How We Develop AI-driven Solutions

Erik David Johnson
Jul 30, 2020 · 4 min read

At Delegate, using AI and Machine Learning in our solutions is not a goal in itself but merely a progression of different levels of our clients’ data-driven business practices. This means that our teams that go out into the field often consist of very different profiles, and the people working here are also outside-the-box professionally. Mastering one technology is not enough and understanding how people and businesses work is just as essential.

This is a growing trend among IT Consultancy Houses, and also the theme and message of this article: No matter how technically challenging an IT project is, mastering the technology itself is never enough — you need knowledge of the business domain you are entering into; you need this type of understanding in your skill set and on your team.

Unfortunately, this truth often seems to elude the IT-projects that are centered around AI-driven innovation. Here companies often make the wrong assumption — that an advanced data-driven solution is derived from data alone. The unhappy result is that the client simply hands over some data to the IT consultants, and expects to be called upon later when the results are in.

What this basically means, is that the client will let whatever patterns are readily available in their data, define what is possible and what direction data-driven innovation is going to take them, as illustrated here:

Hopefully, it is clear that this is not warranted or desirable, and in Delegate, we start the process quite differently. Instead of the client asking “what can I do with the data I have?”, we have the client ask “What do I want to achieve with my data?”. The difference is significant. Instead of a data-only approach, we step into the client’s domain and advise on different possible, ai-driven business goals:

The result is a list of business goals that can be achieved or at least supported by data-driven initiatives. So what if the data that the client has right now does not support these initiatives? Well, this is where we would recommend and help set up a data collection strategy that would ensure that within a limited period of time, the right data has been accumulated in order to support those initiatives.

So now we know that the right goal for AI-driven innovation has been set and we know that it is supported by the data collected. Is the rest of the work data-only, where the technology will carry us the rest of the way? The answer is a resounding ‘no’. Let’s look at a recent project we did for a client in the Business to business segment, in Aarhus.

Their goal was Churn Prediction — the art predicting which customers are going to leave in the future — the aim being, of course, to use this knowledge to attempt to take preventive steps to hold on to those customers. Initial prototypes showed that there were patterns in their customer data that could be used to predict customer churn using machine learning (supervised learning). Our use of the Azure data platform and Microsoft AutoML did a lot to help the work efficiently with their data:

Even the choice of AI model was automated, but still essential to the success and performance of the solution, was to discover what data was available in all parts of the business and how it could be feature engineered prior to being fed to the AI-model(s). This was achieved only by our consultants sparring intensely with the business, and thereby understanding the business domain. Feature engineering can never be significantly automated with current software and technology.

The result was that the first third of the project timeline saw the team, consisting of myself as AI specialist, a data specialist and a machine learning developer, devoting most of our time to meetings with the business, trying to understand their world, so that we would be qualified to get the most out of the data they presented us with and challenge their assumptions on what other data might be utilized.

The AI-case itself is set to be made public soon, but for now let us dwell on what this means for IT-consultants and specialists in Delegate, and for other It Consultancies as well: If you want to be successful with implementing advanced, AI- and data-driven projects, you need to be able to acquire the business- and domain knowledge that qualifies you to get the most out of their data.

By Erik David Johnson
AI Specialist, Delegate

Destination AARhus-TechBlog

Engineers sharing knowledge and insights into current IT challenges, development and features.

Destination AARhus-TechBlog

A team of IT and digital experts from Greater Aarhus, Denmark — sharing ideas, knowledge and insights. Are you a passionate engineer? Check out the many career opportunities of our business network here: https://destinationaarhus.com/vacant-positions/

Erik David Johnson

Written by

Erik currently works as an AI Specialist at Delegate, is behind award winning AI-driven projects, and has worked with AI as both a researcher and practitioner

Destination AARhus-TechBlog

A team of IT and digital experts from Greater Aarhus, Denmark — sharing ideas, knowledge and insights. Are you a passionate engineer? Check out the many career opportunities of our business network here: https://destinationaarhus.com/vacant-positions/