You Can’t Underwrite a Feature Anymore

On April 22, Mosaic announced an $18 million Series A led by Radical Ventures. Good company. Deterministic deal modeling for private equity, used by firms like Warburg Pincus, CVC and Evercore. If you’ve ever spent a Saturday night as an associate chasing a broken link in an LBO, you know exactly why that product exists.

Four days later, on April 26, I sat down to build the same core into DEALITHIC’s Deal Engine.

The first commit, a rollback checkpoint, went in at 1:48 AM. Color-coded Excel, IRR decomposition and a value creation library landed at 12:48 PM. The LBO export with live formulas, the piece I’d labeled “Mosaic parity complete,” went in at 2:02 PM.

About twelve hours.

Then on October 20, a research bot I run flagged Keye, an AI due-diligence platform for PE. Keye raised a $5 million seed led by Sorenson Capital, with General Catalyst and Y Combinator in the round. Smart product. Source tracing, raw-data analytics, deterministic math, benchmarking against your own past deals.

I asked how it compared to our engine around 3 PM. Deterministic answers shipped at 3:38. Source tracing, where every number in the memo cites its CIM page and gets flagged as an estimate if the quote can’t be found, shipped at 5:07. Cohorts, NRR, customer concentration and revenue bridges from a raw CSV shipped at 5:58. Benchmarking, the feature we’d labeled “later,” shipped at 5:59.

About three hours. Everything is shipped to our codebase, with commit timestamps.

I’m not telling you this to dunk on anyone. Both teams built real products and both rounds made sense on the information their investors had. I’m telling you because if one operator with thirty years of code and twenty years of deal experience can rebuild the core of a funded product in an afternoon, then the core was never what the check was paying for. And a lot of term sheets still read like it was.

The old math

For twenty years the venture logic on software was simple. Building was expensive and slow. A good team that could ship a hard product had a head start measured in years, and the code itself was the moat. Hire twenty engineers, burn eighteen months, and anyone who wants to catch you has to do the same.

That math is dead.

I’ve been writing code for thirty years and working in finance and investment banking for twenty. The coding tools I use today don’t make me twice as fast. They change what “a lot of work” even means. The part that used to take a team a quarter, the plumbing, the CRUD, the parsers, the export formatting, now takes an evening. What’s left is the part that was always hard: knowing exactly what an IC memo needs to say, which number a partner is going to poke at, and where the model is lying to you.

That’s domain knowledge. It used to be the input to the product. Now it’s most of the product.

Where it gets honest: Lightfield

Here’s a harder case. Lightfield raised a $47 million Series A led by a16z in September. It’s an AI-native CRM that builds itself from your email and calls, and it’s very good. More than 5,000 companies have signed up since its November 2025 launch, and the founders have disclosed 400% net dollar retention.

QUARRY, our CRM, is not a better Lightfield. It’s a different animal. QUARRY has an AI voice dialer working the phones today, with do-not-call checks, calling-hours limits and voicemail detection built in. On Lightfield’s pricing page that dialer is “coming soon.” QUARRY plugs straight into a map of 93,000+ funds, family offices and angels, a deal engine, and the fundraising tools founders actually use to raise. It’s built for people raising and deploying capital, not for every B2B sales team on earth.

Lightfield is ahead where it counts for its market. The polish is better. It has SOC 2 Type II and HIPAA. It ships a changelog every week. And 400% NDR is not something you rebuild over a weekend. You earn it, customer by customer.

That’s the point. The features are rebuildable. The trust isn’t.

“Yeah, but they have…”

I’ve heard all of these already. Fair questions. Here are the answers.

“They have 5,000 customers.” They do, and that’s the real asset. Code got cheap; getting in front of customers didn’t. That’s exactly why it’s the thing to underwrite.

“400% net dollar retention.” Can’t fake it, won’t try. That metric is earned, and it’s worth more than any feature list.

“SOC 2, HIPAA, enterprise security.” That’s an audit, months and a check, not a code problem. It’s next on our list, and it’s a real moat that should be priced like one.

“The team is stacked.” It usually is. Talent is still a moat. The question is whether you’re paying for talent to rebuild commodity features or for talent that builds distribution and trust.

“Their data compounds.” Agreed, and that’s the best argument any of these companies has. It’s also the opposite of a feature argument.

“You copied the demo, not the company.” Correct. That’s my whole argument.

What this means if you write checks

If the core product can be rebuilt in days, you should underwrite three things harder than you did five years ago.

Distribution. Who already has the customer, and what does it cost them to get the next one?

Trust. Certifications, track record, the fact that a mega-fund lets you anywhere near its models. That takes years and it’s the part a weekend can’t touch.

Data. What does the company see that nobody else sees, and does it get better with every customer?

The feature list should be the last slide, not the first one. A demo that wows a partner on Tuesday can be rebuilt by Saturday.

So what’s DEALITHIC

I’m building DEALITHIC to be the operating infrastructure for private capital formation. Shape a company, build the raise, match it across a network of 93,000+ funds, family offices and angels, run it afterward. We’re early, and we’re moving faster than anyone in the category.

But the bet is the same one this essay makes. Features are cheap now, so we build them fast and spend our time on the parts that compound: the investor map, the deal data, and the trust of the people who use it.

We build in days what others raise rounds to build.

If you underwrite private markets software, ask me for the commit history. It’s more interesting than the deck.