Should organizations first clean up their data so that more realistic AI solutions and data products can be created down the road? 

-OR- 

Do you put a vision in place first...to inform the minimum resulting "data prep" that needs to happen to enable these services? 

In other words, data first? Or a vision of the future first?

I think the jury is out on this still, and it's something I'm opening to changing my mind about.

The dirty truth with the AI space in particular right now (as I hear it) is just how much data wrangling (cleaning, pipelines, quantity issues, etc) is required before organizations can even start to turn their data into predictive models, data products, new analytics solutions, and the like.

Here's my take on this as of today.

I think if you have zero idea of what's possible or needed, and you take a data first approach, you are a lot less likely to just "fall into" a valuable solution. This may be why the success rates for AI and analytics projects are so low. Is it possible? Sure. But, it's probably a lot more expensive, slower, and less reliable to build solutions this way. This is the "build and pray" model. No pun intended.

So, can valuable and useful human experiences be created in this model? Perhaps, but let's perhaps look at this way.

There's the rehearsal, and there's the concert.

There are practices and scrimmages...and then game time.

Which mode are you operating in?

I sometimes think of it as operating in "production mode" vs. "lab mode."

In lab mode, you may go through the entire process of building something or simply exploring a data set, only to see it not get used, create any value, or even get finished. An unsupervised machine learning model might fall into this category. Was it a waste? Perhaps not. You might have built some future infrastructure that will make the next project easier and faster, and you probably learned a lot along the way. You practiced. You rehearsed. However, you may not have created any immediate or obvious business or customer value...yet.

The question is, do your customers or stakeholders agree and understand whether your initiative is operating in lab mode or production mode?

What expectation of value creation was conveyed at the beginning of the project?

Was there any firm success criteria even established, at all?

Are you in search of a problem to solve, or a solution to build?

Best that you make sure everyone on your team, particularly your fiscal sponsor, understands whether they're paying to watch a rehearsal...or the concert.

As a designer, I have always gravitated toward a problem-first mindset, even for data products, so we can be sure the technology solution services the people. By uncovering latent needs and opportunities, we de-risk the possibility of creating a solution or product nobody will use (or worse, harms people).

Design is also just as much subtractive as it is additive and while this is easiest to understand at the user interface design layer, I think it can be applied at a higher level. Design can help us understand "are we making business and customer progress?" Or just engineering progress?

Innovation is a messy road; it's not a straight line and it requires us to accept some strikeouts for sure. However, I don't think that it means "start with absolutely no structure, no idea of what any customer or stakeholder wants. I don't think it means, "jump into any tasks that look actionable now and MAYBE necessary down the road."

The other way?

Your leadership likely has quarterly or annual organization objectives; some may be internal-facing only; others external (i.e. new products/services). Likewise, your sales people, customer service people, researchers, and marketing folks can all be sources of latent end-user problems and needs that are "simmering"—particularly if you don't have that critical resource I like to call the "data product manager." The big innovative stuff is likely going to come from new products and services customers interact with, and if you can align those business objectives and customer needs up, then to me, you're helping remove variables, distractions, and risk from your innovation efforts because you're focused on an outcome ("achieve X customer success") vs. an output ("clean the data, build a pipeline, stand up a private cloud").

As for that jury?

I think there's an argument that can be made for starting with data and working backwards to a product, model or service, however, this seems to me like it would come with higher risk overall. It seems to have more chance of failure, and it requires very forward-thinking management who are willing to go to bat many times before they have any expectation of hitting a base hit. It also seems like a method that would necessitate a very creative and open environment for experimentation and R&D; one that is not driven primarily by profit or other short-term objectives.

This leads me to a question for you:

Are you in a business or organization that is trying to use AI, but they aren't quite sure where to start? (i.e. they don't know what they need or want, beyond not wanting to miss out?)

Hit reply and let me know (and when I say that, I'm not being rhetorical!).

I'm considering offering a new training or consulting service for this year that would use improvisation and creative thinking methods to help teams generate a volume of ideas for new data products and AI models that are human-centered and value-producing.

Do we need this? Do you/your company need this?

I hope you'll let me know - drop me a line at brian@designingforanalytics.com.

Photo by Victor Garcia on Unsplash