# Marcus Thill, Head of Customer Success at Cloverfield (HR tech, ~70 employees, $2.8M ARR) — read of silent-churn-detector-ai, May 24 2026

> 8 years in CS, the last 3 trying to get Gainsight to do something it was never designed for. Currently managing a team of 5. Train commuter. Coaches his daughter's U10 soccer team Saturday mornings which means he is frequently tired and impatient with nonsense on Monday.

## How I got here

Typed "predict customer churn 30 days before cancellation" into Google last Tuesday. This showed up on page two, which I usually ignore, but I'd already read three Totango blog posts and was running out of patience. Clicked through. Skimmed in about four minutes total.

## What I clicked first

The hero stat trio stopped me for a second: "85% Prediction Accuracy. Real-Time Alerts. Reduce MRR Loss by 30%." The 30% MRR loss reduction is what made me keep reading. That's a real claim if true. We've been estimating our preventable churn at roughly 25-35% of total so it felt directionally plausible, not wildly inflated.

The framing underneath the fold was actually pretty crisp: "By the time most SaaS teams notice the pattern, the decision is already made." Yeah. That's it. That's the whole problem in one sentence. Credit where it's due.

## Where I paused

The pricing structure. $399/month plus "2% of MRR over $10K." That's a weird construct. So if I'm at $2.8M ARR, let's say ~$230K MRR, that's $399 plus $4,400 a month, which is about $57K a year. That's not "affordable," that's a real budget conversation. The headline says "transparent pricing" but this formula buries the actual cost for anyone above $10K MRR. I had to do math in my head on a train.

## What I distrusted

"$340K MRR saved in first 6 months by the average customer." Then three paragraphs later, buried at the bottom: "Honest disclosure: we don't have live customers on this idea yet." So what average customer? That stat is fictional. The "+28% increase in customer retention after 30 days" is also fictional. These are projections dressed up as testimonials, and the disclosure at the bottom pulls the rug out completely. The comparison table is also entirely self-authored. No third-party data, no named competitor in the columns, just "Manual Cohort Analysis" and "Reactive Churn Alerts" as straw men.

The phrase "Built by operators who understand CAC, LTV, and expansion" is exactly the kind of thing someone writes when they want to signal operator credibility without demonstrating it. Anyone can write that.

## What would convince me

One real case study with a named company, their MRR range, and what intervention they actually ran after getting the alert. Not "saved X MRR" but "customer was flagged, CS rep sent this kind of outreach, here's what happened." I want the intervention loop, not just the detection claim.

And on the 85% accuracy stat: what's the baseline? If 60% of my customers never churn, a model that predicts "no churn" every time is 60% accurate. What's the false positive rate? How many healthy accounts are my CS reps going to get paged about unnecessarily?

## What I'd ask in an email reply

1. The disclosure says no live customers yet. What's the actual status of the product? Is there working software I can connect to a sandbox environment, or is this still in build?

2. How does the 2% MRR fee get calculated and billed? Monthly true-up based on current MRR? If we have a good month and MRR jumps, does our bill jump retroactively?

3. The "85% accuracy" stat. 85% of what, measured how, on what kind of dataset? Was this a backtest on historical data or a live holdout test?

## Verdict: on-the-fence

The problem framing is genuinely good and the product concept is real. But I caught the "no live customers" disclosure and now every stat on the page feels like a mock-up. I'd reply if someone reached out directly, but I'm not filling out a demo form for something that might be a slide deck.

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