If you spend much time online, you know that artificial intelligence is an ever-increasing presence in our daily lives. Maybe you even use it to organize your grocery list or visualize rearranging your living room. What if AI could also help reduce the workload of damage assessment and parts ordering?

Think of Partly’s Interpreter as the Chat GPT of automotive repair. It’s a repair-specific AI model to help collision repair shops improve estimate accuracy, reduce administrative work, and make better parts procurement decisions. Rather than replacing parts platforms, Partly works alongside them and analyzes estimates, OEM repair procedures, vehicle build data, and shop preferences to identify missed operations, parts, and documentation requirements. It works with all software, with no integration required.

Partly Co-founder and CEO Levi Fawcett says the platform can assist with everything from blueprinting and supplement creation to parts basket optimization, photo annotation, and insurer communications. He emphasized that the system is trained by a large team of engineers and experienced estimators, incorporates current OEM information, and is designed to provide transparent, evidence-based recommendations while leaving final decisions to repair professionals. Early beta users are reporting reductions in research time, administrative workload, and cycle times.

Built for Collision Repair

Unlike general-purpose AI tools, Partly's platform has been trained exclusively on repair-related tasks.

“If you've used something like ChatGPT, or you've seen one of those AI summaries on Google, those are models that are trained to be good at everything,” Fawcett says. “They're very general models, whereas we train a model to be very, very good at just repair.”

The company employs more than 100 people dedicated to building and refining the model, including engineers, AI researchers, and experienced collision repair estimators.

“We're not just some tech nerds who are coming up with ideas in isolation,” he says. “We have real estimators who have been doing this for most of their lives who are helping make sure that the model's doing the right things.”

The goal, Fawcett says, is to create a system that understands the complexities of vehicle repair, from build-specific parts and OEM procedures to common estimating oversights and procurement challenges.

For example, a front bumper may vary based on trim level, production date, ADAS equipment, or market-specific options. Ordering the wrong component can create delays, returns, supplements, and longer cycle times. Interpreter analyzes vehicle data, estimate information, OEM documentation, and historical trends to identify potential mistakes before they occur.

Shop personnel can talk to it as if it were a team member, giving it instructions in plain language – even shop jargon – to assist with writing an email to an insurance adjuster with documentation.

“The model will highlight the damaged area, it will annotate the images, it will explain all the things you could add or maybe the things that were missed off the original estimate and you should consider in your supplement,” Fawcett says. “And then you copy and paste that into your email. There are actually hundreds of these tasks it can do.”

The user can take a photo on his or her phone and plug it into the app to help determine if, for example, how likely it is that the radiator needs to be replaced given the visible damage.

Optimizing the Parts Basket

And for each job, the model can help select the most appropriate parts source.

“They would ask the model, ‘What’s the best basket for the job I’m doing?’ And the model can figure that out based on the insurance carrier rules or based on your preferences: ‘I really prefer OEM parts.’ Or, ‘I don’t like working with this supplier; I really like working with these suppliers. The model will take that into account.”

Subscription cost for the shop is $300 to $500 per month, depending on shop size. And if you’re concerned about AI taking over your shop, you needn’t worry, Fawcett says.

“The model is very accurate. [But] we give everyone controls so they can allow it to do more or less of the work. You can configure it to ask questions every time and not do anything without asking, or you can configure it to be a lot more autonomous.”