Watson Talent Admin
The democratization of AI setup
Watson Candidate Assistant was a real product with a real problem. Customers liked the AI. They couldn't get to the AI without a several-week setup project, run by IBM engineers, billed by the hour, scheduled out months in advance.
Sales would close a deal and then say the words "and now we wait." Support would inherit a queue of half-configured tenants. Adoption capped at the speed of the implementation team.
We had to move setup out of services and into the product. The user (usually an HR ops manager, sometimes a recruiter) had to be able to do it themselves.
What we built was an admin tool that worked the way the AI worked. Confident, fast, considered. A clear path to a working configuration in a single afternoon.
What we walked into
Setup that took several weeks.
The configuration lived in code. Every customer got a custom build: questions, evaluation logic, integrations with their ATS, permissions, branding. Each change went through a developer ticket. Each ticket waited for the same six engineers.
The HR teams who actually owned hiring had no way in. They wanted to add a question to the screening flow on a Tuesday and have it live for Wednesday's candidate pool. Instead they sent an email, waited a sprint, and explained to their boss why it took so long.
Implementation services were the bottleneck. They were also half the cost of the contract.
A simple change request waited four to eight weeks. Customers learned to batch their asks.
Sales had to talk around the deployment timeline. Most prospects asked first, then quietly walked away.
Adoption stalled below where the AI's accuracy could carry it.
How we built it
A tool that talks back.
Six weeks of design sprints. Real HR teams in every session, not proxies. Carbon as the underlying system, with our own pattern library on top for the parts Carbon didn't cover. Nothing shipped until a non-technical user could complete the flow without help.
Inline help everywhere, in plain English. If a setting affected candidate experience, we said so. If it affected compliance, we said so. The tool taught the model as the user worked, so by the time setup was done they understood the AI well enough to defend it to their team.
Outcome
From weeks to days.
Setup time dropped 90%. The implementation queue cleared.
Customer satisfaction in the post-launch survey went up across every category, with the largest jump in "ease of getting started."
The implementation team retrained on advisory work, where they were always more useful than they were on a config screen.
Watson Candidate Assistant moved from "talk to sales for setup" to "see it running this week." Pipeline conversion followed.
The product had been hiding behind the services team. Once setup was a UI instead of a ticket, the product could compete on its own merits.
Takeaways
Up next
