Operational review guide updated in 2026

2026 Local Service Follow-Up Leak Review

A local-service inquiry can cross a phone system, website form, estimating tool and inbox before anyone accepts it. That handoff is worth tracing. It does not, by itself, put a dollar value on lost work. This guide shows how to review six common paths using a bounded sample and a human decision at the end.

Last updated: October 2, 2026

By George King, AI Cleanup Doctor

Why “revenue leak” needs a stricter definition

A phone log can show that a call occurred. A form system can show that a request arrived. An estimate record can show that a quote existed. None of those facts alone proves that the business lost revenue. The caller may have been a vendor, existing customer, duplicate, wrong-service request, outside-area inquiry, spam event, complaint, or buyer who chose not to proceed for an unknown reason.

Revenue is a financial outcome. Supporting it usually requires more than an operational event: identity resolution, service fit, a supported opportunity state, a transaction or counterfactual method, attribution rules, and an authorized amount from accounting or payment records. Even then, “lost” revenue may be impossible to prove because the alternative outcome is unknown.

The safer term for most early reviews is follow-up evidence gap. It describes what can be inspected: a missing owner, unsupported status, unverified callback, stale next action, routing mismatch, or human-approval gap. Fixing those conditions can improve visibility and consistency, but it does not guarantee leads, bookings, sales, or revenue.

Use one evidence method across all six zones

1. Freeze the source and period. Record the system, source label, public entry point, date range, filters, time zone, event definition, and extraction time.
2. Choose a mixed sample. Include answered, missed, completed, duplicate, wrong-fit, unknown, and review-first records where available. Do not select only the rows that support a preferred story.
3. Preserve raw evidence. Keep the original label or event before normalizing it. Do not overwrite a conflict between call tracking, CRM, inbox, calendar, or staff notes.
4. Mark Unknown. Blank, inferred, and unknown are different. Unsupported fields should not become negative or positive outcomes.
5. Report denominators and exclusions. State how many records were reviewed, excluded, unresolved, and placed on Hold.
6. Choose one repair. Fix the first evidence-supported path, read it back in the destination system, and define a rollback or stop condition.

What one bounded cleanup should produce

When requests already cross several systems, another tool may only add a new handoff. Make a small review packet for one path first. It should still be useful if the owner decides to change nothing.

• A path map: show the public entry point, first destination, routing rule, assigned owner, acceptance event, first useful response, current state, and evidence location.
• An evidence-priority list: rank no more than five supported defects by record count, customer or safety risk, reversibility, and evidence strength. Do not rank by guessed job value.
• Draft and rule issues: identify where a draft, routing rule, or status definition needs human review. Keep contact authority, suppression, and approval separate from writing.
• An owner-visible board: expose the last verified event, responsible owner, next action, due date, Unknown state, and Hold reason without displaying unnecessary customer details.
• A next-decision note: state whether the evidence supports repairing one path, reviewing more records, or placing the path on Hold. Decisions about buying demand, recontacting estimates, changing staffing, or publishing a weather page remain with the authorized owner.

This packet can cover missed calls, forms, estimates, demand spikes, inbox handoffs, and AI-assisted drafts, but it should not automate all six at once. One reversible change with a destination-system readback is easier to verify than a broad workflow rebuild.

Use a seven-day observation window without calling it proof

A short observation window can reveal whether the repaired evidence path is readable. It cannot prove future revenue or establish a business-wide rate. Before the window begins, freeze the source, dates, time zone, filters, eligibility rules, state definitions, and extraction method. Keep the pre-change records available so the reviewer can distinguish old data from new events.

For each eligible request, track arrival, assignment, owner acceptance, first useful response, connection, supported disposition, and evidence location as separate events. Keep attempted contact separate from connection, connection separate from scheduling, and scheduling separate from completion or payment. Preserve Unknown, duplicate, wrong-fit, opt-out, complaint, and Hold records as distinct states.

At the end of seven days, report the reviewed denominator, exclusions, unresolved rows, supported defects, and whether the destination-system readback matched the intended change. The result is an operational observation, not a sales experiment. It supports a decision about the next review step; it does not support a promise of leads, bookings, revenue, rankings, or AI citations.

Zone 1: missed calls

A call marked “missed” needs context before follow-up. Preserve the event ID, source, timestamp, destination, ring or connection evidence where available, notification state, assigned owner, human acceptance, callback attempt, connection state, and next action. An attempted callback is not a connected conversation, and a connected conversation is not a booked job.

