(This originally appeared as an article on LinkedIn. I periodically mirror such articles here as blog posts.)
I wrote a lot of software for companies earlier in my career. One lesson I learned very quickly is:
If you’re building custom software,
you are now a software company.
(Even if you don’t want to admit it.)
What I mean is that you can’t just hire a bunch of developers and throw vague specs at them.
In order to succeed, you also need to:
- Build a full software development organization, with roles related to product and project management, release management, testing, …
- Know when to build something extremely specific to your company, and when to stick with industry-standard tooling.
- Have the (company-wide!) discipline to know when to not build your own product, but to adapt some off-the-shelf product instead.
and so on.
The same holds true for the data space:
Once you say that your company wants to “become data-driven” or “do AI,”
you are now a data company.
(Even if you don’t want to admit it.)
You can’t just hire data scientists and hope for the best. You need to:
- Build out a full data organization that also includes data engineers and possibly an MLOps team.
- Establish organization-wide data literacy, such that stakeholders, product owners, and the legal team are all up to speed on what is (not) possible with data and what you plan to do with it.
- Start with strong standards around data collection, storage, and access. (You’ll thank yourself later.)
So when your company says that it wants to start using ML/AI, and their only “plan” is to hire some data scientists and turn them loose … It’s time for some tough conversations.
Do yourself a favor: develop a real plan (based on your business model and challenges), get down to specifics, and build out the team. In that order.
Anything else is just a gamble.
And with the cost of even a modest data operation, that makes for a very expensive lottery ticket.