Use case · Customer service

When the support queue grows faster than the team.

Customer service is one of the use cases we run most often. Here's the shape of the work: what stalls in support orgs, what we ship, and how Workplace Agent shows up in the channels your team already lives in.

What stalls

The patterns we see most

  • Routing drifts. Tier-1 re-routes the same tickets twice.
  • Senior reps writing the same answer for the fifth time. New reps writing from scratch.
  • KB articles 14 months stale. Reps find three contradicting ones, ask a senior person anyway.
  • Churn signals buried in ticket volume until it's too late.
  • 60–90 days before a new hire writes replies without senior review.
What we ship

What a CS engagement actually delivers

01
Intent classification and routing at intake. Tickets land in the right queue with a confidence score; humans review edge cases.
02
Drafted replies grounded in your knowledge base and prior tickets. Tone is configurable per channel and per customer tier.
03
Retrieval that actually finds the right KB article. Surfaces drift between what's documented and what reps are answering.
04
Sentiment flags on intake. High-risk tickets route to a named human owner immediately.
05
Observability dashboard for the Head of Support. Every model call is logged, every draft is reviewable, every change has a kill switch.
06
A handover period (workshops, runbook, optional retainer) so your team runs the system after we leave.

The engagement shape for CS clients

Most CS engagements use three of the four capability areas, typically Custom Development, Integrations, and Training. Advisory is often the entry door when the right first system isn't named yet. One contract, one team, the same people from kickoff through handover.

A typical first-engagement scope is 8–12 weeks, fixed-fee against scope. Pricing is per engagement, not per ticket or per resolution.

What this looks like in practice

A representative engagement, pulled from the Services catalog. Full case studies are on the way.

Problem

45-person services company with a 12-person customer-support team needed AI-assisted ticket triage and response drafting.

What we built

8-week build pulling from Zendesk, knowledge base, and product docs. Routes tickets by intent, drafts first-pass replies for human review, logs every model call for audit.

Outcome

Agents review drafts instead of writing from scratch. Average handle time drops; quality stays flat or improves. The team tunes it themselves.

From the field

Notes on what's working in AI deployments, what isn't, and what teams are actually asking for. Direct. No pitches.

Unsubscribe in one click. We don't share your email.

Tell us where support is stuck.

One sentence is fine. What's the workflow, what's jamming up, what would good look like?