HR teams sit on a mountain of information. Who’s off next week. How someone's holiday entitlement stacks up. What the sickness absence policy says. Which training is overdue. It's all there in the HR software – but getting to it often means running a report, clicking through screens, or sending yet another email to HR.
That friction is the problem we set out to solve when we started building our HR AI assistant. And in April 2026, after months of internal testing and close work with our customer advisory group, we quietly went live with the beta.
This is the story of why we built it, how we made it, and what we've learned from the HR teams, managers, and employees who've been using it from day one.
The problem: self-service that isn't actually self-sufficient
Ask any HR team where their week goes and a pattern emerges. A surprising slice of it is spent answering the same handful of questions: 'What's my holiday balance?', 'Who's off this week?', 'Which policy applies to me?' These are questions that, in theory, employees and managers could answer themselves.
In practice, they often can't. Traditional self-service assumes people know where to look, what report to run, and how to navigate a system they might only use a few times a year. When the path is unclear, the easier option is always to ask HR.
Meanwhile, managers need workforce context fast – before a one-to-one, before a team planning session, before a board meeting. Employees want quick answers in the flow of their day. And HR teams want to spend their time on the work that genuinely needs their expertise, not on balance checks and policy lookups.
What we built: Ciphr's HR AI assistant
The HR AI assistant lives inside Ciphr HR, as a part of our AI HR software. There's no new system to log into and no app to install. Users click the 'ask' button already in the top navigation bar and type a question in plain English. They get an answer in about 10 to 15 seconds.
The first release focused on two areas we knew would create immediate value for the broadest group of users:
- Workforce data queries, for HR and managers – 'how many people joined in the last three months?' 'What's our absence rate this quarter?' 'Show me training compliance across my team.' The assistant pulls the answer from live HR data and, where useful, presents it as a simple table or chart
- Holiday and absence, for everyone – employees can check their balance and book time off in a single conversation. Managers can see who's off when, before they plan workloads
There was an unexpected standout early use case: one-to-one preparation. Managers ask something like, 'I'm about to do a one-to-one with Sam – can you give me some insight?', and get back a quick briefing on role, tenure, recent absence, and even suggested conversation openers. It's the kind of prompt people don't think to try until someone shows them – and then they use it every week.
How we built it: the principles that shaped every decision
From the start, we had a few non-negotiables.
- Usefulness over hype – we didn't want to ship a demo. The beta launched as a working product, useful from day one. If it wasn't genuinely helping people, it wasn't ready
- Reduce admin, don't add to it – no new login. No training required. No IT project to stand it up. The assistant uses the same permissions, the same data, and the same infrastructure as the rest of Ciphr HR
- Trustworthy answers (or honest silence) – we deliberately built the assistant to say when it doesn't know. Every factual response shows where the data came from and when it was last updated. For complex or sensitive situations, it points users to see advice from their HR team rather than guessing. Trust comes from transparency, not from always having an answer
- A phased, controlled rollout – HR admins get access first, with the ability to decide when (and whether) to open it up to everyone else. That control matters. Different organisations are at different places on their AI journey, and nobody should feel rushed
Secure. Trusted. Built for HR
AI in HR raises legitimate questions – about privacy, about accuracy, about what happens to sensitive employee data. We took those questions seriously from the start.
Our HR AI assistant is embedded directly inside Ciphr HR, not bolted on alongside it. That means your data stays in your HR system. It's not sent to public AI tools, and it's not used to train wider AI models. There is no need to paste employee information into something like a public assistant to get a useful answer – the answer is already inside the software you trust.
Because the assistant uses your existing Ciphr HR permission model, a line manager sees their team's data, an employee sees their own, and an HR administrator sees what their role allows. Nothing more. We've also built in specific guardrails around sensitive categories like medical records, protected characteristics, and financial details.[
For factual questions, the assistant retrieves answers from your live HR data rather than generating them from general AI knowledge. That architectural choice was deliberate – it's the single biggest reason we can be confident about accuracy on the things that matter.
What we learned from our customer advisory group
11 organisations from our customer advisory group went live with us on day one, earlier this year. Their early feedback has shaped the product in ways a closed testing environment never could.
Three themes stood out.
Firstly, trust came up before anything else. Customers weren't worried about features. They were worried about AI making things up, about who could see what, and about where the data goes. That's why source attribution, the 'I don't know' behaviour, and the permission-model-only architecture aren't add-ons – they're foundations. We even built a dedicated trust and safety prompt into the onboarding experience for users who were hesitant after their first few interactions.
Secondly, feedback loops had to be live from day one. The in-app feedback button went in before launch, not after. In the first few weeks of live use, the overall success rate moved from around 65% to nearly 90% – the kind of jump that only happens when real customers are using a product and telling you, in real time, where it needs to improve.
Thirdly, customers wanted a tool for employees too – not just HR. That shaped the decision to prioritise holiday booking alongside the workforce data use cases. It's high value, low risk, and instantly useful. Live data from the first week backed this up: nearly half of all assistant queries were about holidays and absence.
What our HR AI assistant can help you do
- Check holiday balances and book time off through a simple conversation
- Pull instant workforce insights – headcount, tenure, absence patterns – without running a report. The assistant can even turn data into a chart
- Prepare for one-to-ones with a quick briefing on each team member
- Spot training and compliance gaps across your workforce
- See who is off and when, before you plan team workloads
- Get quick, trusted answers to everyday HR questions – in plain English
What's next for Ciphr's HR AI assistant?
This is the beginning, not the end. On the roadmap: HR policy queries with cited answers, so users can ask 'What's the policy on enhanced sick pay?' and get a trusted response with a link to the source document. Mobile access. Personal details self-service. Deeper integrations across our platform – payroll, benefits, LMS – so the assistant can help with more of what HR actually does.
The principle behind all of it stays the same. AI should lighten the load for HR, not add to it. It should give employees and managers quick answers they can act on. And it should do all of that without ever asking you to trade away control of your data.
Ready to see the HR AI assistant in action?
If you'd like to see how our HR AI assistant could help your team, we'd love to show you.
Book a demo to see the assistant in the context of Ciphr HR or download our AI factsheet to learn more about how our AI capabilities are built – and how we keep them safe, reliable, and genuinely useful.
Let's make HR work better, together.
About the author
Martin Pyle is senior product manager at Ciphr, where he has spent nearly six years working across product and operations. He leads on AI strategy and looks after the learning and analytics product areas, with a passion for helping mid-market companies punch above their weight with technology. Outside work, he's a music lover, dog enthusiast and garden potterer who can usually be found walking, reading or watching the latest sci-fi.