Free field guide · September 15, 2026
Contractor estimate follow-up: test the handoff before adding AI.
Follow one remodeling inquiry from arrival to an owned next step. Use the ten tests and measurement worksheet to specify what a reliable workflow should do.
By Kisha Wright-Banks · Founder, Agile Brite
The Friday-evening inquiry
A homeowner submits a kitchen-renovation inquiry at 7:20 PM Friday: name, email, town, project description and preferred contact time, but no budget. In this example, office hours are Monday–Friday, 9 AM–5 PM Eastern. A coordinator reviews inquiries and an estimator approves appointments, scope and price. This is a non-emergency remodeling workflow.
The goal is not an instant automated quote. It is one accurate opportunity, one accountable owner and one visible next action.
| Event | System action | Human owner and evidence |
|---|---|---|
| Friday, 7:20 PM: request received | Store the submission ID, source, timestamp and supplied details. Suppress exact retry submissions. | Coordinator owns the next-business-day queue. Retain original submission and opportunity ID. |
| Receipt recorded | Send one approved acknowledgment through a valid channel with accurate office hours. | Coordinator reviews delivery failures. A send event is not proof of delivery. |
| Monday: office opens | Show the request in the assigned queue with an internal due time. | Coordinator reviews project fit, location and missing information. Record owner and review time. |
| Budget missing or scope unclear | Create a clarification task; do not invent values or silently reject the lead. | Coordinator asks the appropriate question and records the disposition. |
| Fit confirmed | Offer approved appointment options; reserve only after calendar confirmation succeeds. | Estimator owns feasibility. Retain event ID, timezone and confirmation. |
| Appointment rescheduled | Cancel old reminders and tie new ones to the current appointment ID. | Coordinator resolves conflicts. Retain old and new appointment history. |
| Estimate prepared | Record approved estimate version and next follow-up date. | Estimator approves price, scope and commitments before sending. |
| No reply to estimate | Example: check in after 2 business days and again after 7; end the automated sequence there. | Owner reviews the remaining open opportunity. These intervals are design choices, not benchmarks. |
| Reply, booking, decline or opt-out | Recheck current state before every send; suppress obsolete queued follow-ups. | Owner responds or closes appropriately. Retain stop reason and timestamp. |
Acknowledgment without overpromising
Thanks for sending your kitchen-project inquiry. We have received it. Our office reviews new requests Monday–Friday, 9 AM–5 PM Eastern. A team member will review your details during office hours and contact you about the next step. This confirms receipt only; it does not confirm an appointment, price or project acceptance.
Replace the hours with the real business's approved information. A promised response deadline needs a staffed escalation path. Keep inquiry handling separate from promotional campaigns; this example does not authorize SMS outreach or marketing enrollment.
Keep the record small and useful
Capture submission and opportunity IDs, source, received time and timezone, supplied contact details, town, project type, original description, contact preferences and permissions, owner, stage, next action and due time, last human response, appointment status, estimate version and stop reason.
Use separate opportunities for genuinely different projects. Do not merge solely because names match. Limit access to customer details and avoid unnecessary sensitive data.
Where AI fits
AI may suggest a project category or draft a summary for staff review. Keep the original request and flag uncertainty. Rules should control ownership, deadlines and stop conditions. People retain pricing, construction scope, start-date promises, safety decisions and exceptions.
Ten tests before launch
These are proposed acceptance tests, not completed tests. Use synthetic data in a test environment. Record tester, date, actual result and evidence. If a test fails, pause the affected automation and use the manual queue.
| Test input | Expected result | Actual result / evidence |
|---|---|---|
| 1. Same form event twice | One opportunity and acknowledgment for that event. | |
| 2. Same person, different project | Separate opportunity or explicit human review; no silent loss. | |
| 3. After-hours submission | Accurate acknowledgment and next staffed-period task. | |
| 4. Missing budget or unknown town | Human clarification, not an invented value or automatic rejection. | |
| 5. CRM write failure | No false success; retry or a visible, owned exception. | |
| 6. Bounced email | Record failure, stop repeats to that address, create review task. | |
| 7. Competing appointment bookings | Only confirmed availability is booked; safe retry path for conflict. | |
| 8. Reschedule | No canceled-slot reminder; correct new time and timezone. | |
| 9. Reply just before follow-up | Recheck state immediately before send; suppress obsolete message. | |
| 10. Decline, opt-out or close | Stop queued follow-ups; no later message escapes. |
Measure the pilot honestly
Start with a baseline and a pilot using consistent definitions. Thirty days is a planning starting point, not a guarantee of enough data. Extend observation when volume is low. Keep counts with percentages and segment inquiry source, project type and business-hours arrival.
| Metric | Definition | Baseline | Pilot |
|---|---|---|---|
| Eligible inquiries | Unique genuine projects; document spam and retry exclusions. | ||
| First human response | Median receipt-to-substantive-human-response time; report staffed-time separately. | ||
| Owner coverage | Assigned within agreed target / eligible inquiries. | ||
| Estimate booking rate | Eligible inquiries with a booked estimate / eligible inquiries. | ||
| Attendance rate | Attended / appointments due; document cancellation treatment. | ||
| Follow-up completion | On-time completed tasks / eligible tasks due; exclude correct suppressions. | ||
| Administrative effort | Staff-recorded minutes per opportunity; same sampling method. | ||
| Safeguards | Count duplicates, messages after stop, delivery failures and unresolved exceptions. |
Automated acknowledgment time is not human-response time. Use equal outcome observation windows for inquiry cohorts. Record staffing and process changes; a before/after difference alone does not establish causation.
From example to evidence
A real case study requires a participating contractor's permission, baseline data, agreed scope, test evidence, sample sizes, observation windows and limitations. Have the contractor approve public names, quotations and results. Until then, this remains a representative workflow design.
Related: AI workflow automation for service businesses · New Jersey AI automation consulting