From black box to glass box: The future of predictive analytics

Our latest webinar topic was “Predictive analytics facilitating smarter decision making”. As AI gives us more opportunities to create analytics models, handle more data and create predictions, a big topic of the webinar was to find out the limitations of both AI and how we interact with it.

As a short introduction of the guest: Our speaker next to Thomas Janhonen was Ilkka Peltola, the CEO of Supervoima. Supervoima is a company that helps other companies adopt AI for business benefit. Ilkka has a background of working 20 years in data analytics with the focus of actual outcomes and business value.

The data overload

The general idea of using data to make better decisions is not new. The fact that data availability has increased every year, is neither. However, now with AI, anyone can access and analyse data without having complicated tools in place.

The assumption: More available data makes decision making easier. No. Based on statistics, 75% of executives don’t trust their data as well as 2 out of 3 CEOs rather make decisions based on their gut. Even if the trust and willingness to make a decision based on data were there, the difficulty comes also from figuring out what should actually be measured, how should the results be interpreted, what insights are worth acting upon in the long term…

This is also where the shift is currently happening: not only knowing what happened and why, but creating models that can anticipate what could happen based on the data the company already has.

Predictive analytics doesn’t necessarily have to include AI at all, the more important and again hard part is figuring out what kind of predictions would actually help make better decisions. And if the prediction is correct, how to act? To make this make more sense, let’s say a SaaS company predicts a lower percentage of customers using the software in the summer based on earlier years’ data, what action would you take? Or, would you do anything at all? To quote Ilkka from the webinar: “If a prediction doesn’t change what you do tomorrow, it isn’t predictive analytics - it’s curiosity.”

Creating transparency in analytics

It is tempting to command an LLM to make graphs on data you serve it and maybe even analyse them for you, but then the question “Can I trust it” comes to mind so fast that our speaker even had a slide about it. Apparently, that is a wrong question entirely! It is the same as asking to trust a new analyst on day one.

Better questions include: what data do you actually have, is the data quality good and based on the results you see and your own experience in the subject matter, does the analysis make sense.

After knowing what data you have and what kind of predictions you would like to make, the next steps are to understand that:

1. AI makes bad data sound credible. Remember, most of the tools are built to give you answers you want to see. So if your data is originally bad quality (missing rows, filtered wrong), you can still get good looking results.
2. You can’t outsource expertise you don’t have. If you don’t understand the answer, how it was built and what it means, you can’t trust it.

So what to do. Here we get to the main title of this article: turn the black box into glass! Make it see-through. In addition to understanding the data you’re managing, treat analytics like software: test your models for logic, validate the data you get and compare the results to reality. Create an iterative loop of your question, decomposing, data probing, refining and verifying until the answer holds up.

The human edge

As with all AI conversation, what we want to hear are the things AI can’t do. In terms of analytics it’s the fact that we can add context and understand the data. If the numbers look way too high or low, we can see something is off. AI recognises patterns, but can’t say if something is correct or not.

Human judgement is where the competitive advantage lies: Gathering data that makes sense for the company goals, making decisions based on it and starting to anticipate the future based on earlier data and learnings.

If you want to see the full webinar, it’s on our Youtube channel. Or, if you need a data professional, give us a call!