• Daily Success Snacks
  • Posts
  • Why “What’s the model’s ROI?” is the fastest way to trigger existential dread in any data team.

Why “What’s the model’s ROI?” is the fastest way to trigger existential dread in any data team.

Expectation management is harder than gradient descent.

Author

Glenda Carnate
December 12, 2025

Read time: 2.5 minutes

When they ask for the model’s ROI… and I’m still trying to descend toward the minimum of their expectations, that’s when I know the meeting is no longer about math... it’s about survival.

Last month, a team presented an early-stage model... still fragile, still learning, still basically a toddler with a GPU.

Before they could finish slide three, an executive interrupted: “So what’s the ROI?”

You could feel the room collectively attempt a silent gradient descent on their stress levels.

Because the model wasn’t trained. The features weren’t finalized. The validation wasn’t clean.

They weren’t even past “does this thing make sense?” And somehow we had fast-forwarded straight to “how much money will this print by Q2?”

This is the moment every data scientist becomes a philosopher. 

How to Handle ROI Questions When the Model Isn’t Ready?

1. Reframe ROI as “expected impact”.
When you can’t quantify the return, quantify the direction: time savings, accuracy improvements, error reductions, and throughput boosts.

2. Share scenarios, not dollar amounts.
Stakeholders understand stories better than statistical intervals.

3. Anchor expectations to maturity levels.
Prototype ROI ≠ Pilot ROI ≠ Production ROI.
Clarify which stage you’re actually in.

4. Use confidence bands, not guarantees.
Executives respect ranges when they’re explained clearly.

5. Shift the conversation to decision trade-offs.
“Here’s what improves the ROI.” Suddenly the question becomes collaborative, but not confrontational.

💡Key Takeaway: 

Never promise the model will find the minimum. Just promise you won’t let expectations explode to infinity.

👉 LIKE if you’ve ever been asked for ROI before the model had a pulse.

👉 SUBSCRIBE now for real talk on AI, analytics, and expectation management.

👉 Follow Glenda Carnate for breakdowns that make technical chaos feel human and hilarious.

👉 COMMENT if you’ve had to defend a model that wasn’t ready for daylight.

👉 SHARE with someone who knows the pain of explaining “it’s still training” for the 14th time.

Reply

or to participate.