HR softwareGoal: grow revenue with AI8 weeks to beta
An AI feature that shows up in revenue, not just in the press release
Competitors were announcing AI features every month. The product team of a UK HR platform had a backlog of 23 ideas and no idea which one customers would actually pay for.
- Client
- HR software company, UK
- Their team
- CPO (owner), product manager, 6 in-house engineers
- Our team
- Engagement lead, 2 AI engineers, designer
- Timeline
- 8 weeks to beta · 5 months to general availability
41%
of accounts use the AI features within 3 months
+14%
revenue per account on the new tier
40 → 4 min
to write a job post
< 3%
of the tier price spent on AI costs
The business goal
Launch an AI capability customers pay for — a new pricing tier — not a demo for the press release.
Where it was
Recruiters at customer companies wrote job posts and screened CVs by hand. The idea list came from sales, support and competitors’ announcements — nobody had checked which ideas saved customers real time.
23 ideas
from sales, support, competitors
12 interviews
plus usage data
2 features
job posts · candidate summaries
What we did
- 1Outcome brief · 2 weeksCustomer interviews and usage data cut 23 ideas down to two with clear time savings and a price customers accepted.
- 2Beta · 8 weeks25 customers. An evaluation suite for quality and bias, usage-based cost limits from day one.
- 3General availability · 3 monthsRollout behind feature flags, the new pricing tier, admin controls. The in-house team took over development, pairing with us.
- 4Run & handoverMonthly review of model cost and quality; full handover to the in-house team after six months.
What did not work
Automatic candidate ranking: beta customers did not trust it, and their legal teams worried about fairness. We replaced it with neutral candidate summaries — and usage tripled.
Who did what
The client’s team
Pricing, sales enablement, customer interviews and the final say on scope.
Our team
AI features, evaluation suite, cost controls, training the in-house engineers.
Together
A weekly beta review: usage, quality scores, cost per account.
Before and after
Under the hood
LLM features behind a provider-agnostic gateway; evaluation suite with 400 test cases, including bias checks; per-account usage limits and cost dashboards; feature flags for gradual rollout; pairing sessions for handover.
We finally have an AI feature that shows up in revenue, not just in the marketing deck.