Turning the Review Follow-Up Process Into a Running System

A six-person restoration company finished 40+ jobs a month and collected 2–3 Google reviews a quarter. Not because customers hated them. Because follow-up lived in one person’s head and lost every time the phone rang.

This is a process story, not a tools story.

The problem (knowledge + opportunity loss)

After a job closed in the field system, the intended flow was:

1. Tech marks the job complete
2. Office pulls customer details
3. Someone drafts a thank-you in email (templates varied by tech and mood)
4. Message sends three to five days later — if remembered
5. If a review appears, someone might notice … eventually

On busy days — most days — step 4 did not happen. Reviews tracked owner attention and office bandwidth, not job quality.

That is classic knowledge loss (process in one skull) stacked on opportunity loss (reputation and referral fuel never harvested). Time loss showed up as copy/paste and template hunting. Margin loss showed up later as uneven brand proof when bidding against competitors with thick review profiles.

Binding constraint

The scope work forced one sentence:

There was no automatic trigger connecting job complete in the field system to a reliable, branded follow-up sequence.

Everything else was symptom. Fix the handoff, and the rest can be designed once.

What “done” meant

Before anyone wrote automation rules, success was defined:

• Every eligible completed job enters a follow-up path
• First thank-you after a short cool-down (not while the van is still in the driveway)
• Review request on a known cadence with a clean Google link
• Owner alert on 1–2 star outcomes before public spirals
• A weekly view of job → follow-up state so silence is visible

Edge cases that blocked scope creep later:

• Insurance-heavy jobs with sensitive wording
• Customers who opted out of marketing-style mail
• Repeat customers already saturated
• Jobs reopened / recalled within the cool-down window

The build (narrow agentic system)

From the approved playbook we shipped a small system, not a science fair:

• Trigger on job-complete events from the field/dispatch system of record
• 24-hour cool-down before the first message
• Personal thank-you using customer name + job type from the job record
• Day-3 review request with the correct Google review destination
• Low-star alert to the owner for personal recovery (skip the generic push)
• Weekly pipeline dashboard — every job’s follow-up status, overdue or closed

Humans stayed in the loop for angry/low-star paths and for any message class we marked as judgment-heavy. The agent owned tracking, timing, and the boring yes-path.

Results (first 60 days)

• ~120% increase in Google reviews vs prior run-rate
• ~5 hours/month returned to the office manager (no more manual chase list for this flow)
• Owner visibility on negative signals in minutes, not weeks
• Coverage: follow-up became the default path, not a heroic memory trick

Numbers like these are shop-specific — your cadence and ticket mix will differ. The pattern holds: reliability beats intensity.

Why it worked

1. Scope first. Binding constraint and edge cases were written before build.
2. One workflow. We did not also “fix marketing,” “fix dispatch,” and “add a chatbot” in the same sprint.
3. Measure what owners feel. Reviews, hours, alert latency — not model vanity metrics.
4. Keep humans on the expensive exceptions. Agents are superb at not forgetting. People are superb at de-escalation.

How to steal the pattern for your shop

Use this checklist on any follow-up process (reviews, estimates, maintenance agreements, payment reminders):

1. Name the start event already in a system of record
2. Write the happy path as numbered steps with timing
3. List exceptions that must never auto-send
4. Define one success metric and one safety metric (e.g., complaints caught fast)
5. Build only what the playbook names — then run it long enough catch glitches

If you cannot complete steps 1–4 on a whiteboard, you are not stuck for AI talent. You are stuck for clarity.

Where this sits on the Agentic Solutions ladder

1. Consult / Four Losses — find whether reviews (or estimates, or invoices) is the binding constraint
2. Scope — BPM-style playbook with triggers, tools, escalations
3. Build — narrow agent + connectors + dashboard
4. Run — light host-and-maintain so the sequence does not die when someone goes on vacation

The tooling is available. The method is available. The question is whether the process is worth making boringly reliable — for most service businesses with volume, it is.

---

One example of operational knowledge turned into a running system. Start with the first 30 pages free, or get in touch if you want help scoping your version.