The Paperwork Lost.

Wednesday afternoon. A technician at Location 2 replaces a power steering pump on a 2023 Honda CR-V with 28,000 miles. The repair order notes the part and labour. The service advisor closes the ticket and hands the keys back to the customer. Job done.

The power steering pump on a 2023 CR-V with 28,000 miles is covered under Honda's powertrain warranty. The shop is entitled to reimbursement: $185 for the part (OEM price) and $94 for labour (Honda's flat-rate time at the warranty labour rate). $279.

The service manager at Location 2 should identify this as warranty-eligible, pull up Honda's warranty portal, enter the VIN, the repair code, the technician's documentation, and submit the claim within 30 days.

He doesn't do it Wednesday because the shop is busy. Thursday he's dealing with a customer complaint about a brake job. Friday he's short a technician and covering bays himself. The following week, he files 4 warranty claims from a stack of 9 eligible repair orders. The Honda CR-V is in the pile he'll get to next.

Four weeks later, the filing window closes. $279 that the shop earned, performed the work for, and never collected.

Across three locations, this pattern repeats 672-840 times per year. $191,520-$239,400 left on the table. Not because anyone decided not to file. Because the filing competed with running the shop, and running the shop always won.

Multi-location independent auto repair chain in Phoenix, Arizona. Three locations. Thirty-four employees. $5.1M combined annual revenue. Roughly 14,000 repair orders per year.

Nick owns the chain. He runs it from aggregated quarterly financials and a daily habit of checking in at each location. The warranty claims gap is the number he finds most uncomfortable at the quarterly accountant meeting: the difference between what the shops should have been reimbursed and what they actually collected.

The numbers: warranty-eligible repairs estimated at 12-15% of total volume (1,680-2,100 per year). Average reimbursement per claim: $285. Total annual warranty entitlement: $478,800-$598,500. Claims actually filed and paid: roughly 60%. Unclaimed: $191,520-$239,400.

Of the claims that were filed, 18% were rejected for incorrect documentation, wrong manufacturer codes, or missing information. Estimated recoverable if corrected and resubmitted: $31,000-$42,000. But rejected claims cycle back to the same service managers who didn't have time to file them correctly the first time.

Each location's service manager spends 3-5 hours per week on warranty administration: identifying eligible repairs, cross-referencing VINs against manufacturer warranty databases, translating repair descriptions into manufacturer-specific codes, navigating warranty portals (each manufacturer has its own), and following up on pending claims. Combined across three locations: 468-780 hours per year. At $32 per hour loaded: $14,976-$24,960 in annual admin time.

Total annual cost of the warranty gap: $237,496-$306,360 in unclaimed reimbursements, rejected claims, and admin time.

Nick discovers the magnitude of this at the quarterly review. By then, the filing windows on the oldest unfiled claims have closed. The conversation with the service managers is, by Nick's description, "the same conversation every quarter." They were busy. The claims were in the pile. The pile grew. The windows closed. Next quarter, the same conversation.

Four stages. The core insight: the service managers aren't failing at warranty administration. They're being asked to do warranty administration as a secondary task inside a primary job (running a shop floor) that never has a quiet afternoon. The filing needs to happen. The identification, the documentation, and the formatting don't need a service manager. They need a system.

Stage 1: Warranty eligibility detection

Scans closed repair orders from Mitchell 1 daily. For each RO: decodes the VIN (year, make, model, engine, trim via NHTSA API), checks mileage against the warranty reference database (powertrain coverage, bumper-to-bumper, emissions, by manufacturer), and cross-references the repair type against covered components.

A power steering pump replacement on a 2023 CR-V with 28,000 miles: flagged within 24 hours of the RO closing. The service manager didn't need to identify it. The agent identified it because it can cross-reference a VIN, a mileage reading, and a repair code against a warranty database faster than a service manager can walk to the parts counter.

Stage 2: Claim pre-population

For each eligible repair: generates claim documentation in the manufacturer's required format. VIN, RO number, date of service, repair description in the manufacturer's terminology (not the shop's shorthand), parts used with OEM part numbers (cross-referenced from Mitchell 1's parts catalogue), labour time at the manufacturer's flat rate, technician certification number.

The service manager receives a notification: "RO #47832 (2023 Honda CR-V, power steering pump) is warranty-eligible. Claim pre-populated. Review and submit. Filing deadline: [date]. Estimated reimbursement: $279."

Review: 60 seconds. Is the repair accurately described? Does the claim match what was done? Confirm and submit.

Stage 3: Submission and follow-up

On confirmation, the claim submits through the manufacturer's warranty portal (API where available, generated PDF where not). The agent tracks claim status. Rejected claims return with the specific rejection reason and a suggested correction. The service manager fixes the specific issue rather than rebuilding the entire claim. Approaching-deadline claims escalate automatically, with the dollar amount at risk.

Stage 4: Claims dashboard and revenue recovery

Weekly view: claims by status (identified, filed, pending, paid, rejected, expired), reimbursement amounts (claimed, received, outstanding), filing rate per location, rejection rate per manufacturer. Monthly reconciliation against QuickBooks warranty receivables.

Nick sees the number every Monday morning instead of every quarter. The gap between expected and actual reimbursements is visible in real time. The quarterly conversation about the pile becomes a weekly glance at a dashboard where the pile doesn't accumulate.

The warranty reference database is the foundational work. Ten manufacturers cover roughly 85% of what rolls through a typical independent shop. For each: powertrain coverage terms, bumper-to-bumper terms, emissions terms, covered components, filing windows, required documentation format. Six to ten hours of initial setup. Updated quarterly. This is the equivalent of Victor's recipe digitisation and Graham's knowledge capture sessions: tedious, done once, worth every minute.

