How a messaging SaaS shipped a voice AI feature in weeks without hiring a voice team

What they were stuck on

A B2B messaging platform with thousands of customers using their SMS workflows. The customers kept asking the same thing: "Can this place a phone call too?" The CEO wanted to ship that feature.

The engineering team was strong on SaaS, weak on voice. They'd spent two months evaluating providers and were stuck on architectural questions they didn't have time to answer themselves.

They didn't want an agency to build it for them. They had engineers. They wanted senior voice judgment to enable the team to ship without a six-month detour.

What the eight weeks looked like

Week one was vendor selection. Five providers, head-to-head on latency, telephony stability, multi-tenancy support, function-calling reliability, and the pricing curve at the volume they'd hit by month six. We delivered the shortlist with the math and our pick. The team had been leaning toward a different vendor whose per-call cost would have priced the feature out of their plan tier by month six. The math killed it on the spot.

Weeks two through four covered architecture: customer accounts mapped to provider sub-accounts, where prompts were stored, how usage metered back to billing, how compliance from SMS carried over to voice (consent, opt-out, recording disclosure).

We also wrote the prompt scaffolding so the platform's customers could pick from three pre-baked agent types (appointment setting, lead qualification, callback follow-up) and customize within guardrails. Three clicks, no prompt engineering knowledge required on day one.

Weeks five through eight: the recording, transcription, and CRM-sync flow. Every voice call wrote back to the same customer record their SMS did, in the same shape. We reviewed every architecture decision the team made in a shared Slack channel during the build window. When their build hit walls (twice), we unblocked inside 24 hours instead of three weeks.

Where it landed

Voice feature shipped to GA in 8 weeks. The team had estimated 4 to 6 months on their own.

Zero internal voice hires were needed during the engagement. Their existing engineers learned the patterns and now own the feature outright.

The feature now ships as a standard capability of the platform. Customers can stand up a usable voice agent in three clicks. Call recordings, transcripts, and CRM events flow through the same plumbing as SMS, which means support, billing, and the rest of the org didn't have to learn a new system.

What we'd standardize earlier next time

Observability. We added detailed call-level dashboards in week six. They should have shipped in week one. The team would have caught two prompt regressions before customers did.

A "we considered and rejected" log on the vendor selection. The team made good vendor choices but had to re-explain the reasoning to internal stakeholders three separate times. A one-page rejection log per vendor would have saved hours of meeting time.

What enablement actually means

We didn't write production code for them. We didn't run their team standups. We sat in their Slack, reviewed their decisions, killed the bad ones early, and unblocked the hard ones fast. Their team owns the feature. We made the feature shippable.

When the gap is voice judgment and not engineering capacity, Voice AI Enablement is the engagement that fits.

Talk to us.