The Glass Takes an Hour. The Insurance Billing Takes the Day.

The instinct for most shop owners exploring AI is to start with something visible — a chatbot on the website, or automated appointment reminders. Those have value. But for an auto glass operation, they're not where the money is hiding.

The right starting point is the claim lifecycle. Specifically: the gap between when a job is scheduled and when a clean claim actually leaves your office. That gap — filled with verification calls, data entry, authorization hunting, and correction loops — is where your staff's time goes and where your cash flow slows down.

Phase 1 should be an insurance intake assistant. Not complex, not expensive, not a wholesale replacement of your current process. A structured intake flow that captures the insurer name, policy number, VIN, and claim type from the customer at the time of scheduling — then checks that information against coverage before anyone picks up the phone to verify manually.

• Build the intake form around the fields your most common TPAs actually require — Safelite Solutions, LYNX, Harmon, and your direct insurer relationships each have slightly different requirements
• Use the AI layer to pre-populate claim fields based on the intake data, flagging anything missing before the job is confirmed
• Set a simple alert if the policy information doesn't match the VIN or the coverage type doesn't include glass
• Keep your staff in the loop — this is a tool that does the first pass, not one that makes final calls

This phase alone, done right, can eliminate a meaningful portion of the phone time your office staff spends on verification. The auto glass industry processed tens of millions of claims annually through third-party networks, and the administrative burden falls almost entirely on the shop, not the insurer. (Source: Auto Glass Safety Council, 2022)

Once Phase 1 is running cleanly — meaning your intake data is accurate and your pre-submission check is catching real errors — you build from there. Supplement tracking, adjuster follow-up automation, and customer communication all connect naturally to the same job record. But none of that works well if the upstream data is messy. Start clean, then expand.

Before. It's 7:45 AM and your office manager, Sarah, is already on hold with Farmers. The customer called yesterday — rock chip on a 2021 Silverado, wants to use insurance. Sarah needs an authorization number before she can schedule the tech. The hold music runs for eleven minutes. She gets the authorization, types it into your management software, and moves to the next intake from last night's voicemails. Two of those voicemails don't have policy numbers. She'll have to call those customers back.

By 10 AM, two techs are in the field. Sarah is reconciling a claim rejection from last week — a VIN transposition error that kicked the claim back from the TPA. She finds the original job record, corrects the number, and resubmits. The clock on that invoice just reset.

The afternoon is supplements. A dealer job from Friday came in short — the insurer didn't approve the full OEM part. Sarah drafts a supplement request, attaches the invoice, and calls the adjuster. She leaves a voicemail. She'll call again Thursday if she doesn't hear back. Meanwhile, a customer texts asking when their tech is arriving. Sarah checks the board, texts back manually, and returns to the supplement queue.

By 4:30, she hasn't touched the three new jobs that came in through the website. Those will be the first thing tomorrow morning.

After. The 2021 Silverado intake came through the online form overnight. The AI assistant captured the policy number, VIN, and insurer, ran a coverage check, and flagged that the deductible is above your minimum threshold — Sarah sees that when she opens her queue at 7:45, skips the hold call entirely, and calls the customer directly with two options. Total time: four minutes.

The rejected claim from last week was caught before submission. The VIN field mismatch triggered a review flag; Sarah corrected it before it went out.

Supplement follow-ups run on a schedule. The adjuster gets an automated email today; if there's no response by Thursday, Sarah gets an alert. The customer text about the arriving tech went out automatically when the job status updated. Sarah's afternoon looks different. Not magical — she's still doing real billing work. But she's doing the parts that actually require a human.

AI won't fix a broken process — it will run a broken process faster and at higher volume. Before building anything, answer these questions honestly.

• Do you know where your claims are rejecting? If you can't describe your top three rejection reasons from the last 90 days, you don't yet have enough process visibility to automate. The AI will need to be trained on your actual error patterns, not generic ones.
• Is your job data in one place? If customer information, insurance details, and job status live in three different systems — or partly in a whiteboard and partly in someone's head — the AI has nothing clean to work with. Integration requires a source of truth.
• Does your office staff have bandwidth to implement? The first four to six weeks will require someone on your team to review outputs, catch edge cases, and give feedback. If your one office person is already maxed, adding a new system during peak season will make things worse before they improve.
• Are you billing enough volume to justify it? If you're running fewer than ten insurance jobs a week, the manual process is probably manageable and the build cost won't pencil out yet. This scales with volume.
• Do you have documented billing procedures — even basic ones? AI systems built on documented workflows outperform those built on tribal knowledge. If only one person knows how you bill Progressive versus how you bill State Farm, that knowledge needs to be captured before it can be systematized.

The auto glass industry has a high rate of owner-operated shops — the U.S. Small Business Administration notes that the majority of auto glass businesses operate with fewer than ten employees, which means the owner is often the de facto billing department. (Source: U.S. Small Business Administration, 2023) That's exactly the scenario where automation has the most leverage — but also where implementation needs to be sequenced carefully so it doesn't disrupt the one person keeping the whole operation running.

If you answered these questions and found real gaps, that's useful information. Fix the process gaps first. Then automate what's working.

The cost of not automating your back office doesn't show up as a line item. It shows up as a slow accumulation of small losses that feel like normal business friction — right up until they're not.

Claim aging. Every day between job completion and clean claim submission is a day your reimbursement is delayed. If a rejection kicks a claim back, that clock resets. Multiply that by the number of insurance jobs you run weekly, and the working capital gap compounds. Auto glass shops operate on thin margins — the national average net margin for glass and glazing contractors runs in the single digits, according to industry financial benchmarks — and delayed reimbursements disproportionately affect cash flow. (Source: Risk Management Association Annual Statement Studies, 2023)

Staff burnout on low-skill tasks. Your office person didn't take that job to spend half their day on hold with insurance companies. When the most repetitive, frustrating work dominates their day, the best people leave — and replacing them means re-teaching every undocumented billing nuance from scratch.

Missed supplement windows. Most insurance networks have defined windows for supplement submission. A shop that's chasing five open supplements manually, with no systematic follow-up, will regularly miss those windows. That's approved revenue that simply never gets collected.

Customer experience gaps. A customer who chose you over a Safelite affiliate made an active choice. If they're waiting two days to hear about their appointment, or texting your office to ask where the tech is, you're competing on service and losing.

• Unanswered voicemails from new customers who called three competitors — whoever responds first usually gets the job
• No-shows caused by appointment confirmation falling through the cracks
• Five-star review opportunities missed because no one followed up after the job closed
• Billing errors that require manual correction eating time that could go toward new customer intake

None of these failures are catastrophic on their own. That's the problem. They feel like the cost of doing business until someone adds them up.