to the AI assistant your managers already use. Planning difficult feedback, it pulled out the relevant parts of the person’s Wave profile and built an outline of what to cover. For a project group, it suggested who might take which responsibility, and flagged two people whose shared perfectionist streak could shape the work.
It also held its boundaries. Asked about someone outside the manager’s team, it would not share their profile. Asked who was most likely to leave after a restructure, it declined, because Wave does not measure that.
AI will generally try to answer almost anything. That’s why we’ve put clear guardrails into our MCP to keep it from speculating about things that Wave isn’t designed to measure.
Lauren Jeffery-Smith, Principal People Scientist, Saville Assessment
Everything the assistant uses lives in Wave Connect Develop. Managers see their own insights first, then their team’s make-up and gaps, then how to work with each person. It is written in plain, narrative language, so nobody needs training to use it, and it all comes from a single Wave assessment.
One buyer asked us exactly that: “What stops us getting our reports and running our own LLM?” Four things.
That science is what makes the advice fit. Modeled against the average development tool, Wave identifies real development strengths and limitations about 60% more accurately and has around four times the development impact, and better development is projected to cut voluntary turnover by about 5% in organizations of 500 or more (Saville Assessment talent development ROI model).
Tell us the conversation on your desk, and we’ll show you how grounded AI answers it, in the assistant you already use.
Book your demo See the AI plugin Watch the on-demand webinar