Two teams used the same word, churn, and told leadership two different stories. Garima Gupta explains why the most underestimated data debt is a business that hasn't agreed on what its own metrics mean, and how to fix the definitions that cost you the most trust.

At an early-stage startup with 130 KPIs, product measured 60-day churn and finance measured 30-day churn. Investors saw one story while the roadmap followed another. Garima introduces the Blast-Radius Test for deciding which metric to settle first. She also shows how AI raises the stakes: a self-serve analytics agent failed to get adopted because nobody had written down the exclusions an experienced analyst knew by heart.

In this episode:

- Data debt vs. trust debt, and why stakeholders start pulling their own numbers

- The Blast-Radius Test: upward or external? compared across teams? driving another team? hard to reverse?

- Why capability matters, but trust decides whether people use enterprise AI

- Getting finance and product to agree: show the definitions side by side, then ask what decision each one serves

- What a semantic layer should capture: meaning, calculation and exclusions, and an owner

- How to get definition work funded without "selling data cleanup"

Chapters

00:00 One word, two stories: churn

01:28 Meet Garima Gupta

01:54 The data debt that gets underestimated

03:29 You don't have to fix every metric

04:49 The Blast-Radius Test

06:32 Scoring churn: 4 out of 4

07:29 AI raises the stakes: the self-serve analytics agent

09:42 Capability matters; trust decides adoption

10:15 How to get two teams to agree on one definition

12:47 The semantic layer as a contract

14:53 How to get this work funded

16:12 The hard truth about enterprise AI

Garima's LinkedIn newsletter, Curious Connect: https://www.linkedin.com/newsletters/curious-connect-7114858899605766144

How do you hire data scientists and analysts when AI can write the SQL? Garima Gupta shares the hiring philosophy she drew from about 2,000 interviews: hire for the floor, bet on the slope.

Technical skill is the floor, a real gate every candidate has to clear. The slope is curiosity and ownership: whether someone understands the business problem before reaching for a solution. Garima compares two candidates. One was technically brilliant but couldn't say what his model changed. The other missed the hardest SQL question, then was promoted twice in about 12 to 15 months.

In this episode:

- Why calibration matters more than better interview questions

- Signal vs. noise: "Noise is expertise performed. Signal is understanding demonstrated."

- Putting your best-calibrated hiring judgment early in the funnel

- Three changes to the data interview loop: case first, watch their questions, test ownership

- Why you shouldn't blend technical and business judgment into one score

- How the bar changes from analyst to senior IC, manager and director/VP

Chapters

00:00 The brilliant model with no business impact

01:39 Are we interviewing for the right people?

02:07 What 2,000 interviews taught me

03:02 Hire for the floor, bet on the slope

04:19 The candidate who failed the hardest SQL question

07:07 You can teach the floor up; the slope is harder

07:55 Signal vs. noise in interviews

09:13 Put calibrated judgment early in the funnel

10:26 Three changes to the interview loop

12:04 Scoring: don't average the floor and the slope

13:11 How the bar changes with seniority

14:11 Why judgment matters more in the AI era

15:26 Take another look at your open roles

Garima's LinkedIn newsletter, Data With Consequence: https://www.linkedin.com/newsletters/curious-connect-7114858899605766144