Case Study · Sample Build
Keystone Plumbing & Drain is a representative sample business used to design, configure, and validate every AI product in this suite. 850 customers, 2,400 jobs, 420 reviews — all synthetic, all realistic, all actionable. The demos below are live.
About this case study: Keystone Plumbing & Drain is a fictional Philadelphia-area plumbing and light-HVAC company constructed as a representative sample of the service businesses Forrest Analytics Group targets for AI implementation. The dataset spans April 2024 – April 2026: 850 customers, 2,400 jobs, 3,200 inbound calls, and 420 reviews. Every AI product below was designed, built, and validated against this data. Results shown are derived from the dataset, not projected or hypothetical.
Business Profile
A residential and light-commercial plumbing operation doing what most well-run shops do: solid work, steady repeat customers, and a few expensive blind spots.
What the Data Revealed
The audit ran across call logs, job history, customer records, and review activity. Here's what surfaced, and which tool was built to address each one.
Out of 3,200 inbound calls over the two-year dataset, 908 were never answered. The pattern was consistent: afternoons during peak field hours, evenings after 5 PM, and all day Saturday and Sunday. The business was fully operational — the phones just weren't being covered.
The dataset's own summary flags $621,163 in missed-call revenue risk — that's the product of unanswered call volume, average conversion rate, and average job ticket. At $1,370 per job, every ten missed calls is roughly $8,200 in revenue that went to a competitor.
Solution
Answers every call within two rings — after hours, on weekends, when the office line is tied up. Triages emergencies, collects the caller's info and problem, books an appointment from live tech availability, and texts a confirmation before the call ends. Nothing hits voicemail.
See This Product →Of 850 active customers in the dataset, 357 had no job, no contact, and no outreach in over a year. Segmentation by service type revealed many were statistically overdue: water heaters aging past the efficiency window, sump pumps installed before last spring's flood risk, drain cleaning customers 18+ months out.
These aren't cold leads — they're past customers with a positive prior experience and a known service history. Reactivating even 10–12% of that segment generates $25,000–$45,000 in incremental annual revenue at zero acquisition cost.
Solution
Segments the customer list weekly by service type and recency, writes personalized SMS and email outreach that references the customer's actual job history, and projects recovered revenue by segment. Owners see a weekly dashboard; customers get a message that sounds like it came from a human who knows them.
See This Product →The dataset has 420 reviews across Google, Yelp, and Angi, averaging 4.5 stars. Of the 27 reviews rated 1–2 stars, only 45% received a response. The replies that did go out were slow (often 4–7 days) and generic. Response recency and rate are active Google local ranking signals — every unanswered negative review is a drag on organic search placement.
The pattern analysis also surfaced something the owner couldn't see manually: one technician, David Chen, had 3 of the 6 attributed negative reviews in 90 days — a coaching opportunity that was going unnoticed without a system to aggregate the signal.
Solution
Drafts a personalized, on-brand response to every review within minutes of it posting. Owner approves with one tap. Negative reviews get an acknowledgment and make-right offer; positive reviews get a genuine, specific thank-you. Pattern analysis flags any technician with repeat negative mentions for proactive coaching.
See This Product →Call data lived in one system. Job history in another. Review activity required logging into three different platforms. Revenue trends required pulling from accounting software. No single view existed — which meant operational decisions were made on instinct and lag, often weeks after a problem had developed.
In one stretch of the dataset, call booking rates dropped 14% over three weeks before the dip was visible. By the time the pattern surfaced, two techs had already had light weeks that could have been filled with a single morning alert. The owner needed one signal, not five logins.
Solution
Every morning at 6am: yesterday's revenue vs. same day last year, today's full schedule, any invoices past 14 days, any unanswered negative reviews, any big jobs in the pipeline. Delivered as a single plain-English SMS. No dashboard. No login. The owner opens their phone and knows what matters in under 60 seconds.
See This Product →Live Demos
Every output below is derived from the Keystone dataset. These aren't mockups — they're what the tools generate when pointed at real operational data.
Every morning at 6am, Claude reads Keystone's live dataset and writes this SMS. Revenue vs. last year, today's schedule, open invoices, reputation alerts — all in under 180 words. No dashboard. No login.
Claude drafts a personalized reply within minutes of a review posting. The owner approves with one tap — or edits the text and approves. The pattern analyzer runs in the background, surfacing any tech with repeat negative mentions.
Dawn answers every inbound call — first ring, 24/7. She triages emergencies, collects caller info, pulls available appointment slots from live tech schedules, books the job, and texts a confirmation before the call ends.
Every week, the engine segments dormant customers by service type and recency, then writes personalized outreach for each one. Click a segment to see the sample message Claude generates for a real customer from the dataset.
The Numbers
Each figure below comes from the Keystone dataset or from documented industry benchmarks for service-business AI implementations.
| Metric | Before | After AI |
|---|---|---|
| Answer rate | 71% | 100% |
| After-hours answered | 31% | 100% |
| Missed-call revenue risk | $621K/yr | ~$0 |
| Office staff hours saved | — | 8–12 hrs/wk |
| Metric | Before | After AI |
|---|---|---|
| Negative review response rate | 55% | 100% |
| Avg response time | 4–7 days | < 1 hour |
| Organic call volume (est.) | Baseline | +8–15% |
| Owner time per week | 2–3 hrs | 10 min |
| Metric | Before | After AI |
|---|---|---|
| Dormant customers identified | 0 | 221 segmented |
| Reactivation rate (est.) | — | 10–15% |
| Annual incremental revenue | — | $25K–$45K |
| Time to draft 221 messages | Never done | < 10 min |
| Metric | Before | After AI |
|---|---|---|
| Morning ops review time | 45 min | < 5 min |
| A/R issues caught early | Reactive | Daily flag |
| Negative reviews caught | When noticed | 24-hr window |
| Open A/R (current) | $175K | Flagged daily |
Ready to Run Your Own Numbers?
The Keystone audit took about 15 minutes to set up and immediately surfaced six figures in recoverable revenue. The free audit call does the same thing for your business — no prep required, no commitment expected.
Request Your Free 30-Minute AuditM.S. Business Intelligence & Analytics (Saint Joseph's University) · Instructor, AI for Business (Saint Joseph's University). The Keystone dataset was designed and built by Josh as the foundation for every product in this suite. All AI outputs shown above were generated by Claude claude-sonnet-4-6 against the real dataset.