The AI Adoption Data Layer

The pain point

AI tools are spreading faster than any software in history — ChatGPT became the fastest-growing consumer app on record, reaching 100M users in two months, well ahead of TikTok or Instagram (UBS, via Reuters) — but almost nobody can credibly measure whether they work. Because these tools hallucinate, their output cannot be trusted at face value, and there is no neutral, trackable record of who is adopting AI well and actually producing results versus who is just paying for subscriptions. Efficacy is asserted, not evidenced.

That is the gap: the market is pouring dollars and hours into AI, and the layer that would tell you which adoption is real and producing value does not yet exist. Whoever captures that signal — credibly, longitudinally, at the level of the individual builder and the team — owns the missing measurement layer of the AI economy.

The space, and the companies in it

A handful of companies are already building at this adoption layer, and they fall into two rough categories: community-based tools that capture who shows up, builds, and ships — like Build Club and BuilderBase — and learning-based tools that capture who is trained and what they actually retain and apply — like Honen. Each one sits on real adoption data as a byproduct of what it already does.

The companies, in detail:

• Honen — automated teaching + learning infrastructure for any company; turns team knowledge into AI-generated courses across 200+ industries. Built by the StudyFetch team (Esan Durrani, Forbes 30u30; Ryan Trattner), which raised ~$10M (2025, Owl Ventures). Captures who completes and applies AI training inside companies — a direct read on corporate adoption efficacy.
• Build Club — APAC AI builder community (project-based learning, accelerators, residencies). $1.75M pre-seed (Oct 2024, co-led by Airtree + Blackbird; founder Annie Liao). Captures who shows up and ships — community-level adoption and follow-through.
• BuilderBase — end-to-end hackathon and innovation-program infrastructure: builder profiles, applicant screening, judging, sponsor marketplace. Captures builder identity and what people actually built — the profile-as-credential layer.

Who, when, where, why

The pain is concrete. Two illustrative profiles and the moments the need surfaces:

• The enterprise buyer. A 200-person company runs an AI-upskilling program and pays for a dozen tools. It cannot tell which teams are getting real lift versus generating plausible-but-wrong output. It needs a credible record of adoption efficacy — and that need surfaces at budget renewal, when the CFO asks "what did the AI spend actually return?"
• The individual builder. A 20-year-old ships real projects with Cursor + Claude + Replicate — but nothing trustworthy records that skill, so no employer, school, or lender can act on it. It matters most when they graduate and job-hunt, or leave a salaried job to go independent and suddenly can't prove their income or ability.

Where: the capture nodes themselves are the channel — upskilling platforms, builder communities, hackathon networks, university CS departments. Why this, not another dashboard: because the signal only becomes credible when it is captured at the point of real work and tracked over time, which is exactly what these companies already do.

Where the space grows: financing & credit

Once you can credibly measure who adopts AI and produces results, you can underwrite them. That is the natural compounding path of the space — from measurement to money:

• Credentialing first. A trusted record of demonstrated AI capability — valuable to employers hiring AI-native talent.
• Credit and capital next. The same record becomes an alt-data underwriting signal: a card, a credit line, or capital for the student-and-builder cohort, priced on AI engagement rather than a thin FICO file. Consumer banking for the cohort exists (Fizz, SoFi, Greenlight) but is not adoption-aware; AI-spend management exists (Cledara, Zylo) but is corporate-only. The intersection is empty.

This is why the space is a good place to be early: the upstream companies capture the data as a byproduct of what they already do, and the downstream financing layer — the part that monetizes it — has no occupant yet.

Why DCM specifically

The move from capture to financing runs through the sponsor-bank ecosystem — and that is exactly where DCM's lending portfolio already lives.

• A two-decade pattern of data → lending, led by one partner. Co-founder David Chao led DCM into Bill.com (its earliest and largest backer — $26.4M became a stake worth ~$900M), SoFi (board seat from the 2012 Series B), and Figure (co-led the 2018 first round with Ribbit). DCM backed founder Mike Cagney twice — SoFi, then Figure. The repeated shape is alt-data underwriting → consumer and SMB lending.
• Fluency in the sponsor-bank layer. DCM's lending companies plug directly into bank rails: Cherry is sponsor-banked by Cross River (a $50M facility, loans originated through Cross River); Figure runs lending through 175+ partner banks; SoFi became a chartered bank. A data-to-credit company must plug into exactly this layer, and DCM's book shows it understands it.
• The financing entity is likely separate. Realistically the lending product is a distinct, sponsor-bank-backed company that buys the credible adoption signal from the capture nodes. DCM is positioned to assemble that: back a capture company, back the financing company, and sit across the bank relationship the two depend on.
• How DCM itself frames the edge. Writing on SoFi's IPO, DCM described the win as the ability to "discover a new way to assess risk that not even the traditional banks could see" — serving creditworthy people the conventional model misses. AI-adoption signal is the next version of that.

Why now

• The credibility gap just became expensive. The 95% failure number means enterprises are actively looking for a way to measure AI efficacy — the demand for the capture layer is here, not hypothetical.
• The cohort hit scale. ChatGPT weekly actives went from ~400M (Feb 2025) to ~900M (Feb 2026); the under-25 cohort is the heaviest per-capita AI adopter. The population to measure and serve is observable today.
• Alt-data underwriting is standard. Affirm, Karat, and SoFi (via Nova Credit) proved you can underwrite non-traditional populations on alternative signals — so the financing layer is buildable once the data exists.
• Issuance is a commodity. Lithic and Stripe Issuing collapse a sponsor-bank-backed pilot to roughly $200–500K and weeks, not years.

Companies already underwriting on data

Underwriting on alternative data rather than FICO is a proven playbook — these companies built real businesses by picking a coherent group and lending against a signal incumbents couldn't see. That track record is part of why this space looks like a good opportunity: the model works; it just hasn't been pointed at AI adoption yet.

Falsifiability

This thesis is wrong if, by mid-2027: (1) credible AI-adoption data turns out not to predict downstream hiring, income, or company outcomes — i.e., it is noise, not signal; (2) the measurement gap closes on its own — enterprises learn to track AI ROI internally and the 95% failure number falls fast, so no dedicated capture or credibility layer is needed; (3) an upstream platform closes the loop itself and forecloses an independent financing layer; (4) the capture nodes refuse to share data, leaving any financing entity renting signal with no moat; (5) card/credit economics compress below the cohort's viability; or (6) consumer-credit regulation tightens on alt-data underwriting past what the unit economics support.