Reusing analytics models starts with team alignment

An analytics model built once and reused across several teams can save a company a lot of money. It only works when the people who build models and the people who use them are well aligned. This post covers practical ways to build that alignment and a story from my own career about what it takes.
I first gave this as a talk at a Transforming Data with Intelligence virtual conference, where the topic was reusing analytics models.
Put analysts in the room
One of the most effective ways to align the teams that build models with the teams that use them is to have senior analytics people sit in on weekly stakeholder meetings. I recommend it for marketing, sales and product teams alike.
Done well, information starts flowing and the teams start to feel like one team. I’ve seen it work many times, and it makes for better collaboration and a better place to work.
At that conference, a vice president running a global team asked a great question:
Do you have any recommendations on getting engagement and changing culture among stakeholders on the process and challenges of machine learning? The value seems to be clear, but understanding of what’s needed and some business process change is still a major challenge.
It reminded me of a line often attributed to Theodore Roosevelt: “Nobody cares how much you know until they know how much you care.” Adding analysts to product meetings isn’t enough on its own. The gains come from open conversation.
Putting a math-literate analyst next to a marketer who “just knows” what to do can feel forced at first. A few things make it easier.
Ways to build bridges between teams
- Hold a hackathon. It puts people from several disciplines on one team with one goal.
- Invite in both directions. Invite data scientists to the weekly product meeting, and invite the head of product to the weekly analytics meeting.
The point is for each team to show it cares about the other and values what the other knows. Product teams clearly benefit from a data-driven approach, and without the product team’s buy-in, the best data-driven approach goes unused.
A dashboard story
Years ago, as a director of data science, I needed company-wide dashboards. Lots of people needed different things to see every day. I hired someone to build the dashboards, and the first thing I asked her to do was walk around the company with a crayon, so she wouldn’t get lost in small details, and ask each person what they needed to see daily to make decisions.
She did, and she built a dashboard for each person. A week later, I asked her for a report of who was using them. It’s easy to see this in a tool like Tableau.
The result was unsettling but predictable: more than half of the people weren’t using the dashboards she’d built for them. She had built exactly what they asked for.
So I asked her to go back to everyone who wasn’t logging in and say, “Sorry I built the wrong chart. Would you tell me how I can make this better for you?”
That was the key. After a few small changes, use rose above 90%. That bit of humility opened the door. From there we had the start of trust, and we could begin feeding people the insights that made the company truly data-driven.
Alignment comes first
When teams feel that they matter to other teams, they work together and achieve more. That trust is what makes it possible to reuse a model. A recommendation model built for a website can also power marketing emails, but only if the two teams talk. I give an example in why AI projects fail to pay off.
It’s also why our engineers work inside your team. You can read how that works in how we work with clients.