What does it take to deliver a seamless experience across thousands of customers, three automotive brands, and countless service touchpoints — without compromising quality or consistency?
At Dataiku’s Everyday AI Summit in Dubai, Dr. Mani Abedini, Head of Data, AI, and Analytics at AW Rostamani Group (AWR Group), walked through how his team is building exactly that. By applying machine learning (ML) at critical moments in the customer journey — and doing it with a steady, use-case-first mindset — they’ve turned operational data into real, everyday value. While AWR Group also operates across a wide range of sectors, the session mainly focused on its automotive business, known for its strong presence in the region’s motor industry.
Building a Customer 360 View From the Ground Up
For any AI strategy to deliver business value, it starts with data. In this case, the data journey begins the moment a customer starts exploring a vehicle. From the first enquiry, through to purchase, and into years of after-sales service, AWR Group captures the full cycle.
This comprehensive view is centralised in what the team calls its internal Customer 360. It includes:
- Demographic details
- Ownership history, past and present
- Vehicle statistics like mileage and service records
- Call centre interactions and complaint logs
- Survey results, NPS scores, insurance, and financing data
By bringing it all together, the team doesn’t just monitor performance — they generate insights that influence how the business engages with each customer, at every step.
Predicting the Next Visit With ML
One of the Group’s most mature AI use cases focuses on a simple but critical question: “When will a customer need their next vehicle service?”
During a car’s warranty period — typically three to five years — customers must bring their vehicle in for servicing every 10,000 kilometres or every six months. Missing that window could void the warranty. For a customer, that creates pressure. For the business, it’s a coordination challenge.
To help customers stay on schedule, the data team developed the Next Expected Due Date (NED) model. Its job: to predict when a customer is likely to hit their next service milestone, based on how they drive. This is where personalisation matters. Some people commute daily across the United Arab Emirates. Others drive sparingly on weekends. The model adapts to both, factoring in driving patterns, mileage, and customer behaviours over time.
The model went through three major phases:
- A basic statistical version, with ~50%–60% accuracy.
- A regression model, which raised accuracy to ~70%.
- A final implementation using Dataiku, applying AutoML, auto-tuning, and ensemble methods to reach ~80%.
Migrating the model into Dataiku allowed the team to experiment with multiple modelling strategies quickly and efficiently. It also gave them the ability to fine-tune and evolve the solution without overhauling their workflow.
Adapting to Post-Warranty Behaviour
Warranty periods come with strict manufacturer guidelines — but once they expire, behaviour shifts. Some customers stick to regular servicing. Others only show up when something goes wrong.
To manage this, the team introduced a second model to identify individual service habits. It’s designed to detect who’s likely to return, who tends to skip, and how often.
With this model in place, AWR Group can better allocate time and resources. Customers who are consistent and timely are prioritised. The model even predicts no-shows — people who book but don’t show up — with roughly 80%–85% accuracy. That kind of foresight helps the contact centre operate more efficiently, whilst improving the customer experience for those most engaged.
Different Brands, Different Patterns
Another insight the team uncovered: not all customers behave the same — and brand plays a big role. For example, some customers prioritise practicality, while others are more focused on luxury. Additionally, certain models attract a distinct type of owner altogether. Recognising these differences, the team built separate models to cater to each customer segment.
Each model is trained independently, then integrated into a shared operational framework. This brand-specific segmentation ensures that predictions are not only accurate, but relevant to the customer behind the wheel.
Understanding Feedback at Scale With LLMs
While predictive modelling handles the when, understanding the why behind customer satisfaction requires something else: language. AWR Group receives detailed feedback from customers recounting their experiences. These stories often mention service advisors by name, highlight any past issues, and include specific expectations for the future. Capturing all of that in a structured way could be difficult.
Originally, the team tried traditional NLP techniques. But they were time-consuming and inflexible, requiring manual setup for every step. Now, using large language models (LLMs), the team can process feedback more naturally and at scale.
Their new sentiment analysis is capable of:
- Classifying sentiment as positive, neutral, or negative.
- Identifying which part of the journey the comment refers to (e.g. mechanical vs. service experience).
- Flagging advisor names and booking preferences.
- Detecting follow-up promises made by staff, so they can be honoured next time.
These insights are fed directly into the Customer 360 view. So, when a contact centre agent reaches out, they already know what was discussed before, and who the customer prefers to work with. Crucially, the system is multilingual. The model understands Arabic natively, which is especially important given the Group’s customer base — many of whom speak Arabic.
Scaling Through Iteration, Not Ambition
One of the most important aspects of this journey is how it evolved. Rather than starting with a massive vision, the team focused on solving one problem at a time. They started small — building useful models, improving them, and then layering on more complexity as needed.
That’s where the platform made a difference. With Dataiku, the team could scale their efforts gradually, experimenting without risk, and deploying only when ready. Today, what began as a statistical estimate has grown into an integrated system that supports real-time decisions, personalised outreach, and long-term loyalty.
AWR Group’s data science team didn’t chase trends or hype. They focused on building what the business and their customers needed — layer by layer, use case by use case. By combining customer insight, behavioural modelling, and scalable tooling, they’ve turned AI from an abstract concept into an everyday operational asset that adds value to the customer experience. This is what applied AI looks like when done right: practical, measurable, and always connected to the customer.