Are you trying to leap directly from data to action? It’s not that simple.

No matter how much you want to, you can't get actionable insights directly from data. You need to apply knowledge.

Nov 19, 2021

The same set of data will lead to different insights – depending on who’s looking at it, what they’re looking for, what they’re capable of seeing, what they want to do, and what they can even conceive of doing …

Two accountants can look at a set of books and understand them – their knowledge of the context lets them turn the raw data in the books into information. They’re likely to get similar information to one another. If they know what a business owner is trying to do, they’ll also be able to provide insights into the health of the business, as well as suggest actions.

But I’m not trained as an accountant and have limited knowledge of the context. When I look at a set of books, I don’t see much information, let alone glean any insights.

It takes layers of knowledge to turn that raw accounting data into actionable insights.

I’ve seen a similar pattern across all sorts of data.

OK looks nice enough. But how can I use this?

Example 1: Actionable Insights for eCommerce Merchandisers

This model first emerged from work I did with Qubit. Our team was trying to generate actionable insights for eCommerce merchandisers, using raw data we collected from their websites.

We knew we had a load of data. And we knew our customers really wanted us to turn it into actionable insights for them. We just didn’t know how.

In the end, we tried 5 reports that completely missed the mark before we landed on a report that had some promise. We had to make a dozen more iterations of the report and gain a lot of knowledge along the way before we successfully made the three hops above and turned raw data into valuable action.

We had to learn a ridiculous amount about their context: their role and environment, but especially how they interpreted the data – the stories they told.

Then we had to learn much more about their strategies so we could filter through all the information and show just the important bits, not the background noise. What were success and failure like for them? Which few bits of information were salient or surprising to them? These were the insights – the bits of information that changed what they considered doing next.

Finally, we realised that many of the insights suggested actions they currently couldn’t take within their time constraints. We learned we needed to increase their capabilities by building some new functionality for them.

The final reports and functionality turned out different from anything we’d imagined when we started out.

Example 2: Making sense of data in Pivot Triggers

After finishing the initiative above, I documented the way the team worked so I could share it with other teams – this became Pivot Triggers.

This data → actionable insights model also makes an appearance in Pivot Triggers. Meta, eh?

When we probe, we do something that’s inside our control and hope that actors outside our control will do something in response. We look for the signals those behaviours will give us, and we’re always asking: what signals do we need to see to feel that we want to continue with our plan?

Those signals are our raw data. Some are numbers, some are qualitative feedback, and some are as subtle as our subconscious firing emotional states while we make the probe happen.

We always end up with more signals than just the ones we went looking for. Even then, we know we’re paying attention to a tiny fraction of all the possible data. On the other hand, it’s infinitely more data than we get when we don’t probe!

Then we do a Post-Probe Review, which follows the little knowledge jumps above, and turns the signals into a new set of probes (aka data into action).

Here’s how that works:

Step 0) Review the data

First we share and review the data — the raw signals — but withhold judgement.

What were the cold, hard numbers?
What qualitative data did we get?
Were there any other signals we noticed? e.g. Was it more or less work than we expected? What surprised us along the way?

1) What?

Armed with knowledge of our context, we make inferences to turn data into information.

What do we think these data mean?
How do the signals compare against the triggers we set? (We set a trigger level for each signal before we start so we can’t wheedle our way out of healthy cognitive dissonance.
How do we feel about the comparisons? I remind them that we didn’t succeed or fail here, we simply gathered information.
What’s really going on here? I push teams to come up with at least 3 different stories about what the data mean. We don’t want to get trapped in one simplistic story.

2) So what?

Based on the information, and armed with knowledge of our strategy, we turn information into insights – about what new options could be interesting to explore.

We answer the big question: do we invest more or do we pivot?
Invest! → when the information points towards investing more, this builds our momentum and confidence. Often we see new options that are more exciting than our initial idea. Alignment starts to emerge.
Pivot! → when the information points us towards a pivot, this helps us think again. Sometimes we drop the whole thing, but most of the time, we diverge from our initial plan and create new options. We stay a little longer in exploration.

3) Now what?

Based on the insights, and armed with knowledge of our capabilities, we turn insights into action.

We design a new set of probes to explore some of our new options
Sometimes the next probes are pretty obvious, sometimes we need to do some more thinking, or even add to our capabilities

Here it is in diagram form, but with the words we use in Pivot Triggers:

Bonus! An alternative pyramid version

Jorunn Newth did a remix of my diagram above, and I liked it so much I wanted to share it here too, with her permission:

Maybe one or the other is clearer for you?

Summary

You can’t leap directly from data to action without applying knowledge of context, strategy and capabilities.

I’d love to know what you make of this model. Like all models, it’s wrong … but I’ve found it to be very useful.

If you’ve ever struggled with the whole “data driven” thing (I know there are a lot of you out there!) then I hope this helped make sense of why it’s not that simple.

Or perhaps you’ve been data-driven in a different way. Do you have a different model?

Drop me a line and let me know!

I wrote a follow up to this article: Signals > Stories > Options, where I share a practical example of using these layers with a product team.

We also recorded a podcast episode where we read this article and annotate it with additional thoughts and context:

Big thanks to Dave Snowden for “what, so what, now what” (originally from Borton’s development framework in the 1970s) and to Qubit for putting me to work with that team figuring out actionable insights and pivot triggers.