Book A Demo

Case study • Who Gives a Crap

Who Gives A Crap built a five-region chat operation that does action-taking work, with email running FAQ-and-routing alongside it

  • 72%

    Chat containment rate

  • 9,722

    Chat policy executions

  • 4.44★

    Product Details requests avg chat CSAT

Challenge

The volume distributes the way most consumer-product brands distribute: chat carries the real-time pressure (a customer asking about a missed shipping window or a stuck subscription expects a response in seconds), email carries the longer-form case work. The team needed AI that handled both for what they actually are. Chat had to contain conversations end to end where it could and route cleanly to live agents where it couldn’t, with action-taking policies executing per region. Email had to cover FAQ on the front end and make a clean ticket-state decision on the back end: mark solved when the issue was resolved, leave open when something still needed an agent’s eye.

Authentication had to land before the dashboard rolled out, since access policy required Google OIDC restricted to the company domain, with SAML on the maturity track behind it. And topic tracking had to mirror the existing Zendesk categorization on day one, because reporting fragments the moment AI conversations get tagged with a different taxonomy than agent-handled ones. The team didn’t want a parallel reporting structure to maintain.

Solution

KODIF deployed across chat and email running on the integrations the team was already on: Zendesk, and Google OIDC. The build sequence was deliberate. FAQ and ticket creation first to establish the chat foundation, then per-region action-taking policies layered on as confidence in the policy logic grew. Email came online with a pre-automation policy and a post-automation policy that owns the ticket-state decision.

Email automation on the post-processing decision

Email runs a post-automation policy that owns the ticket-state decision. When KODIF answers the question, the ticket is marked as solved. When something in the language or the context requires human review, the ticket stays open with full context for an agent to pick up. The status logic is what makes the throughput safe to run unattended on the email side, where action-taking policies are not yet in production. There are no policy automations on email today: the channel is intentionally FAQ-and-routing while the chat policy library matures further

Sensitive topic escalation

Sensitive topics are rare, but they urgently have to land on a person when they do happen. The policy detects the language, creates a Zendesk ticket, and routes immediately to an agent. No AI response, no auto-resolution. This is the kind of edge case where speed of escalation matters more than depth of automation.

SSO via Google OIDC

Dashboard access is restricted to the company domain through Google OIDC, with SAML on the maturity roadmap. The point isn’t the protocol, it’s that authentication landed before the dashboard rolled out, so access policy and AI deployment moved on the same timeline. Reporting on AI-handled conversations sits inside the same identity boundary as the rest of the team’s tools.

Topic taxonomy aligned with Zendesk

Topics on AI-handled conversations are managed against the same taxonomy the agents use in Zendesk, so AI-handled and human-handled conversations report into one structure rather than two. The team did not have to maintain a parallel reporting layer during the rollout.

Results

66%

of chat resolved end-to-end

63%

of email resolved end-to-end

Across 21,294 eligible chat conversations and 1,554 email tickets in Q1 2026, KODIF resolved 66% of chat and 63% of email end to end

Inside the chat policy mix:

  • 9,722 policy executions in Q1 2026: WISMO, agent handoff across Australia, Canada, EU, UK, USA
  • 19,979 FAQ responses answered without agent involvement
  • 15,326 conversations contained end to end
  • 14,019 chat conversations resolved end to end
  • 4.44 stars average on Product Details requests
  • 11 chat policies running across the five-regions

Inside the email channel:

  • 980 emails contained, with 488 fully resolved end to end
  • 31% fully automated, the early baseline before action-taking policies expand to email

Chat is doing the deep automation work. 9,722 of the chat conversations involved a policy execution, not just a knowledge answer, which works out to roughly 46% of all chat conversations resolving with action-taking rather than deflection. Second, the email side is leaning on FAQ coverage and the post-automation policy to keep the ticket queue clean, rather than executing customer-facing actions. Both channels are contributing differently, which is what the architecture was designed for.

What's Next

Who Gives A Crap is extending the deployment in five directions:

  1. 1

    Email policy expansion

    bringing action-taking policies (cancellation, WISMO) onto the email side now that the post-automation handling is stable Continued policy depth on Agent Handoff across the AU, CA, EU, UK, and US footprint as conversation patterns differentiate per region

  2. 2

    Subscription Management in CA and EU

    closing the regional gap so cancellation runs in all five markets rather than three

  3. 3

    SAML SSO

    as the next authentication tier alongside the existing OIDC setup

  4. 4

    Continued policy depth on Agent Handoff

    across the AU, CA, EU, UK, and US footprint as conversation patterns differentiate per region