Belron’s brand, Autoglass® BodyRepair, manually assessed customers’ vehicle damage to generate quotes, a process that was time consuming and also prevented the company from providing an online digital solution.
Autoglass® BodyRepair applied IBM Watson Visual Recognition technology to its website’s booking page, which visually analyzes customer photos of vehicle damage and automatically recommends a repair price.
>70% improvement in quote processing timesby using AI analyses to determine repair costs
Optimizes damage advisors’ time and skillsby allowing employees to focus on more complex claims
Boosts efficiency of the quote generation processby supporting human decision-making with AI analyses
Business challenge story
Manually assessing auto damage
Dings, dents and scratches. No matter how they happen, cosmetic damage to a vehicle can be painful. Unfortunately, for some consumers, the process of having the damage assessed for repair can be equally as painful. Autoglass® BodyRepair, a UK-based Belron brand that offers mobile repair of vehicles with accident damage, is changing that.
On average, 260 people visit the Autoglass® BodyRepair website daily to request quotes. Until recently, a team of three damage advisors examined photos that customers sent in through a portal or via email. They assessed the vehicle damage and generated repair quotes from what they could see and by reading the damage notes on the system. The process could take an hour up to a couple of days as the team waited for images to be sent in.
Dafydd Hughes, IT & Digital Manager at Autoglass® BodyRepair, elaborates: “In the past, customers would fill in a basic form on the website and attach a few images of the damage. These would land in a damage advisor queue where we would examine the images and either calculate a price or have to get back in touch with the customer to request more images or more detail. This can prove very tricky and sometimes we could be phoning them back for a couple of days to try and get hold of the customer to get them to send another photo or provide more detail.”
Hughes, a skilled software developer with a keen interest in AI technology, saw an opportunity to make improvements. “We came to realize that we should be able to use customer photos and AI to calculate repair costs,” he says. “It was about making the processes as efficient and hassle free for the customer as possible.”
To this end, Autoglass® BodyRepair investigated AI options from Google and Microsoft before choosing IBM Watson technology. “I discovered that there was a visual side to Watson that would fit our needs,” says Hughes.
In May 2017, Belron sponsored a company-wide Watson hackathon dedicated to finding ways to use AI across the organization. Hughes and his UK team focused on the technology’s visual recognition capabilities to identify types of damage, such as scratches, dents, bumps and scuffs.
Using AI to analyze damage and costs
Today, Autoglass® BodyRepair assesses vehicle damage and generates customer quotes using the IBM Watson Visual Recognition service. Customers simply upload images of their vehicle damage on the website and enter some basic details about themselves and the vehicle. If they meet the requirements, the Watson service issues a quote. Customers can then book a service time and enter credit card details to facilitate payment after the business makes the repairs. When the service cannot calculate a quote, the company contacts the customer back and determines costs using the normal process.
The IBM Watson service uses a library of roughly 2,000 images to analyze and organize customer photos based on four classifiers: type of vehicle, mobile repairable, product code and technician. From these findings it is then able to determine repair costs.
Hughes explains the process: “First it identifies whether the image is of a car and then validates that we can repair it. If we can, it categorizes the damage using one of 12 product codes related to the type of damage, such as a scratch, dent or a big dent. After that, it assesses the minimum skill level of the technician required to make the repair, such as a flat liner or cosmetic technician. Then it recommends a price based on these selections.”
For each classifier, the system generates a score that reflects its confidence in its analyses of the images. In turn, those scores are measured against each classifier’s threshold, or how confident the company is that the classification is right without having to forward the request to the damage advisors, as set by Autoglass® BodyRepair. For instance, the threshold for the vehicle classifier is 37 percent; for the technician classifier, it’s 60 percent. Any queries generating confidence scores lower than the established threshold are automatically forwarded to human damage advisors for review and processing, although, currently, the company ultimately audits all transactions for accuracy while evolving the solution.
“The scores act as safety nets to prevent offering a price that could be too low or high,” says Hughes.
To get to this level of analyses, Hughes and his team spent six months training the Watson service, searching through nearly 50,000 images of different types of vehicle damage to select the best images to use. After training, the team spent just four months developing and deploying the solution, which went live in September 2017. “And that included rebuilding the entire website from the ground up to replace what was there,” adds Hughes.
Generating repair quotes faster
The work is paying off. According to Hughes, Autoglass® BodyRepair is the first in its industry to use AI to provide customers with auto repair quotes. Not only are quotes generated faster, but also damage advisors are free to perform higher-value work. And the overall quotation process is far more efficient.
“When Watson successfully prices a repair without any human intervention, there is a 70 percent or more improvement in processing time,” says Hughes.
Hughes continues to enhance the solution’s capabilities, such as building a new product code classifier to improve the hurdle rate. “We’re currently at about a 30 percent hurdle rate, but on target to hit 45 percent, or pricing 45 in 100 jobs,” says Hughes. “I believe we’re close to reaching that score.”
Even with the current hurdle rate, the process is much improved. “We’ve improved the customer experience significantly. By the time our advisors are involved, we have already collected all the data required to build a quote. This has resulted in our advisors handling fewer but more complex claims, ensuring we are using their skills in the best possible way.
We are also looking at using verbatim text analysis to boost confidence scores. If a customer types ‘I scraped my car along a trolley,’ we could link that text to the type of damage we’re seeing on the photo and bolster the image results,” adds Hughes. In the future, Hughes wants to employ object detection, whereby the technology pinpoints specific parts of an image, such as a scratch that extends across two panels of a vehicle, and actually counts the number of points of damage.
For now, Hughes is more than satisfied with results. “We’ve learned so much and we’re happy with how everything is working at the moment. We’ll continue to make improvements to the process ensuring we’re giving the best possible experience to our customers.”
Founded in 1897 and headquartered in Egham, England, Belron is the global leader in vehicle glass repair and replacement, serving more than 16.5 million customers in 34 countries. Recently, the company expanded services to include cosmetic auto body repair through more than 10 retail brands. These include Carglass in Europe, Autoglass® and Autoglass® BodyRepair in the UK, and Safelite Autoglass in the U.S. Belron employs more than 26,000 people worldwide and reported sales of EUR 3.5 billion in fiscal year 2017.
Take the next step
To learn more about the IBM solution featured in this story, please contact your IBM representative or IBM Business Partner, or visit the following website: