We Almost Chose Delve. Here Is Why We Went Another Way.

A few months ago, we ran a vendor evaluation for a piece of our compliance stack. Delve was on the shortlist. They had slick sales, lots of AI, and on paper they looked like a good fit. We looked at them long and hard.

We went another way.

I want to write about why, because the reasons were not a single headline moment. They were small. They were the kind of thing that gets waved off as a vibe check when it deserves more than that. And then, a few weeks after we passed, the public story around Delve took a turn that made me think the vibe check was doing more work than we gave it credit for.

What we liked

Credit where it is due. The Delve pitch was well-produced. The website was sharp. The demo flowed. The founders are young, technical, and have obvious conviction. They raised a $32M Series A from Insight Partners last summer at a $300M valuation, they came out of Y Combinator, and the brand had real momentum. If you walked into that meeting cold, you would see a serious company.

We walked in interested. We walked out unsure.

What made us pause

Three things, in order of how quickly they registered.

The pricing was higher than the scope seemed to justify. Not dramatically. Enough that when we laid the quote next to the quotes from other vendors doing roughly the same work, the math did not line up with what Delve said the product was actually going to do for us. That alone would not have killed it. Pricing conversations happen. But we made a note.

The answers were surface-level. We ask pointed questions on vendor calls. We expect some of them to get kicked to a second meeting with a solutions engineer or a security person or a legal partner. That is normal, and it is often a good sign: it means the company knows when to bring in depth. What we got from Delve instead was a lot of confident, fluent answers that did not actually go anywhere. If you pushed on the underlying mechanism, the answer flattened out.

And the deferrals went to the platform, not to a person. This is the one that stuck with me. When a vendor says “our legal team will follow up” or “let me get a compliance engineer on a call,” that is a healthy escalation path. It means there are humans with depth behind the front line. When a vendor says “the AI handles that” or “our platform takes care of it,” that is a different answer. It sounds like a feature. What it often is, in practice, is a place where a human used to be and is not anymore.

None of those three things, on their own, were disqualifying. Together, they were our sniff test. We passed on Delve. We picked another vendor. We have been working through onboarding slowly, deliberately, and so far we are happy with the choice.

Then the story broke

In March 2026, an anonymous account called DeepDelver posted allegations that Delve had been providing fake compliance evidence to customers. Delve disabled the “book a demo” button on their site. Insight Partners quietly deleted their investment announcement post. Y Combinator removed Delve from their portfolio. TechCrunch picked it up. Delve posted three responses across late March and early April.

I want to be careful here, because the story is still unfolding and Delve deserves a fair accounting of their side.

Delve’s position, summarized from their March 20 statement, their March 24 customer-support update, and CEO Karun Kaushik’s April 3 post: the anonymous allegations are the work of a malicious actor, not a whistleblower. They say the attacker bought the product under false pretenses, exfiltrated internal company data, and used cherry-picked screenshots and stolen context to launch a smear. They also acknowledge that they grew too fast and fell short of their own standard, apologized to customers, and put remediation measures in place: a new auditor network, free re-audits and pentests.

Both things can be true. A company can be under a coordinated attack and have grown too fast and cut corners. The courts and the audits will eventually tell the story. If the allegations turn out to be untrue, I hope that gets resolved quickly and Delve gets back to building. If the allegations turn out to be true, that is a very big problem, and everyone in the AI-for-compliance space needs to learn from it yesterday.

Either way, our decision was not based on any of that. We passed before any of this broke. We passed because of three small things in a sales conversation.

What we actually learned

This is the part that matters more than any single vendor.

A vendor who knows their stuff will peel the onion. They will say “that is a great question, and honestly it depends on your environment, let me get our security lead on a call next Tuesday to walk you through the specifics.” They will name a person. They will scope the answer honestly. They will acknowledge what their product does not do as confidently as what it does.

A vendor who does not will smooth over. They will give you a confident answer that sounds thorough and is not. They will defer to the product instead of to a person. They will sell you the outcome without showing you the mechanism.

Both patterns are legible once you are looking for them. The problem is that the first pattern feels slower in the meeting, and the second pattern feels like momentum. If your procurement process rewards momentum, you will keep picking the second one. And some of them will be fine. And some of them will not.

Questions worth asking your AI vendor

If a vendor cannot or will not engage with questions like these, that is your answer. It is not always a sign that the company is doing something wrong. It might just be a sign that they are early, or that they have optimized sales over depth, or that the people with real technical answers are not the people who talk to prospects. But it is information. Take it seriously.

The broader point

We have written about this pattern before from a different angle. If you want the deeper argument, read why “human-in-the-loop” is often a lie, which unpacks what real oversight requires and where most implementations quietly fall short. The short version is that “the AI handles it” tends to be a structural claim, not a marketing one. It tells you where the humans are and where they are not. In a regulated category, that distinction becomes everything.

At SageScreen we build in a category adjacent to what Delve sells: AI that produces consequential decisions in a regulated part of someone else’s business. We are an AI interview platform, not a compliance platform, but the architectural pressure is identical. You either build so a human can see inside the decision, or you do not. You either name the person who is accountable, or you hide behind “the platform.”

Our answer is written into the architecture rather than the terms of service. The AI that conducts the interview is structurally separate from the evaluator that produces the score. The output is a readable transcript, not a rating. A human always makes the final call. We have laid that out in detail on our values page, and in why we are transparent in our AI, and in trust but verify: the new rule of modern screening.

We did not build it that way to look principled. We built it that way because the alternative, in our category, is the kind of platform a candidate can never challenge and a regulator can never reconstruct. That is not a product. That is exposure with a dashboard on top.

Architecture is philosophy made tangible. It tells you what the builders actually believe. When you sit across from an AI vendor in 2026 and ask them a hard question, the answer they give you is the architecture. Whether the product is good, whether the company is honest, whether the system will hold up when it matters: all of it shows up in how they answer, and in who they hand you off to.

Ask the hard questions. If the vendor will not peel the onion, you have your answer. Trust the sniff test. It is doing more work than you think.