Startup Secrets: Adopting agile data strategies for success in retail
Stefan Kühn VP Data & AI air up:
With soaring costs, heightened competition and ever-increasing customer expectations combining to make retail an increasingly competitive market to operate in, how well a retailer utilises its data can make all the difference in how successfully it navigates these current challenges.
However, many retailers have been operating for many years, meaning that their data systems, architecture and mindset may not be optimised to achieve data-driven business growth. As modern retailers and e-commerce platforms rise to prominence in the digital age, there are some valuable lessons that traditional retailers can learn from these younger companies regarding their data strategies.
A mathematician by background, Stefan Kühn is an experienced data leader who has led data functions in both young, digital native startups like air up, as well as more traditional retailers. In this article, Stefan shares insights into the key pillars of a startup’s data strategy and how traditional retailers can apply these lessons to drive innovation and agility in their own data practices.
What are some of the key elements that characterise the data strategies employed by startups?
When you found a startup, you’re never quite sure if the company will still be there next week, let alone next quarter or next year. So the initial idea is usually not to over-invest in data, but to try and get the most value out of the data you have while accepting that there will be limitations
For example, when air up was founded, there was a need for early insights without too much engineering investment. The data integrations were UI and click-based so that you wouldn’t need a huge team of data engineers and could instead make do with a few analysts who can answer some very specific questions.
When I came in, the idea was to build a more full-fledged data platform because the questions that the company wanted answers to had become a lot more complex. The company now wanted to understand what customers were doing across all touchpoints. Where are the customers? What do they need?
Why do they come back? Why do they not come back? How do their usage patterns change over time? These are much more complex questions which require us to have more control over the data.
However, when you start introducing more systems, you also introduce this kind of accidental complexity where these same data points live in multiple systems. And these data points may diverge slightly in different systems, so you no longer know where the source of truth is.
The key pillars of the data strategy were therefore very simple:
- Have a single source of truth for every metric or entity.
This means moving from an Extract, Transform and Load (ETL) approach to an Extract, Load and Transform (ELT) approach. ETL means extracting data from your systems, pre-transforming it, and then loading aggregates. However, you can’t easily combine it with other pre-aggregated sources unless they happen to be accidentally aggregated in a way that makes it easy to join them up. So you need to get control over the raw data and have a single source of truth. This means that when integrating raw data, you check data quality directly at the source. The sad reality is that data quality is fixed at the moment of production, and you can’t do much to fix it afterwards. So long story short, you need to integrate raw data into a single source of truth and do data quality testing starting at the very first stage of the processing chain. This will give you control over the data and provide you with the foundation you need to do more operational reporting for the different business divisions.
- Implement self-service analytics.
Smaller companies usually don’t want to overinvest in data. While the focus is on data integration and modelling to provide a single source of truth, this is still a self-service platform where the business stakeholders can go to answer their questions themselves. This is really powerful because it then allows the data team – which may be small in number at this stage – to work on the foundations.
What are some of the key differences between working in a startup environment and a more traditional retailer?
In the early days of a startup, there is usually no standardisation – you probably have very specific questions, and you just want to get to the answer as quickly as possible. But without standardisation, you can’t automate or scale. When a startup is just getting off the ground, the need to scale is not guaranteed, so you don’t tend to build for this. But at a certain point, there will be a need to invest in standardisation and automation so you can really scale to the needs of the growing business. For example, if a startup grows to be a very large company, we need to have things in place like highly standardised A/B testing systems that are self-built with a lot of functionality. So it’s important to keep things like this in mind because you don’t want to make decisions now that will become very costly later.
Another thing is that working for a startup, you have the opportunity to shape a lot of the data strategy because there is not a lot of support already in place. This means that you still need to have the ability to do a lot of of technical things for yourself because you can’t always hire more people, even if you feel like you really need someone. It’s a much more dynamic environment which can make headcount planning a challenge.
However, what I really like about working in a startup environment is the proactiveness. In a larger, corporate environment you might be able to make a plan for the next two years and slowly build towards a goal while operating standard processes every day. In a startup, there is always urgency and you’re always out of your comfort zone. Personally, that’s something I appreciate because you get this immediate feedback of how your actions really changed something. In larger companies of course you can still do impactful things, but often there is a much longer wait to receive relevant feedback, and your scope of influence can be narrower.
How do the challenges that data leaders face in retail differ from those in other industries?
First of all, there is a lot of history and tradition in retail. Older companies might be using slightly outdated technologies. For example, a POS system can be so old it can be impossible to pull data out of it in a scalable way. The people that built these systems are usually no longer with the company, so sometimes it can be like archaeology to try and find out how to extract data from these systems.
On the other hand, sometimes retailers have big ERP ecosystems in place, but may have contractual regulations that prevent them from exporting the data without specific contracts or licenses. So there are a variety of issues, and often the main root cause is that these established companies did not start as digital first companies. Everything data-related was simply considered as something needed in order for the company to operate. For example, you need a POS to sell your product and count stock and revenue, but nobody back then really thought about actual data use cases in the sense of integrating information from all touchpoints.
Another challenge is that you often have very thin margins in retail, so costs are a huge factor. Sadly, the data department is often considered a cost centre if you are not contributing to revenue, and the only way to create revenue is to have access to the data you need and create meaningful data products out of that. You also need to have a smart strategy when it comes to building because you have to accept the reality of small margins and not much resource. So tackling one use case first with a very lean, clean approach can be a good place to start. You also need to have the support of people like the CEO in order to achieve up-front investment because often when things get harder and revenue is lower than expected, perceived cost centres like data departments can undergo cuts.
What would be your top piece of advice for CDOs of larger retail companies looking to adopt a more agile approach to their data strategy?
My main advice would be to find a way to contribute to revenue, ideally through some form of customer centricity, and then build for the outcome that you want to achieve. Avoid starting from a place of wanting to build a huge data platform, warehouse, or standard reporting system. Quite often, standardised reporting is only showing you the same numbers that you already see in other systems, so if you really want to highlight that data has a value, you should aim for something that you can’t get out of any other system. Then you can start to do conscious experiments with the data. I think one of the best uses of data is to derive hypotheses and inform experiments because that’s something you can’t do with any other system. Then it depends on the specific business whether it’s focused on something like personalised recommendations, pricing strategies or general customer insights. Once you get something like this up and running, then it becomes a self-improving system.
Personally I always want to end up in a place where I have a system that is not just reporting stuff, but is genuinely useful in driving things like product development or enhanced customer relations. So I would say to focus on the outcome you want to achieve, and then find a way to build a lean platform for that purpose.
If you want to learn more about Stefan’s insights, you’re in luck. He’ll be attending the highly anticipated CDO Retail Exchange taking place this July – download your personal event guide here.