Use controlled states such as missed-unreviewed, owner-assigned, owner-accepted, callback-attempted, connected, wrong-fit, duplicate, Hold, and closed-with-support. The missed-call triage matrix provides a deeper row-level gate.

Zone 2: form handoffs

A successful form submission proves that the website recorded an event under its configuration. It does not prove that the correct team received it or that the buyer obtained a useful response. Review the public entry point, submission timestamp, first destination, routing rule, notification, assigned owner, acceptance event, first useful response, evidence location, status, and next action.

Google Ads explains that advertisers define conversion actions they consider valuable, including website and phone events. Those measurements are important, but the business still needs downstream evidence. Keep the platform event and the operational handoff as separate layers.

Zone 3: old estimates

Old estimates are not a ready-to-contact list. First establish record identity, relationship, service context, last verified event, current eligibility, contact and suppression evidence, owner, current state, and next action. Closed, duplicate, complaint, payment, legal, safety, warranty, insurance, opt-out, and unclear-authority records belong in an appropriate review or exclusion lane.

The old-estimate follow-up template separates eligibility from message drafting. A prepared draft is not approval, sending, delivery, reply, booking, or revenue.

Zone 4: weather and demand surges

Storm, heat, freeze, flood, or emergency demand can increase intake pressure, but the event itself does not prove more qualified demand. Review source, service type, area fit, urgency evidence, first destination, priority rule, available capacity, owner acceptance, safety boundary, status, and next action. Do not let AI assign emergency or safety conclusions from weak context.

The storm-season intake guide covers a focused roofing example. For any service, surge reporting should separate activity volume from handled, qualified, scheduled, completed, and financial states.

Zone 5: owner visibility

An owner-visible board should answer simple questions without exposing unnecessary private data: what record is this, where did it come from, who accepted it, what was the last verified event, what is the supported state, what happens next, when is it due, and why is it on Hold? Counts should link back to reviewable rows.

A dashboard total without definitions can create false certainty. Define each state, keep the denominator visible, and avoid multiplying unresolved rows by an assumed job value. The contractor follow-up board guide describes the owner-facing view, but any financial label still needs financial evidence.

Zone 6: AI-assisted handoffs

Separate data organization, draft creation, human review, approval, sending, delivery, reply, and final disposition. An AI-assisted draft is not an approved customer message. A human reviewer needs the source context, contact authority, policy boundaries, sensitive-data check, claims check, and a clear Hold route.

NIST describes the AI Risk Management Framework as voluntary guidance for managing risks to people and organizations across AI design, use, and evaluation. For a local-service workflow, the practical application is governance: named ownership, evidence, review, stop conditions, and monitoring. The automation readiness guide provides a whole-path gate.

Do not upload passwords, payment data, sensitive customer histories, private notes, or unrestricted inboxes for a first review. Public URLs, redacted exports, field definitions, and representative screenshots are usually enough to map the path. If email or text is considered later, an authorized reviewer should verify sender identity, message accuracy, suppression evidence, a visible stop route, and a human response path before any send. FTC commercial-email guidance and FCC consumer guidance are starting references, not a substitute for legal review of a specific campaign.

Choose a decision lane

How to discuss findings without overclaiming

FTC advertising resources emphasize truthful, non-deceptive claims and support for objective statements. A public “revenue leak” claim can be objective and material to a buyer, so it needs evidence. Safer language stays close to the method:

• “This sample contains records without a visible accepted owner.”
• “The current export does not distinguish attempted callbacks from connected calls.”
• “Several estimates have no supported next action; contact eligibility was not determined.”
• “The platform recorded these actions, while downstream outcomes require separate review.”
• “No revenue amount was calculated because the required transaction and attribution evidence was not available.”

Google Search Central's people-first guidance also favors useful, reliable content that helps the intended reader. A transparent method, visible limitations, and precise terms are more useful than a dramatic benchmark without data.

Frequently asked questions

Is this a statistical 2026 benchmark report?

No. This is an operational review guide updated in 2026. It presents no industry sample, prevalence rate, benchmark, survey, or market estimate.

When can an operational gap be called lost revenue?

Only when authorized business records support the relevant transaction, attribution, counterfactual, and financial amount. A missed call, stale estimate, or unknown status alone is not proof of lost revenue.

How should a contractor report follow-up leak findings?

Freeze the source and period, review a mixed sample, mark unsupported fields Unknown, report counts with denominators and exclusions, and choose one evidence-supported repair.

Can AI decide which records should be contacted?

No. AI may organize authorized evidence, but contact authority, complaints, payment, legal, safety, insurance, warranty, privacy, and unclear records require appropriate human review.