AI Production Scheduling for Roofers: Crew, Material, Weather, Dumps

The production board is the bottleneck nobody on the roofing payroll wants to own. Crew skill mix, dump-truck rotation, ABC Supply or Beacon lead time, the 10-day weather forecast, and the signed-job queue all collide on a whiteboard or a spreadsheet that the production manager rebuilds every Sunday night. AI dispatch is the workflow that converts that 4-hour weekly rebuild into a 20-minute confirmation pass — and reclaims roughly 10–14% of crew-hour capacity that would otherwise be lost to idle days, dry runs, and weather scrambles.

This article is a transactional buyer guide for owners and production managers evaluating an ai-dispatcher layer on top of AccuLynx or JobNimbus. It assumes you have already read the operator pillar at AI for Roofing Contractors: The 2026 Operator Playbook and want the configuration detail.

What the production board actually optimizes

Five variables, two constraints, one objective.

The variables: crew (which crew on which day, with skill mix — tear-off speed, steep-pitch comfort, low-slope TPO certification), material (shingle SKU, underlayment, drip edge, fasteners, accessories), dump (10-yard vs. 20-yard, drop time vs. swap time, vendor rotation), weather (10-day forecast at the property zip with hourly precipitation probability and wind), and homeowner availability (the third of jobs that need the homeowner home for the start walkthrough).

The constraints: each crew works one job per day, and material must be on-site by 0700 of install day.

The objective: maximize signed-job throughput, weighted by gross-margin contribution, subject to weather risk.

Most production managers solve this in their head. The good ones solve it well. The problem is that the cost of a single bad swap — crew shows up, material is short, weather turns at 1100 — runs $1,800–$3,200 in lost crew-day plus dump charge plus reschedule cost. Three of those per quarter and you have paid for the AI stack twice.

How AI dispatch actually works

It does not replace the production manager. It does the math and surfaces the top three viable install dates per signed job, ranked by confidence. The PM confirms or overrides.

The pattern: a Claude-driven workflow pulls the signed-job queue from AccuLynx via webhook each evening. It joins against crew availability (from the schedule), crew skill (from a roster maintained by the PM), material lead time (from the ABC Supply portal or Beacon's API where available), dump-truck rotation (from your rotation contract), and the 10-day forecast at each property's zip. It produces a ranked install date for each job. The PM reviews in the morning, accepts 75–85% of the recommendations, and overrides the rest based on knowledge that isn't in the data — the customer is a referral source, the lender disbursement is two days behind, the foreman flagged a roof access issue at the pre-install walk.

The interface is usually a Slack or Teams card with the top three dates per job and a one-sentence explanation of the trade-off ("Day 1: highest crew skill match, dump rotation tight. Day 2: easiest dump. Day 3: best weather window."). The PM clicks accept or override.

The integration map

AccuLynx and JobNimbus both expose the signed-job queue through webhooks and REST. AccuLynx is the more mature surface for storm-restoration shops; JobNimbus is cleaner for retail-leaning operators. Either works. The weather feed is typically NOAA's NDFD plus a paid layer (DTN or Weather Source) for hourly precipitation probability at zip-level. Material lead time is the messiest piece — ABC Supply and Beacon do not expose a clean availability API, so most shops feed a static lead-time table per SKU and have the AI flag outliers for the material coordinator to confirm by phone.

Dump-truck rotation is the easiest integration if you already use a service like Waste Management's commercial portal; for shops on a rotating local-vendor model, it usually starts as a Google Sheet the AI reads at 0500 each day.

Crew routing once the install date is set

The second-order optimization is which crew on which day. A four-crew shop with two A-crews and two B-crews has 24 plausible permutations per install week. AI dispatch ranks them by total expected gross margin — A-crews on architectural-shingle premium retail, B-crews on insurance-baseline 3-tab, with route density factored in. Most shops see drive-time per crew-day drop 20–35 minutes once route density is part of the objective function.

Cross-reference field-service-management for the underlying scheduling primitives.

The 9-day rollout

• Day 1–2: Audit. Pull the last 60 install days from AccuLynx. Categorize each as on-time, late-start (material), late-start (weather), late-start (crew), or aborted. The pattern usually shows a 15–25% non-clean-start rate.
• Day 3: Wire the data sources. AccuLynx webhook into the dispatch tool, weather feed, dump rotation sheet, crew roster with skill mix.
• Day 4: Configure the objective. Start with throughput-weighted; add gross-margin weight once you have one week of data.
• Day 5–6: Shadow. AI publishes a recommended board each morning. PM compares to the board they would have built. Note the deltas.
• Day 7: Cut. AI-generated board becomes the source of truth. PM reviews and overrides.
• Day 8–9: Tune. Look at the overrides from the first week. Half are signal the PM should feed back into the model; half are signal the AI got it right and the PM was being conservative.

ROI math

For a 5-crew shop: 10–14% of crew-hour capacity reclaimed is roughly 280–390 crew-hours per quarter, worth $42K–$58K of incremental gross profit at typical crew loaded cost. The AI dispatch layer itself runs $400–$900/month all-in. Payback inside a single storm month. Full sensitivity in Roofing AI ROI: What a Storm-Restoration Shop Actually Gains.

Industry data published by Roofing Contractor magazine and the Roofers Coffee Shop production-management panels both put crew-day idle cost at $1,400–$2,200 per missed install. The math compounds quickly.

Pitfalls

Three failure patterns.

First: shops that try to dispatch the AI without first cleaning up the crew roster. If the skill matrix is wrong, the recommendations are wrong. Budget two PM-days to update the roster before the pilot.

Second: shops that don't feed weather overrides back into the model. The AI will recommend an install on a 30% precipitation day; the PM correctly overrides because the homeowner is a referral. If that override doesn't go back into the training set, the AI will recommend the same job on the same day next time.

Third: shops that try to optimize for throughput when they should optimize for margin. A B-crew installing a premium retail job at a 22% margin should beat an A-crew installing an insurance-baseline job at a 12% margin almost every time. The objective function matters.

FAQ

Q: Does this replace our production manager? No. It compresses their planning time from 4 hours a week to 20 minutes. The PM still owns the relationships with crews, suppliers, and dump vendors. The AI does the arithmetic.

Q: What if our crews are subcontracted? The dispatch still works. The skill matrix shifts from W-2 crew skill to sub-crew reliability and availability. The objective function adds sub-crew payment terms as a margin variable.

Q: How does this interact with the install foreman's autonomy? The foreman still owns same-day decisions on the roof. AI dispatch sets the install date and crew assignment; the foreman runs the install. Most foremen prefer the AI board because it removes the "why am I on this job today" friction.

Q: What about emergency tarps and storm response? The AI tags emergency tarps as a separate queue with sub-24-hour SLA. They route to an on-call crew, not the main board. GAF and CertainTeed both publish emergency-response certifications that the dispatch layer respects when routing.

Q: How does this handle multi-day commercial jobs? Multi-day jobs are pinned to a crew for the duration and the AI works the other crews around them. TPO and EPDM jobs typically run 3–5 crew-days; the dispatch math respects the pin.

Q: What if material is short? The AI flags the shortage 5–7 days out based on the lead-time table and the material coordinator confirms by phone. If the SKU is unavailable, the job gets pushed and the AI rebalances the board.

Ready to wire AI dispatch into your production board? Book a 30-minute operator review or read the full vertical playbook at /ai-for/roofers.