Alternative Data Weekly #235

April 25, 2025

John Farrall

Apr 25, 2025

Thanks for being here. If you enjoy reading the ADW, please share with a friend.

Thanks to our sponsor Big Data Federation!

Boost Your Strategies with AI-Powered Fundamental Forecasts.

The theme that emerged in this week’s email is … start with the business problem; get the quality right.

QUOTES

“Organizations implementing AI need explicit strategic direction connecting AI initiatives to specific business objectives: ‘Here's where we're going. We are going to be working in this market space with this client avatar, and here are our products that we're serving. Here's the problems where we're trying to solve.’" - David Sweenor’s The Grandmother Test: Building AI Trust Beyond Technology.

News Articles
Podcasts
Cool Charts
Final Thoughts (Focus is a Super Power)

#1 – Dylan Anderson’s Approaching Data Quality in Today's Complex Data World. April 2025.

My Take: Interesting take on data quality viewed through a data lens and a business lens. Business tends to focus on data as a reflection of truth/accuracy, while data teams think of data quality in terms of validation rules, fill rates, or taxonomies. AI is boosting data quality importance and issues as there is less transparency in AI systems, and I would add, AI always seems to deliver output with a high level of conviction, so, without some level of domain expertise, you may miss errors. It is the AI version of Gell-Mann Amnesia.

“…this a business-data gap, where business stakeholders and data teams view quality through different lenses”

#2– Vin Vashishta published How Do You Measure AI Initiative Success? Painting the Monetization Picture with Quantitative Measures. April 2025.

My Take: This is an important topic. Whether selling AI tools or data products, defining success up front is important. What is the business @outcome we are striving towards, and how do we measure that outcome? I have seen teams wanting to get a trial going or get some data intro production just to get it done. Particularly with “AI” as there Is a lot of pressure from the c-suite to be doing “something with AI”. Start with the business problem. Understand how you will measure the impact your tool is having, and measure throughout the trial process. For example, we are ModuleQ might define success as a metric of end-user daily engagement, increased engagement with other internal systems like CRM, and ultimately more client/prospect engagement (as measured by CRM inputs). We can watch these metrics closely from the first day we launch and tweak as needed, seeking feedback from the end-users throughout.

#3 – Isabelle Bousquette of the WSJ published Johnson & Johnson Pivots Its AI Strategy. April 2025.

My Take: “The company is making a shift to focus on only the highest-value GenAI use cases and shut down pilots that were redundant or underdelivering”
“And as the company tracked the broad value of AI, including generative AI, data science and intelligent automation, it found that only 10% to 15% of use cases were driving about 80% of the value, he added.”
So much to learn from this. Focus on the business problem. Solve problems. The tech is the tool, not the driving force.
This also demonstrated the importance of defining success up front. What does success look like? This can be engagement metrics, time savings, quality improvements…whatever, but the agreed-upon metrics need to be measurable.
I have discussed the fact that, in my experience selling “AI” solutions to large enterprises, I have seen the focus move from experimentation to focusing on projects that deliver ROI.

What else I am reading:

Abraham Thomas published Data and Defensibility. April 2025 (repost from last week … check it out)

Dan Entrup published Interview with Big Data Federation. April 2025.

Daniel Beach published Complicated != Good. April 2025. (paywall)

Ben Lorica published The Real AI Race: It's About Diffusion. April 2025.

Source: Satyen Sangani of Data Radicals Podcast interviews Sanjeev Mohan. April 2025.

My Take: Data products for dummies is the title of the podcast episode. They first spend some time defining a “data product”. This include traits such as re-usability, defined owner, trust (versioned like software), well-defined re-usable asset.
Theme: While AI is not perfect by any stretch, the trajectory towards improvement is faster than any other technology in history.
Another theme that is consistent in these types of conversations … that is to start with the business problem. Then deliver data that can be trusted to be accurate.

Lots of discussion about moats in recent weeks:

“Only moat for a company is the data.” - 26:30

Finally, discussion of AI Agents. Four parts of an agent (28:30):

Sense my environment
Reason what to do
Plan (series of steps)
Act

Highlights (46 minute run time)

Minute 02:00 - podcast starts; defining “what is a data product?”
Minute 06:45 - what problems does data product thinking solve?
Minute 10:00 - change from process point-of-view; how to build data products practically
Minute 15:00 - where do I start?
Minute 18:00 - first mention of GenAI
Minute 21:00 - importance of context
Minute 24:00 - eliminating data products that are not used
Minute 26:00 - AI agents overhyped?
Minute 27:15 - what are AI agents?
Minute 32:00 - every job will change and new jobs will be created
Minute 37:00 - how quickly will change occur?
Minute 41:00 - which near-term progress are you most excited about?