The buzz on the street

x Sponsored content on Research Live and in Impact magazine is editorially independent.
Find out more about advertising and sponsorship.

Tugce Bulut started her career as a strategy consultant but, when working with clients, she repeatedly found data quality a problem. So, to meet her need for qualitative insights at scale, she ditched her job and set up artificial intelligence and tech-based market research company Streetbees in 2015.

Research Live: How did you build the technology behind Streetbees?
Tugce Bulut: The hardest bit was formulating exactly what our clients needed. When I first started, we didn’t build in technology for about nine months. We did it all manually to see if it worked – we used WhatsApp to distribute and Google to analyse the data. We didn’t even have a dashboard.

Initially, every algorithm was being run by a data scientist, whereas today we’ve automated that. We now have 1.2bn data points in the database that we analyse. We built the team of machine-learning researchers, who invent new algorithms, a few years after we started the business.

How has this developed since?
We are training our own deep-neural networks. We are teaching a machine to think like a human; it doesn’t have to see previous examples – from the training it can understand open text and categorise things. We build our own algorithms and our own ontologies – a lot is based on university psychology research. Now the machine can recognise an emotion as loneliness or fatigue even though you don’t use those words. This is our Always On global platform giving clients continuous access to real people – our bees.

How do you ensure you have the right cross-section of ‘bees’?
Obviously, a lot of young people jumped on board. Over time, we had to go offline, on the ground, to balance. We had ambassadors, of all ages, and they distribute our cards. We did a lot of offline in territories where online penetration is low, such as the Philippines and Africa.

Do they have to give you a certain amount of information about themselves when they sign up?
We get it over time. Every piece of information a bee shares with us stays in a single database. So, if we were researching food delivery and you told us about what kind of food you order, how often, and so on, and then we do another research project, let’s say for a food manufacturer, we already know your delivery-food profile. We can use that information to shape the targeting or to inform our clustering analysis.

So, if you’re doing work with a client, you can hold that data for another client?
Streetbees owns the data. We never use a dataset connected to a client for another client as a dataset, but we flatten all the data – if you took a photo of your food delivery, that image is a data point in our database. So, if another piece of research also refers to that one data point, it uses that, but no-one would use an entire dataset.

How do you ensure the information is easy for clients to access?
They might be getting a monthly report where we cover their key questions and the changes around that. With the long-term clients who have multi-year subscriptions, we deploy resources – some of our employees are based in their offices full-time, working with their teams.

Initially, we thought we’d use existing dashboards, but those were built for quant-numbered research and were losing all the richness. We built our own systems where you have a quantified graph, that was extracted from open text data, and if you press on one of the bars, we show you the images that data came from. So you never lose the context of your quant and statistical analysis.

What is your bee retention rate?
We have about 65% annual. In most panels it’s less than 1% active each month; for us it’s 50%.

What are the bees’ incentives?
There are three ways it works. We pay cash – if you took a photo of your coffee for us, we might pay you $1. The second is we donate – bees choose a charity from our list and we donate if they contribute to the research. We also sometimes give bigger rewards. We run the global research for college rankings, collecting all the data for the programme with hundreds of thousands of people responding. In that case, we have some big rewards.

Is there a risk that cash introduces bias?
It’s always been a risk in research. That’s why the incentives systems change, depending on the type of research. If we are doing a political poll, for example, we won’t pay. We also have very strict data-quality control. If someone goes above a certain level of logs in a day, they get flagged.

Does this eliminate bots?
When someone registers, there is a triangle of checks we do: device ID, phone number – we send you a text and you have to confirm your number and location. We also run checks on all our researchers’ images against Amazon and Google libraries. If it matches, it’s an automatic rejection, so you can’t cheat. We eliminate a lot of users.

What are your future priorities? 
We discovered from our database that 75% of human behaviour around consumption decisions in FMCG categories is explained by your immediate context and emotions. The contribution of demographics or your claimed behaviour is only 25%. We want to turn this into an ongoing, non-stop observation of why people do what they do, with consequences beyond FMCG into health.

We’ve collected data on how people feel for four years – by geolocation, by employment, by the food you’re eating. We can use that data for mental health and to predict the economic wellbeing of a country. This year, we are doubling down on connecting desires and emotions with consumption choices, and mapping that geographically.

We are also interested in the older age groups’ service needs – it is a massively underserviced market and it holds a big chunk of the world’s wealth.

What about the ethical issues in this?
When GDPR came about, we were already following those rules. It’s privacy by design. Users know that they are giving us data for companies to develop the right products for the market. They know we are compensating them for that. They still have the legal right to be forgotten – we give this right globally, not just in the EU. And we always aggregate the data, it’s not individual. Our data cannot be used for advertising or targeting purposes.

This article was first published in Issue 25 of Impact.