10 human reasons your data product or solution may fail

10 human reasons your data product or solution may fail


Ready to taIf you listen to my podcast Experiencing Data regularly, you know that I love to ask my guests why they think the failure rate for AI, data science, and analytics projects continues to stay in the 80%+ range. Despite the growing phase we are in (particularly with advanced analytics), I cannot believe that every project that fails is simply due to technical reasons, lack of data, poor data, and all these types of computer things. The reality is that—as I just heard spoken by some other speakers at Predictive Analytics World last week—getting the customer experience right and making it simple is the hardest part. We need more empathy driving solutions, particularly when we start talking about AI that effects large communities and groups of people.

This got me thinking about what the cost is of poor solutions and how many different factors affect project successes that are not the math, models, data, and code. I think part of this has to do with how we think about the word "human" and whether we have designed our solutions with all of the human representatives and interests in mind.

We have to think beyond "customer" and even beyond "UX." UX and UI is not enough. It has "user" built-into the word, but what about the stakeholders who don't directly interact? What about the organization? What about downstream, non-present users (particular with AI solutions)?

Well designed systems and products gather feedback early and often from all the different humans in the loop. When we go up an elevation in our perspective, we enter the realm of what I think of as product design. Even if you're working on internal non-commercial tools, the product mindset can help you focus on delivering outcomes instead of outputs. (A quick aside: if you're a data science or analytics leader, you may want to consider focusing your next AI/advanced analytics efforts on products, solutions, tools, and applications that directly service your customers. This is where the biggest ROIs and impacts are being made, and if you're not worried about "digital natives," you probably should be.)

But, I digress.

Let's get back to YOUR project and the humans in the loop right now. When we start getting into advanced analytics (i.e. AI), our stakeholder plate (and risk) grows. Are you factoring in all of these different possible stakeholders when you design your solution?

  • Your customers ("users")
  • Your customers' customers
  • Direct business stakeholders
  • Cross-departmental stakeholders or staff who could be actively or passively affected by your solution (particularly AI)
  • Data oversight teams
  • Ombudsman
  • Data raters/trainers
  • Recipients of model feedback
  • Your community?
  • Society?
  • How will you handle all these competing interests?

How do you figure out who needs what?

How do you get to the point there is finally a real data science or analytics problem to work on?

What does success look like anyways, particularly in the short term vs. long term?

How do you get to the point you can disseminate all of these interests, requirements, needs, and wants into a data solution that is easy to use, creates value, and gets adopted by the organization/customers?

I can tell you this: it takes more than math, stats, models, and code to create indispensable data products, analytics, and AI solutions. Fortunately, they are skills that most people can develop, with time.

Ready to take the first step?

Join my DFA Insights mailing list or download my free Self-Assessment Guide for data products.

Need to go deeper?

My training seminar–Designing Human-Centered Data Science Solutions—might help, or your can learn how how I work clients on my services page.

 

Photo by Jason Blackeye on Unsplash