Cutting contact rate with AI and self-service
An AI chatbot answered 60% of tier-1 questions and helped defer 4–6 hires.
The result: at Workable, our AI chatbot (Forethought) answered 60% of tier-1 questions. Together with root-cause fixes, it cut (tickets that reached a person, divided by monthly active users) in half. CSAT held at 97%, and we deferred 4–6 hires while the business grew.
My take: the bot was the last step, not the first. It worked because we'd already written 700+ articles, videos, and guides for it to draw on.
What I did: wrote the requirements, worked across teams on the rollout, owned the budget, and built the content plan with my team. After launch, I owned the bot.
What kept it good for customers:
- We rolled it out in phases, so we could see the impact without disrupting customers or the team.
- It never blocked anyone from reaching a human.
- We connected more than the help desk, bringing in HubSpot and customer context to personalize it and get customers to the right person the first time.
- We audited the logs every month to find improvements.
- Cancellations, purchases, and requests for a person always followed fixed flows, never improvised bot answers.
The harder rollout was internal. Agents were slow to adopt Forethought's agent-facing tool, which we launched alongside the bot. Training laid the groundwork, but a usage leaderboard and handing peer training to the team's heaviest user made it stick.
Why contact rate: I track contact rate instead of deflection because it captures all of our customer enablement work in one number: the bot, the help center, onboarding, the academy, and the product fixes. Deflection only tells you how the bot thinks it did.