# Marcus Delgado, Director of Revenue Operations at Arstella (B2B SaaS, ~130 employees) — read of Forecast AI, May 8 2026

> 9 years in ops, started in CS ops, ran my first RevOps function at a 40-person startup, now managing a Salesforce instance with 22 reps and a pipeline data problem I've been trying to solve since Q3.

## How I got here

Googled "salesforce pipeline accuracy miss q4 early warning" on a Tuesday morning because our CRO just got off an earnings call where our forecast was off by 18%. Someone in a RevOps Slack (the Pavilion one) had posted a link to this in a thread about Clari being overkill for sub-200-person teams. I clicked it mostly to have something to show my boss that wasn't a $60k enterprise contract.

## What I clicked first

"Sales managers forecast with gut feel. RevOps managers forecast with data." That line landed. It's the exact argument I've been making internally for two years. It's not original but it's clean and it's true and it made me keep reading. The "30 days early" claim in the hero is also the specific thing I need, so I didn't immediately bounce.

## Where I paused

The "No setup, no training, no black boxes" line. I stopped there for a real minute. Because those three things are in direct tension with each other. If there's no setup, the model doesn't know my deal stages, my sales cycle length, my industry, or how my reps actually move deals. If there's no black box, I need to understand what signals it's using. But if I have to configure those signals, that's setup. I can't have all three. The page then says "It learns deal characteristics, velocity patterns, and close probability from your CRM data" -- okay, so it IS doing setup, it's just automated. Fine. But then say that. "No setup" is misleading and anyone who's touched a CRM integration knows it.

## What I distrusted

"Our AI is trained on 500K+ deals across industries." This is doing a lot of heavy lifting and I don't believe it the way it's written. Trained on 500K deals from what companies? What sizes? What sales motions? A 14-day transactional SMB deal and a 9-month enterprise deal have almost nothing in common. If the model was trained mostly on high-velocity SMB data and I'm running a 120-day mid-market motion, that 500K number is noise.

Also: the hero says "real-time forecasts" and "updated live as deals move." Then the how-it-works section says "every morning, you get updated forecasts." That's not the same thing. Pick one.

The "Built by Wishdeal Studio" footer credit is also a yellow flag. This looks like a product studio that ships multiple tools. That's not disqualifying but it makes me wonder who is actually maintaining this, whether there's a real team behind it, and what happens when I have a data question at 4pm the day before a board meeting.

Pricing at $299/month for 5 reps sounds reasonable but the "scale as you grow" line with no actual scale pricing on the page is a classic soft bait. I've been burned by that before.

## What would convince me

Show me one forecast accuracy comparison. Not a testimonial. An actual number: "Companies using Forecast AI hit within X% of their quarter-end number vs. Y% without it." Even better: show me what the model flagged 30 days before quarter close for a real customer, then show what actually closed. That's the proof I need. A case study from a company with a similar sales motion to mine (mid-market, 90-120 day cycles, Salesforce) would do more than all the copy on this page combined.

## What I'd ask in an email reply

1. The page says "no black boxes" but I need to know which CRM fields the model actually uses. If it's just stage and close date, I could build that in Looker. What signals differentiate your model from a weighted pipeline report?

2. What happens in month one before you have enough of MY deal history to make accurate predictions? Do you fall back entirely on the 500K deal training set, and if so, how well does that actually work for a company with a 90-day average cycle?

3. Is there a human I can talk to when the model flags something that looks wrong to me? Or is the product fully self-serve?

## Verdict: on-the-fence

The core positioning is right and the price point isn't offensive, but the page contradicts itself on "real-time" vs "every morning," the "no setup/no black box" claim doesn't hold up to 10 seconds of scrutiny, and there's no actual proof that the model works. I'd reply to an outbound email from the founder if it included one real accuracy stat.

---
*Memo by skeptic persona, generated 2026-05-08. Studio breaks own self-grading loop.*