Everyone agrees that there is value in data science and advanced analytics. But still, companies are struggling to see that value in their business. Francisca Zanoguera from Expedia at her presentation at the Data Innovation Summit 2019, draws attention to a McKinsey study showing that only 8% of companies have managed to implement machine learning into their processes and only 12% have managed to go beyond the experimentation phase.
With this introduction, Francisca asks the question “Despite the obvious potential in data science and analytics, why are companies struggling to realise it?” To find answers to her question, she decided to look into her experience as Head of Analytics at Expedia Partner Solutions – the B2B division of Expedia Group, a leading online travel platform. By talking to peers across different industries and looking back at her own established background in managing data science teams, she has come up with a formula of three key elements that companies should get right so they can get the highest ROI out of data science and see an impact on company results.
1. Defining the problem: What are you working on
When defining a problem, companies should make sure they are solving the right problem in a systematic manner. To define whether a problem is worth solving, Francisca proposes a chart that pinpoints a problem on the basis of actionability and size.
Ideally, problems that are big and actionable, meaning they can trigger a change, should be the focus of companies’ efforts. These problems are found in the top right quadrant and present an opportunity for your data science team.
If your data science team is working on a problem that is small and should back-up a decision made in the past – you are wasting your time. This problem is in the bottom left quadrant and is irrelevant and not worthy of your attention.
Despite the obvious potential in data science and analytics, why are companies struggling to realise it?
Spotting a distraction (bottom right corner) is tricky, as Francisca points out, because you may be led to think you are on the right path, spending time on small problems and getting some results. But in reality, you are not working on the problems that matter and actually have an opportunity cost.
Diagonally across the distraction is what Francisca describes as interesting because they are non-actionable, but are big problems for the company. Problems in this quadrant can result in learning that can be applied in the future, but again are taking away efforts that can be better spent on a real opportunity.
What big and actionable problems look like – In Expedia’s case, Francisca emphasises two big and actionable problems their data science team is focused on solving. Sorting hotels on the first result page and finding the best hero photo are two opportunities for Expedia because both are essential for conversion and click-through rate.
If your company is from an unrelated industry and you are having challenges finding opportunities that data science can solve, Francisca provides a few pointers:
- Understand your financial statements and find the biggest drivers of performance, which may prove to be opportunities.
- Map the value chain to understand how and where your company creates value. You can either improve key elements of the existing value chain or innovate new ways to create value.
- Tap into the collective intelligence and ask your employees to contribute with innovative ideas.
2. Data infrastructure: Do you have what you need
“Data is to a data scientist is what ingredients are to a cook,” shortly and correctly states Francisca. Without good data infrastructure, you cannot do anything. And for companies that don’t have a data infrastructure in place, they are in for some expensive and lengthy investment. Additionally, it creates a problem for data scientists because they can’t deliver results while they are busy building basic infrastructure.
A solution for companies that don’t have infrastructure is the MVP (minimum viable product) approach Francisca suggests. Choose which problem you want to work on and determine the minimal amount of data infrastructure you need to deliver it. Once you start creating value, you’ll get more investment.
Things to consider in setting up your data infrastructure:
- Acquisition strategy: How are you going to get data?
- Data quality and governance: Make sure you are not feeding bad data into your infrastructure, which may lead to suboptimal business decisions. And know your data retention policies, definitions, catalogues, GDPR and other legal requirements.
- Access tools – Secure systems that allow access to data in a fast, efficient and scalable manner.
Once you have the infrastructure in place, the next step Francisca points out is developing an operating model to take your data science project from the exploration stage all the way to production and maintenance. Like many companies, Expedia Partner Solutions data science team was initially getting stuck at the PoC stage, as there wasn’t a proper operating model in place to help them transfer successful proofs of concept to product and engineering teams. To solve this bottleneck, they created a virtual team consisting of product managers, engineers and data scientists with the shared goal of taking machine learning products through the entire pipeline, and building any missing infrastructure in the process. Their operating model proving successful, Expedia Partner Solutions is now focusing on fine-tuning the last stage.
3. The people: Who is involved?
People-related issues can be the most complex to tackle, but Francisca emphasises four main elements to keep in mind:
- Talent – Getting the right people to work on the right problems. There is a large gap in skills and knowledge in the data science market. In order to recruit efficiently while retaining quality, you need to have a clear strategy.
- Top-down support – To have a big impact you need top-level support, in particular if this is a new area for your company. Some companies delegate AI transformation efforts too low into the organization, and the last thing your data science efforts need is to end up with someone who has no resources, no power and no knowledge to execute them.
- Engagement – When you are driving change, you are going to come across resistance. So it’s that much more important to engage across business units as you move forward.
- Ownership – In most companies, data science and analytics are considered as support functions. But Francisca objects to this organisational setup and advises to start considering data science as a revenue-driving function. This means setting revenue targets and giving ownership and accountability to the team.
Data is to a data scientist is what ingredients are to a cook.
To help companies self-assess on where they stand with getting the full value and potential out of their data science and analytics, Francisca provides a cheat sheet with key questions for each of the above-explained areas:
- Do we have alignment on a prioritisation framework?
- Who has the final word in priorities?
- How will we measure results? When you measure results, you can communicate the value and get more investment and by-in.
- How good is the infrastructure? If it’s not good enough, what should we do?
- Do we have an end-to-end operating model?
- Are there any bottlenecks that should be removed?
- How is top-management supporting the effort? If they are not, how can you get that support?
- Do we need a culture change? If so, who is going to drive it? Francisca emphasises that data science and analytics are usually not the best placed or skilled at driving company-wide culture change.
- Do we have the right team in place?
Considering all the above, do you think your data science and analytics team is realising their full potential? And did this piece help you identify opportunities for improvement? Let us know in the comments below.