To use information from previous repair jobs to provide more detailed, accurate quotes, leading to increased conversions.
Perform exploratory analysis on data gathered by the platform, design an automatic learning model, test it and then implement.
The model was able to accurately predict parts and labour costs, with estimates within a few minutes of actual time required.
Providing accurate quotes for car repairs is extremely challenging due to the number of variables involved such as car part prices, location, age, condition and additional work that is commonly linked to different types of repair.
ClickMechanic’s quote engine is the key driver for its on-going business success. Improving its capability so it can deliver more detailed, accurate quotes is essential for mechanics and car owners alike.
Although the business had captured a considerable amount of information from previous jobs, limited data inventory storage and other challenges meant that ClickMechanic was unable to fulfil a large number of customer-generated quotes. This was a cause of frustration to users and represented lost conversion opportunities for the business.
ClickMechanic chose codescrum to collaborate with them, asking us to unlock the data they held and enhance the capabilities of their quote engine without risking business continuity.
Codescrum received a preliminary dataset of several months of historical requests from the quote engine that contained a few hundred thousand records.
We used this to begin exploratory analysis to identify the best approach to the challenge. This led us to deciding that a machine learning model was an appropriate solution and we set about implementing a preliminary baseline model.
Starting small enabled us to test assumptions in a short space of time. Results were promising and demonstrated that we could improve the accuracy of predicting car part prices and feed them back into the existing quote engine relatively simply.
We then received a more complete dataset numbering millions of requests gathered over a longer period of time. We used this information to test different algorithms to find the one that generated the more accurate model. After multiple iterations, we developed a model that was able to deliver accurate estimates based on all combinations of input parameters (mainly information about the customer’s car) which we shipped to CarMechanic for testing.
Once this was done, our next challenge was to find a way to improve the accuracy of predicting time required by mechanics to complete each job. We followed a similar approach that quickly showed we could tune a model for accurately predicting labour times with estimates that were within minutes of actual time required.
Our final step was to identify integration points that would enable the business to use the machine learning model with the existing application without interrupting its current operations.
Our work has enabled ClickMechanic to increase the number of quotes it delivers to customers. It is now also able to deliver more detailed estimates for car repairs based on accurate, up-to-date data.
This has resulted in a significant rise in sales conversions along with increased visibility and confidence in cost-efficiency.
July, 2017 - Dec, 2017