Gain Support for Data Science Projects: The DATA 4-Step Process
In a session at the Data Science Innovators Conference 2023, Pfizer's Nechama Katan outlined a four-step process, Define, Advocate, Team and Act, for getting data science projects funded and adopted. Her core advice: solve a business problem in business language and make the leader the hero.
By The AI Enterprise Magazine Team4 min read
Key takeaways
- Define the business problem. Keep asking questions until a technical project becomes a business need such as retention or faster decisions.
- Make the leader the hero. Katan said that when leaders own the problem, they support the solution.
- Team up with customers. Shadow real work and test end to end to avoid last-mile failures.
- Invite criticism. Silence from partners means they have disengaged.
- Act on feedback. Offer workarounds, iterate and avoid arguing over requirements.
In a session at the Data Science Innovators Conference 2023, Nechama Katan, Director of Data Science at Pfizer, walked a small audience through a four-step process for winning support for data science projects. Katan, who introduced herself as a "wicked problem wizard" focused on helping organizations launch and implement technical solutions, called the method DATA: Define, Advocate, Team and Act.
The workshop was interactive. For each step Katan explained why it matters, showed how to apply it to a sample scenario, and asked attendees to practice on their own projects in chat. Her scenario was a chatbot built with natural language processing to give customers 24/7 support.
Why do data science projects need buy-in?
Katan said there are two hurdles. The first is getting leadership to say yes and fund the work, which means money, time or people. The second is adoption, because, as she put it, 80 percent of technical projects fail without successful adoption and integration. Define and Advocate address leadership, and Team and Act address customers and partners.
She asked the group to use "customers" and "partners" instead of "users." Customers are whoever is affected by the technology, internal or external, and partners are people co-developing it with you. She argued that all good projects are co-developed.
How do you define the real problem?
Most technical people lead with technology, Katan said, and managers do not respond to it. Leaders are driven by fear of layoffs, lost resources and looking or feeling stupid, and their eyes "glaze over" when teams talk about models and data layers. Making things accessible, she said, is not the same as dumbing them down.
Her method is to keep asking questions until you reach a business need. For the chatbot, she asked whether customers are online around the clock, whether low usage is the real issue, and whether a tool that needs less support would be better. That led from "build a chatbot" to a business statement: increase customer retention and net promoter scores through 24/7 support.
What is the real problem behind an enterprise data warehouse?
An attendee described an effort to get the business to use an enterprise data warehouse. Katan said the project was not the warehouse. The business wants to use its own data to make decisions quickly, and she cited an older survey finding that managers want answers to data questions within about 20 minutes.
The secondary problem is delivering answers that are high quality, secure and governed. So the real goal, she said, is decision support, and the way to start is to sit with senior managers and collect the 10 to 15 questions they most need answered.
Why should the leader be the hero?
Katan said the second step, Advocate, is counterintuitive. The person you are pitching is the hero of the story, and the data scientist is the guide. She compared it to Sam and Frodo in "The Lord of the Rings."
"If they own it, they will support it," she said. She described technical people as salespeople "in the good sense of the word," because they connect what the company needs to data science solutions. Phrasing the project around the leader's business goal also gives you a reason to spend hours with that leader learning what is needed.
How do you team up with customers and partners?
For Team, Katan warned against building the data science equivalent of a high-speed train that stops far from where people need to go. She described the "last mile" problem: a dashboard or model that works 90 percent of the time loses its savings if users dump the data into Excel for the rest.
Her advice was to look at the whole process before and after your tool, ask open-ended questions and shadow people at work. Tools are usually designed around the "happy path," she said, while real work includes lost passwords and broken search. She also urged teams to share early prototypes and to welcome criticism, because "when they stop giving you feedback you've lost them."
What does it mean to act on feedback?
The last step, Act, means listening and following up. Katan said any solution will have bugs and will not work exactly as a customer wants. In her own experience, a marketing team and a finance manager each needed the same analytic output delivered differently.
When conflict arises, she advised teams to assume the customer or partner is right, offer workarounds while a fix is built, and not take criticism personally. She said arguing over what was in the requirements is a sign the relationship has failed. Bringing in a neutral third party can help restart a stalled conversation. As a test of whether you are listening, she said, partners will keep talking.
Where to learn more
AI Connect's 2027 programming continues this kind of practitioner-focused learning. The organization has 24 free live webinars on the 2027 schedule, in four series: the AI × Enterprise Panel Series, AI Demo Days, Agentic AI Showcases and the Plug & Play Series. Details are at /webinars/, and registration is coming soon.
Frequently asked questions
What is the DATA process for data science projects?
Katan's DATA stands for Define, Advocate, Team and Act. The first two gain leadership support, and the last two build customer and partner adoption.
Why do data science projects fail to get buy-in?
Katan said teams lead with technology, while managers care about time, money, customers and quality, and are put off by technical jargon.
How do you define the real problem?
Keep asking questions until you reach the business need. A chatbot request, for example, became a goal of improving customer retention and net promoter scores.
What is the last mile problem?
A solution that works 90 percent of the time loses its savings if users export data to Excel for the remainder, Katan said.
This article is based on remarks by Nechama Katan, Director of Data Science, Pfizer at Data Science Innovators Conference 2023 on April 12, 2023. More coverage: More AI Connect Events.
Filed under: People, Culture & Leadership, AI Strategy & ROI
