Case Study

AI that meets your techs where they already work

How Capstone Plumbing turned ServiceTitan into a searchable customer memory — without changing how their team uses it.

Capstone Plumbing

  • Residential Plumbing
  • Cave Creek, AZ
  • ServiceTitan
  • Live since May 2026

Capstone Plumbing is a residential plumbing company based in Cave Creek, Arizona, serving homeowners across the North Phoenix metro — from Anthem and Carefree to Scottsdale, Paradise Valley, and the surrounding desert communities. They run their day-to-day on ServiceTitan, and Foundation built and operates a dedicated AI agent for them on top of it.

Visit capstone-plumbing.com
01

Never lose a customer call again

ServiceTitan only keeps call recordings for about 60 days. We keep them forever, and we make every word searchable.

  • Every customer call gets pulled out of ServiceTitan automatically and turned into searchable text.
  • Trained on plumbing vocabulary, so it catches things like "PRV," "recirc pump," and "slab leak" instead of garbling them.
  • Search any call by any phrase — instantly — no matter how long ago it happened.
  • We loaded in everything ServiceTitan still had on day one (about 1,150 calls).
  • New calls show up automatically, every 15 minutes.
For the technically curious
  • Recordings are pulled via the ServiceTitan API and transcribed locally with a Whisper distil-medium.en model fine-tuned for plumbing vocabulary.
  • Transcripts and metadata land in SQLite with FTS5 full-text search, giving instant phrase queries across the full archive.
  • The pipeline runs idempotently every 15 minutes via cron; per-call errors are captured without blocking the rest of the run.
  • On day one we backfilled the entire 60-day ServiceTitan retention window in a single pass — about 1,150 historical recordings.
02

AI summaries where techs already work

No new app, no new login. Techs see a smart summary on every customer, pinned right inside the ServiceTitan app they already use all day.

  • Summaries appear at the top of the customer in ServiceTitan — the same place techs already check before a job.
  • Four sections per customer: who they are, their job history, equipment to keep an eye on, and site-access notes (gates, dogs, parking, codes).
  • Updates automatically — and only when something actually changes. No wasted compute, no stale info.
  • Cleans up garbled phrasing from call transcripts and double-checks any dollar figure against the actual invoice before quoting it.
  • Never overwrites notes a human wrote. It only refreshes its own section.
  • Spots customers who would have saved real money if they’d been on the Advantage Plan during recent jobs, and tells the tech the exact dollar amount — turning a guess into a data-backed pitch.
For the technically curious
  • Summaries are written to ServiceTitan customer and location records as pinned CRM Notes — no custom UI, no extra app surface.
  • Two-tier pipeline: a cheap "scanner" sweeps ServiceTitan’s modifiedOn watermarks every 15 minutes and queues only customers that actually changed; an LLM-powered "worker" runs every 5 minutes and only generates new prose when the content fingerprint shifts.
  • Source-confidence rules clean noisy plumbing vocabulary, normalize mangled customer names, and reconcile dollar amounts against the actual ServiceTitan invoices before they appear in summaries.
  • Foundation-written notes carry a [CAPSTONE AI SUMMARY · …] marker; refreshes patch only marked sections and leave human-authored notes untouched.
  • Membership-fit math runs deterministically against the Capstone Advantage Plan terms ($360/yr dues, $89 dispatch waived, 10% off services & installs) so quoted savings are reproducible, not hallucinated.

Built on ServiceTitan

If your team runs on ServiceTitan too, this kind of integration likely applies to your business. We know the ServiceTitan API surface, the CRM Notes flow your techs already see, and how to ship work that fits your existing workflow instead of replacing it.

Why this approach holds up

Only does the expensive work when something changed

Costs stay low and summaries stay fresh. Most of the time the system is quietly doing nothing, because nothing needed doing.

Speaks the language of the business

Trained on plumbing vocabulary, not generic transcription — so it catches the words that actually matter on a job.

Respects what your team writes

AI updates only its own section. Notes a human wrote stay exactly as they were, every refresh, every time.

Foundation AI

Want something like this for your business? Let's talk about what AI agents could fit into the tools your team already uses.