# Marcus Delgado, VP of Customer Success at Fieldpath (84 employees, B2B project mgmt SaaS) — read of Customer Propensity Scorer AI, 2026-05-24

> 9 years in CS, spent the first four in support before anyone called me strategic. Currently running a team of 6 CSMs on Gainsight. Coach my daughter's U10 soccer team Saturdays. 35-minute commute each way.

## How I got here

I was Googling "predict customer churn 90 days ahead" because our Q2 numbers came in and I need to show the exec team something more proactive than a spreadsheet I update by hand. Google surfaced this as a paid result. I clicked because the meta description had "90 days" in it, which matched exactly what I was searching for. Not because of the brand.

## What I clicked first

The hero didn't do much for me: "Predict churn before it happens. Keep revenue by keeping customers engaged." That's the same sentence I've read on 15 different tools since 2021. I scrolled immediately.

What made me stop was the stat block: "78% Churn prediction accuracy 90+ days out." That's a specific enough number that I kept reading. 75% would have felt made up. 90% would have felt like a lie. 78% felt like someone ran a model.

## Where I paused

The case studies section. Specifically: "Mid-market SaaS firm noticed 40% of mid-tier accounts showing identical behavior patterns... 28 of 40 customers renewed ahead of schedule. Retained $420K ARR in one quarter."

I paused because those numbers are unusually clean. 28 of 40 is a 70% save rate, which is extremely high. $420K ARR from 40 mid-tier accounts means they're averaging $10.5K ARR per account, which is plausible but specific. I kept trying to find a company name, a logo, a link, something. Nothing. Three case studies, zero named customers.

## What I distrusted

Two things, and one of them is bad.

First, the soft one: the accuracy claim. "78% precision at 90-day prediction windows" is stated as a product fact, but there's no mention of what dataset this was measured on, what the false-positive rate is, or whether "precision" here means precision in the ML sense (of the customers you flagged, 78% churned) or something else. Precision without recall is a partial picture. Anyone who's run a churn model knows you can get high precision by just flagging very few customers. I'd want to know both numbers.

Second, the one that actually bothered me: I scrolled to the bottom of the page and found this:

"Honest disclosure: we don't have live customers on this idea yet. We shipped the strategy package; you ship the customer conversations."

So the case studies above are fictional. The $420K ARR retention story. The $8.2K MRR growth from at-risk accounts. The enterprise firm that dropped churn from 22% to 8%. All of it is illustrative. And I had to scroll to the very bottom of the page, past three case studies with dollar figures and checkmarks, to find out they aren't real. That's not a transparency win. That's burying the lede.

## What would convince me

If this were a real product: one named customer, willing to get on a 15-minute call with me. Not a case study. A call. Churn tooling is relationship-heavy. The reason I switched from ChurnZero to Gainsight wasn't the feature set; it was that a Gainsight CSM spent 3 hours with my team before we signed. If someone running this tool would let me talk to their second or third customer, not their best one, I'd take that seriously.

On the model accuracy front: show me a confusion matrix or a lift chart from a real customer's data. Even anonymized. Something that shows the model actually works on data shaped like mine, not a hypothetical SaaS with clean Snowflake exports.

## What I'd ask in an email reply

1. The case studies on the page have specific numbers but no company names. Are those real customers, or are they illustrative of what the product is designed to achieve? I want to understand what I'm actually looking at.

2. Your 78% precision figure: what was the false-negative rate in that same test? How many churners did the model miss entirely? That number matters more to me than precision.

3. We're on Gainsight already, which has health scoring built in. What does your model catch that Gainsight's health score misses? I'm not going to run two churn tools unless there's a specific gap you fill.

## Verdict: dismissive

Not of the idea, which is legitimate. But of the page specifically, because it presents fabricated case studies with real dollar amounts and saves the "no live customers yet" disclosure for the footer. I don't think that's honest marketing. If they'd led with "this is a new product, here's the model logic, here's a free trial on your real data" I'd have kept reading. The gap between the confidence of the hero and the admission at the bottom is too wide.

---
*Memo by skeptic persona, generated 2026-05-24. Studio breaks own self-grading loop.*