How Allegro extract insights from data

Research data will only keep increasing. Learn how Allegro intentionally tackles this growing data to extract insights and ensure that teams are able to capitalize on them.

How Allegro extract insights from data

Allegro is one of the largest trading companies in Poland and Europe based on modern technologies. Since 1999, Allegro has been making online shopping easier by offering the best prices, convenience and full security of transactions. So far, 21 million accounts have been created on the site.

The problem with data

As companies grow and gather even more data our job is to know how to manage it and get the knowledge out of it. We have big data; experiments, statistics, algorithms, and machine learning. And on the other, small data: user research with interviews, surveys, and other. But we need to know how to manage all the data properly.

The main thing that helped me organize my team’s research was a framework with three steps; setup, plan and manage. And each step comes with a challenge.

Step1: Setup

Step one revolves around identifying the right questions to ask, the knowledge you need and the methods to use.

The right questions depend on where you are in the process of strategizing, exploring, designing, developing or delivering.

Let's take, for example, exploration. Not only you can use small data, but you can also dig deeper with your analytical team to find out some behavioral patterns. So, at each step, we need to mix both small and big data.

A diagram showing the product development process with Strategize, Explore, Design, Develop and Deliver steps.

Next try to figure out what knowledge you need. Sometimes, you need to know how deep you want to go with your research. On the surface, you can get a quick answer to what people think or say. Simple qualitative methods are enough. But remember, that's an opinion. If you want to explore how people behave and why, this is a starting point to start using different methods.

Knowledge Types: Explicit, Observable with some Methods that are applicable for each.

Finally, it's time to pick the suitable methods. By plotting different methods on a matrix with dimensions of qualitative to quantitative (why to what) on the y-axis and observational to declarative (small to big data) on the x-axis you can choose the best method that answers your research questions.

A matrix with different research methods plotted against qualitative to quantiative and small to big data continuums.

Step 2: Plan

Now is the point where you need to choose the right approach. You have to decide between explanatory research, exploratory research, and dynamic research.

Explanatory research is where the data says one thing and users say another. A few weeks ago, the product team at Allegro who are responsible for parameters decided to remove one of the filters we had. It was a filter that helped select offers listed at a given time. The analytical data showed that a tiny percentage of customers select this filter when searching for the right offer.

We removed this filter for customers and we got a ton of customer feedback. They wrote to Allegro that we didn’t respect the opinions of customers, give us back this filter. This of course was not true because we do a ton of research.

When you want to verify user insights, this is exploratory research. After speaking with customers it turned out that this was useful for two main categories, automotive and collections. So we reintroduced this feature in those categories now.

Dynamic research is a mix of big and small data. For example, sometimes product listings are missing and customers need to contact merchants often via chat. We use machine learning to help us understand the reason for the contact, categorize those reasons and thanks to that, we have a quick feedback from customers.

Step 3: Manage

The final step is to manage all the mess. The job to be done here is to make sure that you collect user insights from different sources. Transparency is also essential. You need to know who is on board, what kind of data they have and simply how to get them. Don't forget to use data from outside of your company. Maybe some reports already exist and you can save yourself time.

There should be a person who is an insight leader, the owner of the research process, making sure all perspectives are taken into account and often confronted. In other words, a research lead to connect the dots. And the role of a research manager is to ensure collaboration and connection between all the teams.

Collate user insights from different sources. An image with different sources of data being surrounded by a dashed circle.

You always should have one place where you gather all the details about your research project. You can use Data Studio or Google Docs. The thing is to keep everything in one place, no matter what source it comes from.


I think we all know how to prioritize insights on the experience level using, for example, a severity scale. Now we should go a step further and be able to assess the overall impact on the business as well as show the cost risk. Of course, we don't do it alone, but with the cooperation of business and technology. But the point is prioritization is more complex, especially when you prioritize strategic insights. I know it already looks like the evaluation of solutions but that is because it actually is.

Visualization is secondary here. You may as well show it in the form of a matrix or a table.

A table showing the experience, business and tech impact of ideas to give them a total score.


There is a lot of talk about research democratization and the problem often lies in the ineffective distribution of knowledge. As a cognitive psychologist, I spread the idea of cognitive economy. The expression of it in the user research area is zero waste research. Before you start doing further research, make sure that someone else is not already working on a similar topic and vice versa. If you are doing research, make sure that it reaches all interested stakeholders.

We have a knowledge base that connects all the sources in the company. It is a simple website with tags and a very good search function. Stakeholders can search for a report or some research that may be helpful.

We also have a research newsletter where the right people get the essential insights every month, And finally, we obtain insights from research that do not relate directly to the study area. Then we do the so-called trash book for stakeholders data.


By picking the right methodology for our research based on our development stage and the questions we need answered helps us to get the data we need more quickly and accurately.

And by bringing all the data together into a centralised knowledge base lets us to glean insights which can then be prioritized and distributed to the teams that need to deliver.

I hope the framework will be helpful to you as it is to me. Now let's go out and create opportunities for the next level mixed method approach.