Car owners get a message that fits them: a new car after 5 years, a service offer when it is due.
The manufacturer, with more than 28 million customers and over 40% of the market, wanted each owner to hear from the brand at the right moment with an offer that fits their car and life.
What stood in the way.
The team picked customers and wrote recommendations by hand, sifting names, car models and purchase years for millions of users.
The manual work could not scale and made little difference to growth.
Privacy rules meant only a limited sample of customer data could be shared for model training.
What we built.
- A web application in Python and Streamlit, scoped in a Discover and Frame workshop and run on Amazon EC2.
- Filters for birthdays, anniversaries and maintenance due dates that list the customers likely to respond; staff open a profile and choose from generated messages.
- A car recommendation engine using collaborative filtering and matrix factorization over customer details, car data, past purchases and test-drive feedback.
- A value estimate for the customer's current car, for exchange and upgrade offers.
- Rules for when to reach out (occasions, festivals, service due dates, new offers) and GPT-3.5 messages written in the brand's tone, with banned keywords filtered out.
- Synthetic data generated with machine learning to widen the limited training set, with models trained in Databricks and Jupyter.
- Python
- Streamlit
- OpenAI GPT-3.5
- Databricks
- Jupyter
- Amazon EC2
What this work shows about us.
A five-year-old car means an upgrade; a recent one means service. The rules and recommendations follow that cycle.
With only a sample of data allowed, synthetic data stood in for the rest during training.
Want to build a similar AI recommendation engine? Tell us what you want to change. Our AI assistant, Ask Unthinkable, helps you define your requirements, then connects you with the right team.
Start your plan ↗If we built it today, with AI.
What we would add now, moment by moment.
An agent would take a customer's reply to an upgrade message, book a test drive at the nearest dealer and pass the lead on with the trade-in estimate. Dealers still make every offer.
An agent would offer service slots when maintenance falls due and confirm the booking with the workshop. Workshops still decide the work needed.
Where this work fits.
- Manufacturers & distributorsYour company is meant to get dealers the right stock, on time, without making them chase it.
- Supply chain & assets (inventory, assets, procurement)The right stock in the right place on time, and everyone knows what has been ordered without chasing.
- Dangote GroupDangote's engineers, contractors and suppliers find, share and approve project documents in minutes, not days.
Facts and figures from the project records; nothing added.
Want this for your business?
Tell us what you run or what you're building. Our AI assistant, Ask Unthinkable, helps you define your requirements, then connects you with a senior AI engineer.
Send us a note instead
An AI agent. This prototype uses prepared answers and sends nothing.
We use essential cookies to run this site. With your consent we also use analytics cookies to see which pages help. Privacy policy