The Future of AI in Rewards Isn’t a Model Problem. It’s a Data Problem, an Identity Problem, and an Operating-Model Problem.
This week I had the privilege of sitting down for a fireside chat at HRcoreREWARD 2026 in Amsterdam — thirty minutes with a sharp moderator and a room full of rewards practitioners who came for implementation, not inspiration. What follows is a distilled version of what we discussed, expanded where I had to rush on stage.
The frustration that started it all
Every “Future of AI in Rewards” talk I’ve sat through in the past year follows the same script. An exploding possibilities curve. Three vendor logos. A reassuring bullet about “human in the loop.” You nod along, go back to your desk on Monday, and nothing changes.
That’s because the conversation keeps starting in the wrong place. The “can AI do this?” question is over. AI can. The real question — the one that actually matters for anyone running comp today — is whether your rewards function is built to let it.
My read: there are three things standing in the way, and they need to be fixed in order.
Your data is the blocker, not the AI
I’ll tell on myself. At Palo Alto Networks, we set up a NotebookLM project — loaded our allowance policies so recruiters could chat with it, ask questions, get answers in real time. Self-service model. The recruiters loved it.
Then we discovered one of the policies we’d fed into the model was outdated. Wrong numbers. The AI had been confidently giving recruiters incorrect information — and some of them had been using it.
We had to clean up a real mess. Not because the AI was broken — the AI worked perfectly. The policy document was wrong.
This is the fundamental truth nobody wants to hear: most rewards functions are not sitting on clean data. They’re sitting on policies nobody’s audited in two years, job catalogs with stale titles, comp histories in spreadsheets that three people quietly email around. AI doesn’t rescue you from that. AI publishes it at speed.
For anyone from a manufacturing background — you know FIFO: first in, first out. With AI, the crap gets out really fast.
The work ahead isn’t prompts. It’s plumbing.
The rewards analyst role is ending. The architect role is being born.
This is the part that makes people uncomfortable, so I’ll say it directly: the analyst role, as we’ve known it for twenty years, is compressing. Not to zero — but to a fraction.
Think about what an analyst actually does. Pulling benchmarks. Matching jobs to surveys. Building range grids. Chasing clean inputs. All valuable — and all fundamentally clerical. That is exactly the work AI is good at.
Meanwhile, the work AI makes no dent in — designing comp philosophy, running board conversations, being the ethical conscience when the business gets loud, defending fairness under regulatory scrutiny — that’s the work most of us said we wanted to do when we entered this field. We just never had time for it because we were buried in spreadsheets.
The architect role is what opens up. Someone who can design a compensation system, explain it in plain language, defend it to a regulator, debate it with a CEO, and keep the workforce’s trust.
Teams will shrink in headcount. They will grow in weight and influence. That’s a good trade — if you own the transition instead of waiting for someone to own it for you.
If your value has been your spreadsheet, you should be worried. If your value has been your judgment, this is the best decade of your career.
The architect’s operating model is product management
Here’s where it gets personal. Before I was a comp person, I was a part-time software developer during university and studied engineering. That background changed how I approached every single project in rewards.
Every project I’ve taken on, I’ve run it like a product sprint. Get to know the user. Ruthlessly prioritize. Ship fast. Iterate. Measure. My first two promotions at Facebook were directly based on this approach — building tools the entire rewards team used, designing incentive plans within a single year. That’s not analyst work. That’s product work. I just didn’t call it that at the time.
Here’s the reframe: rewards is the only function in HR that still runs on an annual cycle. Every other function has moved to iterative, outcome-driven work. We’re stuck. And we’re stuck because we’ve been running rewards like a cycle when we should have been running it like a product.
Your employees are your users. Your CFO is the exec sponsor. Your HRIS is the tech stack. AI is the leverage layer. Total rewards is the product. You’ve always been a product owner — you just never called it that.
Four habits worth stealing from product management
- Know your user. Test pay comms the way product teams test onboarding flows. Measure TRS comprehension, not just distribution. When was the last time you did a usability session on your benefits portal?
- Ruthlessly prioritize. Stack-rank every rewards initiative by impact vs effort. Most rewards teams run every program at equal priority because it’s all on the cycle. Product teams kill things. We need to learn how to kill things.
- Ship in sprints, not cycles. Pilot something in four weeks. Iterate. Measure. The annual cycle is a planning horizon — not a shipping horizon.
- Metrics that mean something. Adoption. NPS on the total rewards statement. Time-to-offer. Pay-equity closure rate. Stop measuring “we ran the cycle on time” and start measuring outcomes your CEO actually cares about.
One caveat I gave the room in Amsterdam: I’m not saying run rewards like a Silicon Valley startup. Your users don’t opt in, there’s no churn, and “move fast and break things” will get you sued. Fairness matters more than delight. Steal the four habits, leave the Silicon Valley cosplay at home.
PM discipline is the operating system. AI is the processor. Run them together and your function changes. Run one without the other and you get either disappointed or frustrated.
The regulatory tailwind nobody’s talking about
Everything we discussed in Amsterdam happened at a very specific moment. Four weeks after the conference, the EU Pay Transparency Directive transposition deadline lands (7 June 2026). Twelve weeks after, the EU AI Act starts treating what we do as high-risk — with fines up to €35 million or 7% of global turnover.
Most people treat regulation as an obstacle. I think it’s the opposite. It’s a design brief.
If you’re running an AI tool that influences any pay decision, you now need to explain it, audit it, and prove a human decided. That’s not a tax — that’s exactly the kind of rigour that makes AI trustworthy enough to scale. The compliance bar and the trust bar are converging. Build for both.
What to do on Monday
If you’re running rewards at a European company and you’ve done nothing with AI yet, here’s a 90-day plan. The first two moves need zero budget approval.
Weeks 1–4: Audit your job catalog and your policies. Count stale titles, duplicate roles, missing levels. Check every policy document for currency and version control. You need this anyway for Pay Transparency.
Weeks 5–8: Pick one AI use case with clean data. Benchmarking is usually the safest — structured data, clear output, easy to compare. Run a parallel pilot: old way and AI way. Measure the delta.
Weeks 9–12: Bring one number and one story to your CFO. “We cut benchmarking time from X to Y. Here’s the capacity we freed up. Here’s what we want to invest in next.”
Notice what those three moves really are — your first product sprint. Scoped problem, clean data, measurable outcome, stakeholder pitch.
Don’t buy an AI tool in April if you haven’t cleaned your job catalog by March.
The bottom line
In 2029, the rewards function that wins isn’t the one with the best AI tools. It’s the one that runs itself like a product team — clear users, ruthless priorities, a real roadmap, metrics that mean something. AI is what makes that possible. Product discipline is what makes it work.
We’ve been running the rewards cycle for twenty years. It’s time to start running the rewards product.