The aftermarket-to-OEM part number translation is the detail that prevents rejections. Shops install aftermarket parts. Manufacturers reimburse at OEM pricing. The claim must reference OEM part numbers, not aftermarket. Mitchell 1's parts cross-reference handles this, but the service managers weren't consistently translating. The agent cross-references automatically. In the first quarter, this prevented an estimated 40-50 rejections that would have bounced back to the pile.

The escalation timeline creates financial accountability. Notification at day 0. Reminder at day 7. Escalation to Nick at day 14 with the filing deadline and the dollar amount at risk. When Nick receives "Location 2 has $4,300 in warranty claims approaching deadline in 16 days, un-reviewed," the conversation happens. The same pattern as Dana's renewal escalation: financial visibility creates urgency that a pile of paperwork doesn't.

Recommended stack: n8n for orchestration. Mitchell 1 API (or equivalent: Shop-Ware, Tekmetric). NHTSA VIN decoder API (free, no authentication). Claude Sonnet for eligibility analysis and claim generation, Haiku for RO classification. Postgres for claim tracking and warranty reference data.

Step 1: Set up infrastructure and warranty data (Days 1-3). Deploy n8n. Configure Mitchell 1 API. Set up NHTSA VIN decoder. Build warranty reference database in Postgres for the top 10 manufacturers: coverage terms, duration, mileage limits, covered components, claim formats, filing windows. Set up Postgres for claim tracking.

Step 2: Build RO scanning and eligibility detection (Days 3-5). Schedule Trigger daily at 06:00. Fetch closed ROs from Mitchell 1. For each: decode VIN via NHTSA. Query warranty database. If potentially eligible: Claude Sonnet analyses repair type against coverage. If covered: create claim record, notify service manager.

Step 3: Build claim pre-population (Days 5-7). For each eligible claim: Claude Sonnet generates documentation in the manufacturer's format. Translate aftermarket part numbers to OEM via Mitchell 1 cross-reference. Output: pre-populated claim ready for 60-second review.

Step 4: Build submission and follow-up (Days 7-9). On approval: submit to manufacturer portal (API or generated PDF). Track status. Rejected claims return with parsed reason and suggested fix. Approaching-deadline claims escalate to Nick.

Step 5: Build dashboard and reconciliation (Days 9-10). Weekly dashboard: claims by status, amounts, filing rate, rejection rate. Monthly reconciliation against QuickBooks warranty receivables.

Step 6: Test and refine (Days 11-14). Test against 30 historical ROs (15 eligible, 15 not). Compare agent eligibility against service manager assessment. Verify claim accuracy against 5 approved historical claims. Pilot at one location for 2 weeks.

Estimated build time: 12-14 days for a competent n8n developer. 4 weeks if learning alongside.

Monthly running costs:

Claude API detail: RO screening via Haiku (~1,167/month): $0.12. Eligibility analysis via Sonnet (~160/month): $1.50. Claim generation via Sonnet (~160/month): $2.00. Rejection analysis: $0.10. Total API: approximately $3.72/month.

Build costs if hiring: 12-14 days at $400-$600/day = $4,800-$8,400. Self-built: $0.

Year-one total: $5,256-$9,360 (with developer) or $456-$960 (self-built). Compared against $237,496-$306,360/year in total warranty gap cost. Capturing just 50% of previously unclaimed reimbursements ($95,760-$119,700) against the agent cost produces a ratio of 10-250:1.

Warranty term data is incomplete. Start with the top 10 manufacturers. Verify against published warranty guides. Update quarterly. The service manager review catches errors. Track override rate: above 10% means the reference data needs refinement.

Manufacturer portal access for independents. Some portals require dealer credentials. Identify which manufacturers your shops can file directly with and which require a warranty processing service (NAPA AutoCare, Technet). The agent identifies eligibility and pre-populates regardless; the filing route varies.

Mileage data accuracy. Warranty eligibility depends on mileage. If the RO mileage is wrong, the determination is wrong. Flag claims where mileage is close to the warranty limit (within 5,000 miles). Service manager verifies from the odometer reading.

Service managers still don't review promptly. Escalation timeline: notification day 0, reminder day 7, escalation to Nick day 14 with dollar amount at risk and filing deadline. Financial visibility creates accountability.

Manufacturer claim format changes. Monitor rejection rates per manufacturer. A spike signals a format change. Update the claim generation prompt. Quarterly portal review.

Aftermarket parts versus OEM reimbursement. The agent translates part numbers via Mitchell 1 cross-reference automatically. Flag to service manager when no OEM equivalent exists.

If your business performs work covered by manufacturer warranties and your filing process depends on the same people who run the shop floor, the gap between what you're owed and what you collect is larger than you think. The service managers aren't failing. They're prioritising correctly: the customer in the bay is more urgent than the claim in the pile. The claim will wait. Until it expires.

The agent doesn't change what service managers prioritise. It removes the identification, the documentation, and the formatting from their plate. They review in 60 seconds what used to take 20-30 minutes to assemble. The pile shrinks because the work that created the pile is done by a system instead of by three people who have a more urgent job to do.

The Monday morning bottleneck here is quarterly instead of weekly, which makes it worse, not better. Nick discovers the gap four times a year. The claims expire twelve months a year.

This is Blueprint #51 in the AdAI series. Every week we publish the full architecture of a real AI agent design. Free to read. Free to build from.

Want the next one? Subscribe to AdAI News. New blueprint every Thursday.

by TC
for the AdAI Ed. Team