Blog

Uncategorized

Increase Car Dealership revenue using AI

Increase Car Dealership revenue using AI

Gamasome had an opportunity to work with MyAutoIQ and I took up a roll as AI Engine Lead. I must say that one of the best learning and challenging experience in many ways!

A quick intro about MyAutoIQ and its USP:

MyAutoIQ is set to changing the way auto dealers engage and acquire their customers using information, predictions, and recommendations through machine learning algorithms to perform the following:

  • Provide dealers with smart, predictive, and efficient decision-making capabilities
  • Drive more sales by personalising customer acquisition and existing customer engagement
  • Generate leads from existing service customers and convert them into repeat buyers

MyAutoIQ brings proven technology solutions for increasing customer loyalty and lifetime revenue to the auto dealer space. It’s done by finding the best way to fit the needs and expectations of the dealer’s customer through an infusion of AI-driven predictive capabilities, real-time decisions, recommendations, personalized marketing and intelligent end-to-end solution that benefits both the customer and the dealership. 

Gamasome have been working with them building ETA and EDA tools and Optimised models for variety of business problem statements like the following:

Define and/ or fine tune offers based on customer purchase information including frequency, preferences, service items or offers purchased etc. and combine with other elements like timing, vehicle usage or OEM recommendations etc.

  • Deliver offers analysing maintenance needs, vehicle usage pattern, spend appetite, customer demographic profile, offer popularity, time of the year & other variables
  • Learn from customer spend appetite and service interests of similar customer and vehicles and fine tune personalized offers based on that.

Personalized maintenance recommendations: Deliver personalized maintenance recommendation based on engine recommendations.

  • Determines customer vehicle usage pattern (e.g. customer brakes more frequently than norm and hence higher brake wear is expected) and forecasts maintenance needs based on that while also evaluating OEM recommended service schedules and similar vehicle history (continuously updated based on expected usage)

Identify elements which will help create 360-degree view of the customer. E.g. if a customer is “good maintainer” based on customer’s adherence to OEM recommended maintenance schedules or if a customer is “loyal”, if the data indicates that he/ she gets the work done at the same dealership. Engine driven scores with following information

  • Lifetime value $: DS driven lifetime value modelling
  • Customer 360 view: Donut indicator giving a score on a scale of hundred for:
    • Loyalty
    • Value shopper
    • Spend
    • Car Value
    • Any other meaningful indicators driven by data

Identifying customer spend appetite based on their income, net worth, neighbourhood etc

Identify popular models based on zip code, age group, family types etc and use it for suggesting dealership vehicles for that customer.

Find pattern of purchase, service and other vehicle preferences amongst the different communities – Asian, Hispanic etc. Analyse predictive vehicle buying or servicing patterns based on demographics

Fine tune marketing based on members in the household. E.g. a household with members reaching drivable age may be looking for new vehicles or households with small children may be looking for bigger car or vans

Identify potential trade-in cars and predict a profitable buy-in price for the dealership by evaluating vehicle condition data, historical data, mileage, number of ownership vehicle has had, servicing pattern, market value of similar cars etc.

  • Algorithm driven vehicle value estimate, analysing market data sources (NADA, KBB etc.), past sale data, vehicle condition etc. to come up with estimated price of the car and associated bell curve
  • Customer 360 view: Donut indicator giving a score on a scale of hundred for:
    • Car demand: Evaluation of sales to determine relative ranking and popularity of the car model and the relative variance of sales price.
    • Expected inventory days
    • Car value: Predictive model output
    • Cost comparison: Statistical model refined average cost and sale figures of similar vehicles

Evaluate similar vehicle service history and spend pattern to predict service options and offers for a customer. As an example, if a customer has 2014 Honda civic with 30,000 miles, engine can analyze similar cars in the dataset to predict what service needs would the vehicle have and what offers are similar.

Analyse and predict revenue, service items and offers to fill the shop to capacity

  • Forecasted Service revenue and number of customers – Predictive model applied on past data linking external factors like sales, offer popularity etc.
  • Forecasted top service demand areas – Predictive model applied on past data linking external factors like sales, time of year, service item popularity, offers etc.
  • Service offer recommendations – Predictive model, considering offer popularity, time of the year & other variables
  • Correlating weather data with service sales and predicting what service items would sell based on weather condition. E.g. a strong snow season may predict need for tire alignment as the roads become bad with heavy snow

Execution & Approach:

ETA:

All the Data Extraction tools are completely built from scratch inhouse. The data necessary for working on the above problems are:

  1. Customer Demographic Data
  2. Vehicle Details
  3. Customer transactional Data

MyAutoIQ have established external data agreements with various data providers for lifestyle, behavioural, manufacturer, recall, valuations, in-market etc. All the data providers provide data in many unique forms and formats such as APIs, XML, json, SQL, Web Scrapping and RSS feeds.

Our EDA tool is built from scratch using Python and dependent Network Libraries. The tool also performs data preprocessing tasks such as consolidating all the data sources and merging them together using unique identifies like VIN number.

Machine Learning:

Post the rudimentary data consolidation and preprocessing, the data has to be further processes uniquely for training individual models addressing an individual business problem.

We majorly used ML frameworks like Scikit, Kiras and Tensorflow and was hosted on AWS. We built models for Customer Segmentation, Upsell Engine, Maintenance Perdition and Personalized Offers based on the Purchase and Behavioral patterns of the user.

Results:

Using AI in car dealership have shown such a huge difference in terms of finding the right customer and finding the right offer and deal parameters for customers. The accuracy of the perdition and segmentation are as high as 85% and have helped Dealerships to successfully increase revenue.

Bitnami