Applying machine learning to business outcomes at Travelopia

Travelopia shifted its focus from a technology approach to business outcomes, adapting agility and lean to deliver machine learning solutions. This enabled them to deploy machine learning business models faster and better.

Sreekandh Balakrishnan, Director – Innovation at Travelopia, and Simon Case, Head of Data at Equal Experts, spoke about machine learning at Travelopedia at Lean Agile Scotland 2022.

The first iteration of machine learning was very technology-centric, Balakrishnan said:

We took a technology stack approach and built a data lake before understanding the business use cases. We took a big bang approach to deployment, one large team with a promise to deploy multiple use cases once the data lake is in place. At the end of 18 months we had 3 models in production but no less business units using it.

Proposed case to understand which business improvement you want first:

Start small, pick a piece of value, and learn not only about the technology, but how to isolate the problem, how to onboard users, how to deliver the results to users or downstream services and what other organizational challenges (people, process first & finally technology) are there.

Balakrishnan mentioned that they weren’t getting the technical or business impact they wanted. They made some changes to be leaner and focused on business outcomes:

We had a 2nd iteration with the new Lean Agile approach and within 3 months were able to deliver 2 models that are being used and creating business value. After this success, we adopted this as our method. Since then we have expanded production to 10 models for 5 brands. All are used in-store and some brands generate almost 21% additional business wi. In fact, I can now get a new model on the market in less than 10 weeks.

InfoQ interviewed Sreekandh Balakrishnan and Simon Case about machine learning at Travelopedia.

InfoQ: What did you learn from the initial application of machine learning and how did that impact your approach?

Sreekandh Balakrishnan: We applied lean, agile principles – find what’s valuable, deliver in small increments, and keep learning (and pivoting until you get it right).

We have shifted the focus from a technology to business outcomes. We took the time to understand what the company wanted and how it should be delivered. The team was too big, so we reduced it from 40 to 6 people. We found that a lean, cross-functional team could move faster and focus on what the business wanted.

We’ve stopped building a data lake that satisfies all ML needs and focused on providing only the data we need. This also had the side effect of reducing our cloud costs to 10% of their original costs.

We’ve also made a conscious move away from GUI-driven tools. We had used them in the first iteration, but found it difficult to apply modern software development techniques (TDD, pair programming) when using them. Instead of speeding up our delivery, they slowed it down.

Eventually, I realized I needed the approval of the company, which was more used to big bang deliveries. So I made sure I had an executive sponsor who understood and believed in this approach. This has really helped our business relationships and eased the adoption of the ML models.

InfoQ: In your presentation you suggested not to worry about data platforms but to think about how your teams organize themselves. Can you explain why?

Simon fall: Machine learning is a team sport. The data scientist and the data engineers must work together. They are distinct disciplines – data scientists are skilled in algorithms and math, but often lack the software skills needed to build reliable products. It can be tempting to tell them apart, but if you do, they won’t learn the techniques needed to get their work into production.

Balakrishnan: Things started to change for us when we streamlined the team and created a single small cross-functional team. Because the data engineers, data scientists, and business analysts worked as one team, they better understood the users and were able to make quick decisions and trade-offs regarding the technology stack and ML model.

InfoQ: What is your advice for companies considering the use of machine learning?

Balakrishnan: Start small and don’t worry about buzzwords! Keep business stakeholders close by and invite them to daily standup/planning meetings. Create the first profit and go into production. Remember, it’s a learning curve for you, your team, and your company.

case: Build your competency iteratively – start with a steel thread – the first vertical slice to offer useful business value. Use this to learn quickly, and when you’re happy with it, serve as a template that you can use for other ML models. If you’re lucky, there will be one cobbled street – a way of working that allows you to quickly introduce new models into the company.


Leave a Reply

Your email address will not be published. Required fields are marked *