How to plan an analytics charrette

A charrette is an intense, collaborative session, that enables designers to draft a solution to a very complex problem.

It’s a technique first used by artists. Then designers picked it up. And then later still, urban planners. And then a few brave souls wisely invited stakeholders in on the process.

Finally, this approach would evolve into software development and web development.

It is very applicable to solving analytical problems.

First, consider the natural law below.

 

 

 

 

 

 

 

 

 

 

 

In analytics, the proportion of what we don’t know always grows as more knowledge is added. The more imaginative the analyst, the steeper the curve.

Get three or more analysts into a room together and it really scales.

An analytics charrette, if unstructured, will end in a set of questions as opposed to a focused solution on a single problem. It will end badly.

So we focus the entire charrette on a problem, and, ideally, a single optimization objective.

A dataset relating to the problem and the optimization objective is extracted, transformed, loaded, cleaned and distributed before the charrette. If resources allow, the cleaning should conducted by three analysts who all agree on the final dataset. It’s important that everybody participating in the charrette is working off the same version of the data.

The charrette opens with exploration of the dataset, salient features, and basic facts. It’s during exploration that facts are discovered and accepted, relationships accepted, and causal arrows asserted. In other words, participants gain an intuition and share a common experience.

This part of the charrette can take anywhere between 1 hour to 8 hours depending on the number of columns and tables relating to the problem.

Then ideation begins. The activity is centered on asserting solutions to the problem. It’s extremely important that systems thinkers and design thinkers are able to explain and express the solution sets. This process can take anywhere between 1 hour to 8 hours, depending on the complexity of the problem.

Then time must be allotted so that people can prepare their artifacts. These artifacts may take the form of stickynotes, paper prototype models, regression models, designs, simulated results, or, tables. The duration depends on the complexity.

Then the culling must begin. A common technique amongst designers to cull ideas is to hand participants stickers to place next to the concepts they prefer. The concepts with the fewest stickers by the end of the hour are culled. The concepts with the greatest number of stickers survive.

This technique can be used in an analytics charrette.

Sometimes there is a pause, where the most popular (best) ideas are optimized into a single, agreed upon, design to the problem.

Planners should plan out the agenda with plenty of collaboration time set aside, have lots of wall space and whiteboards set out, and should have a singular dataset that is open, accessible, and cleaned for all participants.

At the end of the charrette, a solution to a complex analytical-design problem comes out, and it’s well worth it.