The Journey to Becoming Model-Driven
One thing I’ve been thinking about recently is how to help companies become model-driven. This is a problem that requires change on many fronts, though unfortunately many organizations seem to think hiring a bunch of data scientists is all that it takes. That may seem a bit harsh, and every organization makes lip service to changing business processes to become more data-driven or something like that, but the truth is, they aren’t doing it.
So what does it take to actually become model-driven:
1- Solid Data Foundation. You can’t do data science without data. And there has to be reasonable agreement on what the data is, and some semblance of what the truth is. Too many data science initiatives are built on top of sand, dooming them to failure.
2- A full-spectrum of data analytics services. You can’t jump straight to building AI when your users don’t have basic reports. Start with Business Intelligence tools, and work your way up.
3- A well-rounded data science team with technical skills and knowledge of the domain you’re working in. This is unlikely to be solved by hiring a handful of unicorns that know everything, but through carefully building out complementary skill sets
4- Deployment strategies in place from the get go. If you haven’t already through through how to get the models into the hands of users before you built them, it’s too late. No one wants models to be shelf-ware, and they’re most accurate the quicker they get out there. Deploying them through existing, unloved, clunky applications is a surefire failure point
5- Educate the business users early. Having models won’t make you model-driven unless people are using them. If you’re changing existing business processes leveraging data science, you need to be able to show the expected benefits and build inflexibility for people to adjust.
6- Keep the data science close to business. This is important hierarchically and in the use cases. If you aren’t working on the problems that matter to your business, you will fail.
7- Choose good initial projects. Don’t take on giant, mission-critical projects right out of the gate. You need to have projects with quick impact to get buy-in before you start fundamentally changing your business model.
8- Evolve into the hard stuff. To be truly model-driven means to have your core business process using data science. Think Amazon and its fulfillment modeling or Netflix and the recommendation engine. Don’t be afraid to make data science core to everything you do.
9- Iterate rapidly with your stakeholders. Data science is challenging, rapidly evolving,.and messy, and involving people in the sausage making will give them more faith in the end result than if it’s answer handed down from on high.
This is just a start, and doing all these things won’t guarantee success. But in my experience, if you don’t do these things, you are guaranteed to fail. What else do you think makes the difference?