How agile AI experiments can help to identify who knows what within big organisations

Photo by İrfan Simsar

For a fast-growing global consultancy company like Futurice, with more than 500 people across 7 offices, knowledge sharing is both crucial and challenging.

Do we have in-house experts on deep learning? Who has knowledge of the Delphi method? Who can help with a migration of Azure to AWS? These are just some of the questions that crop up every day. And although there is always someone somewhere who will know the answer, it can take time to find out — and in fast-growing companies, time is at a premium.

Then take the issue of duplicating work. Futurice is a highly autonomous and distributed organisation. And it’s fine if everybody does their own thing, but it sure is useful to know if someone has already done a similar project or task to the one you are planning. Or if that person who is a wizard in a new emerging tech you want to try is located in Oslo. Or just if there is someone you should talk to before doing your thing.

Which is why Futurice set up an experiment to see if AI could be used easily and quickly to bring transparency to these numerous dynamic data sources and as such, de-mystify the problem of who knows what/who has already done what within a big company.

In comes Aito, Futurice’s first product spin-off company. Aito is a predictive database with machine learning abilities. The fact that Aito’s database features are deeply integrated with AI, enables fast AI implementation and experimentation without the need for data science skills.

In the case of Futurice’s knowledge management challenge — how to find an expert in a company over 500 — Aito enabled quick testing of whether data could provide insightful results to help resolve it.

The result is a tool which Futurice calls Bubbleburster 3000. The reason for this name is that the tool aims to burst the bubbles or silos which people in organisations tend to work in — more specifically to allow knowledge and expertise being shared more widely and more quickly throughout the company.

The whole project only took a couple of weeks and was carried out by developers and UI designers — no data scientists were needed.

So how did we go about setting up this project for Futurice?

Connecting the dots

The first step was to connect the 500 Futurice people with their professional skill sets, in a sensible way.

To avoid any privacy issues, only the public internal data sources were used:

  • Flowdock: the internal company chat platform. Information varies from “what’s the recipe for Finnish peasoup” to “I’m facing some weirdness with a non-CLS-compliant API”
  • Google public calendars: this included around historical 500.000 non-private events that Futurice organised or that employees have been invited to
  • Planmill: which houses publicly available project info about who is working on which project/client
  • Power-tool: a project allocating tool which matches employee-skills with their wishes to work on certain projects or with certain technologies

Data from each source were placed in separate tables, each with its own schema. In this application, the schemas were very straightforward. For instance, a flowdock message contains English text and a link to the author in the users’ table. The other tables essentially had the same schema. All the data was exported from their original data source.

To make the results more significant, the Flowdock data needed some data preparation (mostly “cleaning the data”). For example, there was no need to know who’s allergic to tomatoes or who’s a yogi — at least not for this project.

We also pre-matched different user IDs from all 4 data sources before the import to create one consistent ID for each of our 500 employees.

All the data was imported directly into Aito, using the File Upload API.

After the upload, we used one of Aito’s queries to match the relevant people to the search word.

The query syntax resembles SQL, and it reflects Aitos’ internal query pipeline

  "from": "flowdock",
  "where": {
    "message": "react"
  "get": "user_id",
  "orderBy": {"$p": ["user_id"]}

And that was it! Welcome to the Bubbleburster 3000

Having pooled all four data sources, a simple search gives an overview of the Futurice people who know about everything from data transmitters, React or machine learning (to name a few).

Well… not quite.

The Aito-powered Bubbleburster delivered the results along with some slightly superfluous detail such as the statistical relevance of how a certain person relates to a search word.

So once everything was up and running technically, Futurice’s UI designers created an interface to make it user-friendly.

Bubblebuster UI
Version 1 of the user interface. Currently Futurice is developing the UI further.

Another feature that Aito can help Bubbleburster to provide in the future, is not only showing which people in Futurice are linked to which areas of knowledge but also why they are linked. Is it because they are struggling with the same technology, or because they have been solving a lot of problems around it? Or maybe they recently attended or organized a training around the subject.

Futurice is continuing to develop the Bubbleburster application as a tool for knowledge-intensive organizations. This includes using Aito to understand and build relevant data sources that will make the search function richer.

Do you want to hear more about how to set up a similar project in your company? Or you have a use case?

Contact us

Or visit our website at

Back to blog list

New integration! Aito Instant Predictions app is now available from Airtable Marketplace.