Key Results for usage of data in decision making

Let me flip this thing inside-out for clarity:

You simply can't reconcile "We enable data-driven decision making" with the statement that "we have very little data." But: "Are you the decision-makers?" Probably not.

Therefore, I would turn to your [internal] customer community – to the people who actually are tasked to make the "decisions" that you speak of. Directly ask them what they need to help them do their jobs better. Quickly cobble-together prototypes to show them – even if these consist of nothing but static HTML. Freely discuss with them ... brainstorm with them ... what data you have, and what might you be able to do with it that would in some material way benefit them. Guide the conversations, but let them do most of the talking.

You're focusing on the process ("data-driven decision making") and not on what you actually want to accomplish (a better product). You should try it the other way around.

Start by thinking about your vision for the product. Where do you want it to go? How do you want people/customers to think of it?

Then start thinking about how you might recognize whether your vision has actually happened: the degree to which the product has arrived at where you want it to be, how people think of it, what characteristics it needs to have, etc. Then start thinking about what sort of data you would want to collect in order to recognize all this, tempered by what sort of data you actually can collect.

Then go out and collect it. In other words, devise a data-collection strategy and implement it. That implementation should be your outcome, and your Key Results should relate to the successful collection of data.

Once you have enough data, and given that you know exactly how each piece of data relates to the vision/strategy, a dashboard/report becomes easy to design. Data-driven decision-making follows naturally.

Some of the usual metrics for decision-making are time, errors, and cost.

So you can measure how often your projects hit deadlines (% compared to previous results) with your data-driven approach.

How often it resulted in an error (negative target metric).

And I don't think it's going to work for cost but it depends on the nature of what you're doing.