Analytics Translation is still around! Part 1: Ideation techniques 

Back in 2018, McKinsey coined the term analytics translator, a new type of professional able to understand data and business needs simultaneously. My colleagues in the data training department got very excited about it. There was much talk about this “must-have” role and what to do to become one of them... and then not much else happened. Some professionals adopted the title of analytics translator. Others (Data POs, analytics consultants, or CTOs) carried out the functions typical of this role. And yet, the hype faded. The public interest for this position dwindled, but the communication problems between data and business professionals remained.  

Why people stopped talking about analytics translation is a bit of a mystery to me. Maybe GenAI emerged so strongly that we forgot about these professionals, which brings us to the question: Can a good project manager or product owner equipped with a Large Language Model do as good a job as an analytics translator? No, they can’t. That’s our answer at Xebia, and I'd like to show you why.  

This is the start of a series of articles about analytics translation and how it is still very much needed in the age of GenAI. This first article is about Ideation Techniques.  

The AI Solution framework

Just to clarify: an analytics translator is a professional with solid data knowledge and business experience. Their role is close to product owner or project manager, but their data acumen allows them to evaluate and discuss the solutions with data engineers and scientists much more fluently.  

Everyone does analytics translation differently. I like the AI Solution framework, because it helps me develop data products and projects while focusing on how the analytics translator can help at every step of the way. Even if you use a different development system, it still probably has many areas in common with this one. 

I’m not going to show you the whole framework yet, because I’m a big believer in building up the information, rather than throwing it all once and then breaking it down. For now, I’ll just tell you there are three stages to develop AI solutions: Ideate, Experiment, and Industrialize. Ideation is all about finding a valuable business idea and outlining the requirements for the data team and other stakeholders to work on it. We’re going to focus on that phase, more specifically on the first half: Collect and Prioritize.  

Actually, let’s break it down to the fullest. Collect and Prioritize may sound like one step, but I always recommend doing separately. Let’s see why and how an analytics translator can help during the process.  

Collect

Here’s a thing about coming up with great ideas: having them in one go is very tricky. Why? Because the processes of coming up with new ideas and evaluating them are not only different, but somehow opposite. I don’t know this from research (which still acknowledges “our understanding of the scientific nuts and bolts behind creativity remains limited”); I know this from my many years writing fiction on my own and working on data projects with my team.  

There’s something wild and a bit crazy about generating ideas. It’s a way to make the unconscious conscious. You may have incorrect, bad, or even inappropriate ideas, and that’s okay at this stage. Why? Because among all those terrible ideas, you might have written down a good one that will save your business. This is a moment to think without limits. Are you thinking about revamping your whole data platform? Write it down. Are you thinking about hiring five machine learning engineers? Write it down. Are you thinking about having Taco Friday to boost team morale? You’d better write it down.  

By the way, coming up with wild ideas is something LLMs don’t do particularly well. LLMs think necessarily within the box, because they have been trained to reproduce and recombine previous structures and thoughts. They can connect ideas in interesting ways, but truly new out-of-the-box concepts still come more naturally to humans, when they are given the space to do so. For now, I’ll take an imaginative analytics translator over an LLM any day.  

There’s a clear advantage at this stage in having an analytics translator instead of a regular PO. Since the analytics translator has spent years working directly with data, they can gather inspiration directly from the source. They can think in terms of clustering and silhouettes, random forest predictions and ROC. Their data knowledge opens up the brainstorming space massively.    

What if I struggle to come up with ideas? 

Coming up with ideas comes naturally sometimes when we’re surrounded by our data team and actively using our company’s products. Other times, ideas don’t come to us, and we need to chase them. That’s completely normal.  

First, coming up with ideas isn’t a natural gift, but a skill that can be practiced and honed. Second, if inspiration doesn’t come to you, you can do like the writer Jack London and “go after it with a club.” The way we chase them in Analytics Translation is by looking at the Value Chain of Data Science. 

This is a summarized process of the data journey throughout the development of AI solutions. There’s a lot to say about this framework. For now, it suffices to say that if you’re actively chasing ideas, start from the value you can generate for your business in different products/departments, and then work your way backwards to find the type of insights you’d need to make that happen, and the data you’d need to collect. See an opportunity to reduce customer churn in one of your products? Good, now you can think about the insight and actions that would lead to that value and eventually figure out the data your engineering team should collect.  

Brainstorming alone or in a group?

Some teams are guilty of having all their idea-collecting sessions in a group. By all means, bouncing ideas off each other in a team can be stimulating and productive, especially when an analytics translator coordinates the input from business and data stakeholders. However, some people struggle to speak up in a group environment and therefore do not contribute as much as they could.  

The solution to this is simple and sadly underused: ask people to think about options for your solutions before coming into the meeting. This is positive in two ways: 

The idea log

So far, we have discussed the collection of ideas alone or in a group session. While these are great ways to develop our data and AI solutions, the reality is that most ideas come to us in the middle of another meeting or a family reunion. What to do in those situations? 

First, tend to present matters, such as answering a question from your boss or stopping a quarrel between your nephews. But the moment you have a bit of peace, take out your phone (or notepad, if you’re going full analogue) and write down your idea in a log. It’ll take you one minute, and then you can bring your attention back to the meeting or the family reunion.   

The log itself can be a spreadsheet or text file, something as simple as this:

Having a log of ideas is something well-known among creative people. Yet, I haven’t seen anybody in the corporate environment do this. What a waste! Do something with your ideas. Write them all down in the idea log file. Have a backup for hard times. It is important that you write everythingdown, so you don’t forget it, but also so you leave space for new ideas to come. Most importantly, these ideas tend to be the outside-the-box concepts LLMs struggle with.  

Having an analytics translator in the team is an extra advantage. Since they are in conversations with both data and business professionals, the ideas they generate are more varied and abundant.  

Prioritizing 

You’ve now written down all your ideas, whether they are proposals for AI solutions or ways to fix one of your products. If you still want to ask an LLM for extra options, this is a great time to do it. It might give you similar ideas, which can help you process better the ones you already have. Or it might give you radically different ideas, which can trigger further brainstorming.  

Once you have all the options in front of you, you and your team can easily discard many of them. Some will be ludicrous, and others financially unfeasible, but a few of them will be worth investigating further. Your team can rank those following one of the prioritization frameworks out there. At Xebia, we use a simple quadrant to measure impact and feasibility. 

IMAGE

This separates ideas into four clear categories:  

After going through the prioritisation process, you will have one or a few ideas worth pursuing to develop your solutions. You can then start refining them, which is the next step in the AI solution framework, but that’s a story for another day. 

Once again, an LLM can help you prioritise ideas, but it lacks the domain knowledge of an analytics translator.  In my experience, I’ve seen some small business owners get some useful tips from an LLM. When the company is larger, analytics translators’ data-savviness gives better results.

Concluding remarks about Analytics Translation

 So, do you need an analytics translator in the age of GenAI? Yes. Do you still need a good system to Ideate and Prioritize projects? I’m pretty sure you do. You may use GenAI at any point in the process to enhance your results, but my recommendation is to use it as late as possible in every stage.  

It will take you longer to do the steps manually, but then again, you will get to some precious outside-the-box ideas. You can see this process as a burden or as a way to keep sharp and more in touch with your business. The choice is yours!