Twitter You Tube Facebook Autobodynews Linked In

Wednesday, 29 May 2019 13:38

Evaluating AI Tools and Applications: Tips From Tractable AI

Written by

Index

Technological advances in the collision repair industry are an ongoing challenge; however, when utilized properly, many of these advances, such as artificial intelligence (AI), can be a huge benefit during the repair process.

Advances, such as Tractable, help develop AI for accident and disaster recovery.

 

“Tractable’s market-leading tools use photos to automate damage appraisal—allowing partners to improve accuracy, reduce turnaround time and deliver a revamped customer experience,” said Ahmed Zifzaf, senior partnerships and marketing manager at Tractable. “One of Tractable’s AI tools is their ‘review product,’ which helps automate estimate desk reviews, by comparing the photos of a collision with the estimate that was written.”

 

Although it’s common knowledge that AI photo-analysis solutions are increasingly being marketed across the industry and promise to lead to major changes in the repair process, it isn’t always clear how well these solutions work. While it may seem daunting to determine how well an AI photo system works, Tractable’s educational video shares three simple steps collision facilities can use to test such solutions. The first step is to obtain a set of photos the AI has never seen. Secondly, upload the photos on the spot to the AI system and lastly, evaluate the results yourself.

 

 

Tractable’s AI tool generates estimates of repair or replace decisions and a likely range of labor hours, based on the uploaded photos. During a demonstration earlier this year, Alex Dalyac, Tractable CEO, noted that the range is meant to address the “notoriously ambiguous task” of determining labor hours since experts will estimate different labor times for the same procedure. Users can configure their AI tool to specify the number of hours for a certain function.

 

“Our AI can predict bodywork, R&I and paint time,” said Zifzaf. “In general, these need to be adjusted to the client as opinion times vary greatly across North America (e.g. based on the type of insurer, repair facility, etc.). In the demo, our AI outputs a range of hours instead of specific hours, but in production, we calibrate to the particular client or market area.”


 

When it comes to the decision of repair vs. replace, Tractable AI begins with part recognition, then moves onto damage assessment and then lastly, it makes a recommended decision based on its confidence level. The system’s confidence level will depend on part visibility, photo conditions and extent of the damage. This means the AI can also interact with the person taking photos to request specific additional photos on the spot.

 

“Confidence level can be seen as a proxy for how certain the AI is making the correct decision,” Zifzaf notes. “The decisions made can be calibrated to specific geographic areas, carriers or shops.”

 

Tractable’s AI technology specializes in exterior damage, but it can also be used to predict damage to things like sensors via decoding the VIN and predicting the damage associated with an impacted panel. However, Tractable recommends that shops still perform a diagnostic scan on the affected components that are unseen by photos.

 

“Our research is driven by visual algorithms, so we can get an accurate assessment of what happened to the exterior, but we recommend that shops still do their due diligence to see what other damage there may be internally,” said Zifzaf.

Tractable’s AI software is able to “learn” from previous photos, said Zifzaf.

 

“The AI is trained on millions of photos from opt-in customers and partners,” said Zifzaf. “Algorithms can work on any make and model to look at different photos and severities. This is how it learns to calculate the corresponding damage and labor hours in the future.”

 

AI photo technology can benefit shops in several key ways. Shops can use AI for triage, Zifzaf explains, “It is inefficient for body shops to receive vehicles that are total losses. AI total loss appraisals take seconds and can help ensure the right vehicles make it to the salvage yards instead of the shops.”

 

AI technology can also be used for self-auditing purposes, especially for larger MSOs. Zifzaf says, “They can get started by augmenting or automating desk reviews using photo AI systems. This can allow shops to self-manage accuracy; while, allowing shop employees to spend more time focusing on customer needs and repairs. It can also help MSOs identify and promote top-performing shops.”


 

Additionally, shops can use AI to automate and streamline estimate writing and interactions with insurers.

 

“Body shops can calibrate AI to their accurate repair standards and reduce supplements and unnecessary interactions with the insurer,” according to Zifzaf. “This allows them to spend more time repairing cars and taking care of customers, instead of wasting precious time on other tasks that can be automated.”

 

Zifzaf said using AI photo technology for audits can reduce risk and liability “by increased compliance and standardization across estimates. As algorithms give consistent answers, it allows for consistent decisions to be made and therefore reduces risk and liability. A well calibrated and accurate AI is at the core of this.”

 

Tractable is not the only AI technology available on the market. CCC offers an AI photo-estimating product, and in 2017, Audatex announced its partnership with hail photo-estimating provider CSI. When it comes to evaluating which system to invest in, Zifzaf advises, “As a first step, make sure the photo-analysis system really works. Ask to upload your own photos to the system and judge the results for yourself.”

 

“Tractable was found in 2014 and is backed by $30M in venture capital,” said Zifzaf. “We are in production with top insurers globally and have processed over a million auto claims. Now, we are looking to engage with repair facilities on how our technology could best help them.”

 

Feedback can be provided to ahmed@tractable.ai.

Read 1110 times