This year Attest was a gold sponsor of the Data Science Festival (DSF). This meant that we not only had a booth at the conference, but we also gave a talk about how Data Science (DS) works at Attest. If you are interested in listening to our talk at DSF you can access the recording from here.

Throughout this process we realised that not many people have heard about Attest. In this article we would like to give an introduction to Attest, how DS works with the rest of the company (in particular the Data Engineers, DE), and give an overview of Attest’s DS projects.

Attest Product

Attest is a London-based tech scale-up in market research and has recently raised Series B funding. Attest is B2B and works with clients of all sizes, from huge businesses such as Microsoft, Boots and EE to smaller, fast growing startups like Made by Nacho’s, which is a cat food brand.

Attest has its own product, which is a survey platform. Our clients can log in to the platform, design their survey, choose the targeted audience (i.e. demographic), press run and we do the rest of the job. The Attest platform can access more than 125 million people across 58 countries to provide a fast, easy and scalable means of getting answers for a survey. Once results are available we display them to our clients.

A display of Attest’s survey platform.

Data Scientists at Attest

Attest has a cross-functional squad model, which means that Data Scientists are embedded in squads and work alongside product managers, front-end and back-end software engineers and designers to solve specific business problems. In addition, Data Scientists spend some time outside of squads working on projects which are valuable to the business. We also collaborate closely with Data Engineers to deploy and monitor models in production.

Data Scientists at Attest work in squads and collaborate with design, product and engineering people to answer a business problem.

Currently our team consists of 5 DS and 2 DE and we are looking to grow. Data Science projects can involve experimentation design, supervised and unsupervised machine learning (ML) and natural language processing (NLP).

Below you can see some examples of projects we work on:

Examples of inside and outside squad DS projects.

Data Scientists at Attest have a lot of autonomy and being proactive is an important part of the role. We also have amazing rituals and benefits, some of which are:

  1. Two days per month internal hackathon
  2. 20% time per month working on a project of interest
  3. 10% time per month contributing to / participating in charitable or community focused work that enhances or improves the lives of those around us
  4. Budget for conferences and training
  5. Budget for personal or professional growth and development

Data Science Business Impact at Attest

To give you a better feel for the DS impact on the Attest product, we’ve provided two examples below:

1- Detection of bad quality data:

In the past, market research used to run surveys in-person or by making phone calls. However nowadays running online surveys is much faster and cheaper, and therefore platforms like Attest came in to existence. One of the main challenges of online surveys is that respondents (a person answering a survey) might not be remain attentive throughout all of the questions. In addition, the monetary incentives that we offer to our survey-takers can easily attract bot-farms or click-farms instead of genuine respondents. These types of undesirable (bad quality) behaviours can bias the results of the survey and therefore we need to detect and remove them. Below you can see two examples of a bad quality response:

Examples of undesirable (bad quality) behaviours in respondents. In Example 1 on the left, one respondent has chosen contradicting answers. In Example 2 on the right, there are examples of several respondents who have provided gibberish text and some who have provided answers not related to the open text question.

Data Scientists at Attest have developed several quality algorithms (e.g. statistical and ML models) that analyse the behaviour of respondents and detect and remove undesirable ones. As you can see in the image below, once respondents fill a survey their answers go through all of our quality algorithms and each algorithm outputs a label of whether this respondent is good or bad. These labels are then given to a final weighted formula and a final decision on the respondent is made. If the respondent is good, we keep them, otherwise we reject the respondent. More information on the details of these quality algorithms will be available in upcoming medium articles.

Attest’s pipeline on detecting good or bad quality respondents. Attest has several quality algorithms, some of which are NLP or ML, and others are changes made in the surveys, whose value has been confirmed through A/B testing.

If a survey needs, for example, 500 respondents, and the quality algorithms remove n respondents, the Attest platform keeps recruiting more respondents until the 500 goal is met. Therefore, our algorithms should be fast enough to reject respondents before they complete the survey in real-time (i.e. < 300 ms). This is where DE helps us out with all of their cool knowledge and technologies. We will have future medium posts on how DE deploys and maintain our models in production.

2- Displaying useful results to Attest’s clients:

Clients, who are the users of the Attest platform, would like to receive useful insights from the good quality respondents of their surveys. This is where some of our NLP algorithms come in to play, such as generating groups/themes of open text answers and showing sentiment. Below you can see our new proposal for how our clients may visualise the themes of answers in the responses and the type of sentiment they convey.

Proposal for visualising the output of two of our NLP models: grouping of open text answers and sentiment analysis.

Summary

In this article we provided an overview of Attest and its product (the survey platform), and how Data Science have advanced the survey platform to make more data-driven decisions and analysis. We have a wealth of data at Attest, nearly 100K surveys, ~30M respondents, and hundreds of interaction data for each of those respondents. We have also recently started capturing web interaction data from respondents using the platform. There are many projects and areas at Attest where data science can have further impact, such as analysis on the sampling of respondents (through aggregators) or predicting churn for the finance department.

If you are interested in joining our team or have any question feel free to contact us!

References