47% lower support cost, 5-point CSAT lift
Client: Series A SaaS scale-up, ~80 employees, US-East · Timeline: 9 weeks to go-live + ongoing
The challenge
The client ran a 22-person internal customer support team handling ticket, chat and inbound voice for a B2B SaaS product with roughly 4,200 paying accounts. Average handle time on voice sat above 6 minutes, the ticket backlog peaked over 1,800 every Monday, and weekends were effectively a skip day because the in-house team did not cover Saturday or Sunday.
Roughly 55% of inbound contacts were repetitive L1 questions: password resets, billing edits, plan downgrades, basic configuration. The team burned hours on those before getting to the cases that actually needed product knowledge. CSAT had drifted from 84 to 71 over two quarters, and the head of CS was about to be asked to hire six more reps to keep up with growth.
Leadership did not want to grow headcount that fast. They wanted to keep their senior team intact, deflect the low-value contacts, and add weekend coverage without standing up a second internal pod.
The solution
We brought in two of our service lines in parallel: AI & Automation built the deflection layer, and Call Center Outsourcing built the human overflow pod.
For deflection, we deployed a Vapi-based voice agent and an Intercom Fin chat agent, both wired into the client’s existing Zendesk instance and product API. The voice agent handled inbound calls end to end for password resets, plan changes, invoice resends and seat management. Anything outside that scope was warm-transferred to a human with full context handed off in the ticket.
For the human side, we stood up a 9-seat hybrid pod inside our delivery center to cover L2 cases, weekend coverage, and a 24/7 overflow queue. The pod was trained on the client’s product over a 3-week ramp, shadowed the internal team for a week, then took live tickets.
We also rebuilt the macros and routing logic in Zendesk so the AI agent, our pod, and the internal team each owned a clearly scoped queue. A weekly QA review caught regressions early — month two we found the voice agent was over-escalating refund requests, retrained it on the policy doc, and dropped misroutes by half.
How we delivered
Weeks 1-3: discovery and queue surgery
We started by sitting inside the Zendesk instance for a week and tagging every ticket by intent. That gave us a defensible deflection target — 55% of contacts were L1, not the 30% the team had assumed. We mapped each L1 intent to either an AI flow or a macro, and rewrote the routing rules so the AI agent, the internal team, and our overflow pod each had a clean queue.
Weeks 3-6: build and shadow
Our AI & Automation team built the Vapi voice agent and the Intercom Fin chat agent against the client’s product API. In parallel, the Call Center Outsourcing pod started a 3-week ramp on the product, shadowed live tickets in week 5, and took their first independent shift in week 6. We kept a senior internal agent assigned as a floating QA reviewer the whole time.
Weeks 6-9: cut over and stabilize
Go-live was staged. Voice deflection went first at 20% of traffic, then 60%, then 100% over a two-week window. The hybrid pod took weekend coverage from day one and absorbed weekday overflow as the internal team rebalanced toward L2 and L3 work.
The thing that nearly went wrong
In week 7 the voice agent started over-escalating refund requests because the refund policy doc it had been trained on used softer language than the actual policy. CSAT on those calls dropped before we caught it in the weekly QA review. We retrained against the canonical policy, added a guardrail prompt for refund edge cases, and the score recovered inside a week.
What ongoing looks like
- Monthly review with the head of CS on deflection rate, CSAT, and AHT
- Quarterly retraining of the AI agents against new product releases
- Hybrid pod scales up or down by shift, not by hire
Six months in, the client has held headcount flat, kept CSAT above 76, and the deflection rate has crept up to 58% as the AI agents have learned more of the product surface.
Outcomes that matter
Want results like this?
Book a discovery call and we will scope the right path for your goals.