Teahose.
SIGN IN
NEW HERE — WHAT TEAHOSE DOES
We read the entire AI & tech firehose — so you don't have to.
PODPodcastsAll-In, No Priors, Acquired…
NEWNewslettersStratechery, Newcomer…
PAPPapersPhysical AI research
PHProduct Huntdaily launches
VCInvestor ScoutSequoia, a16z, Benchmark…
CLAUDE DISTILLS →
7 reads, 30 sec each — free, 6 AM ET.
+ a live graph of the companies, people & themes underneath.
HOME/DATA DRIVEN VC/🔥How to Gain an Edge When Every…
NEWS
// NEWSLETTER ISSUE
DATA DRIVEN VC

🔥How to Gain an Edge When Everyone Uses the Same AI Model

DATE July 12, 2026SOURCE DATA DRIVEN VCPARTICIPANTS ANDRE RETTERATH
// SUMMARY

1. Key Themes


Theme 1: Proprietary Data as the New Moat in AI-Enabled VC

As AI models become universally accessible, the competitive advantage for investment firms shifts entirely to the uniqueness and quality of the data fed into those models. The panel's core conclusion is explicit:

"The company data most funds compete over is becoming a commodity. The real edge sits in the data only your own team can generate."


Theme 2: Data Architecture Discipline — Separating Four Distinct Layers

Most funds treat all data as interchangeable, which corrupts downstream analysis. The article argues that maintaining strict separation between data types is foundational to trustworthy AI outputs:

"The four layers of data most funds collapse into one, and why keeping them separate changes what you can trust downstream."


Theme 3: Metadata Context Is Required for Metrics to Be Meaningful

Raw numbers without provenance are worthless for investment decision-making. The article singles out a specific, concrete example of this problem:

"Why a metric like ARR is meaningless without the value, source, and date attached to it."


Theme 4: Untapped Internal Knowledge Assets Inside VC Firms

Funds are sitting on rich, proprietary institutional knowledge that is systematically ignored — representing both an operational failure and a competitive opportunity:

"The internal data asset most funds already have and never use: old investment memos and pass reasons."


Theme 5: AI Consistency as an Infrastructure Problem

AI unreliability — giving different answers to the same question — is framed not as a model problem but as a data and scoring infrastructure problem that must be engineered away:

"A 500+ characteristic scoring framework built to stop AI from giving a different answer to the same question twice."


2. Contrarian Perspectives


Perspective 1: External Company Data Is No Longer a Differentiator

The prevailing assumption in the VC industry is that access to better company data (e.g., revenue signals, web traffic, hiring data) creates an edge. The panel directly challenges this, arguing the commoditization of AI access has made such data equally available to all:

"Every fund today can plug into the same AI models. The question the panel set out to answer is what still separates one investment firm from another once that's true." "The company data most funds compete over is becoming a commodity."

The implication: spending on third-party data providers may be increasingly low-ROI relative to investing in proprietary internal data infrastructure.


Perspective 2: True Comparable Analysis Requires a "Time Machine," Not Just Benchmarks

Standard benchmarking compares companies at different stages as if they exist at the same point in time, which the panel argues produces flawed comparisons. The solution is temporal normalization:

"How Kruncher's 'time machine' makes true apples-to-apples comparison possible across funding stages."

This is a non-obvious architectural insight: the industry standard of stage-based benchmarking is structurally flawed without time-normalization of the underlying data.


Perspective 3: Data Source Trust Must Be Tiered Before Integration, Not After

Most firms integrate data sources opportunistically and then try to quality-control outputs. The panel argues this is backwards — a trust hierarchy must precede any schema or taxonomy design:

"The three-tier trust hierarchy to decide which data sources are even worth integrating." "The questions you should ask before writing a single line of schema or taxonomy."


3. Companies Identified

CompanyDescriptionWhy MentionedQuote
Kruncher AIAI platform for private market investorsCore case study; builds the "compounding data layer" and knowledge infrastructure for VC firms"Francesco de Liva is Founder & CEO of Kruncher, bringing 18+ years as a technical architect at Microsoft and Accenture to building the knowledge layer for private market investors."
Eight RoadsFidelity-backed global VC firm with $5B+ AUMCase study in building data infrastructure and AI strategy at scale within an established fund"Harman Bahd is Head of Data and AI at Eight Roads, the Fidelity-backed global VC firm with $5B+ in AUM, where he leads data infrastructure and AI strategy across the firm."
Offline VenturesVenture firm with operating partner modelCase study in fund and portfolio operations, from investment execution through LP reporting"Meredith Parsons is an Operating Partner at Offline Ventures, leading fund and portfolio operations from investment execution through LP reporting."
Wispr FlowVoice-to-text productivity toolSponsored product; converts speech to clean professional text inside any app"Wispr Flow turns your voice into clean, professional text inside any app. Emails, Slack, client updates — speak once, send without editing. 4x faster than typing."

4. People Identified

PersonDescriptionWhy MentionedQuote
Francesco de LivaFounder & CEO, Kruncher AILed roundtable as practitioner building proprietary data infrastructure for private markets; 18+ years at Microsoft and Accenture"Francesco de Liva is Founder & CEO of Kruncher, bringing 18+ years as a technical architect at Microsoft and Accenture to building the knowledge layer for private market investors."
Meredith ParsonsOperating Partner, Offline VenturesRoundtable contributor on fund and portfolio operations, LP reporting, and investment execution"Meredith Parsons is an Operating Partner at Offline Ventures, leading fund and portfolio operations from investment execution through LP reporting."
Harman BahdHead of Data & AI, Eight RoadsRoundtable contributor representing an institutionalized, large-scale approach to AI strategy within a $5B+ AUM fund"Harman Bahd is Head of Data and AI at Eight Roads, the Fidelity-backed global VC firm with $5B+ in AUM, where he leads data infrastructure and AI strategy across the firm."
Andre RetterathAuthor, Data Driven VC newsletterNewsletter author and convener of the roundtable"Hi, I'm Andre and welcome to my newsletter Data Driven VC which is all about becoming a better investor with data and AI."

5. Operating Insights


Insight 1: Mine Your Historical Memos Before Buying More Data

The single most actionable takeaway for operators and investors is that high-value proprietary signal is already in-house — specifically in the institutional memory captured in old investment memos and pass decisions. This is a zero-cost, high-differentiation starting point for building a proprietary data layer:

"The internal data asset most funds already have and never use: old investment memos and pass reasons."


Insight 2: Design Your Data Taxonomy Before Your Technology Stack

The article warns against a common mistake: building schemas and integrations before resolving foundational data governance questions. The right sequence is trust hierarchy → taxonomy → schema → tooling:

"The questions you should ask before writing a single line of schema or taxonomy" and "The three-tier trust hierarchy to decide which data sources are even worth integrating."


Insight 3: Attach Provenance to Every Data Point

For AI outputs to be reliable and auditable, every metric ingested must carry its context. This is an operational standard that needs to be enforced at the point of data entry, not retroactively:

"Why a metric like ARR is meaningless without the value, source, and date attached to it."


6. Overlooked Insights


Insight 1: The 500+ Characteristic Scoring Framework as an AI Consistency Mechanism

Most AI-in-VC discussions focus on what models to use or what data to feed them. Buried in the article is a more fundamental engineering point: without a structured scoring rubric of sufficient granularity, AI systems produce non-deterministic outputs that cannot be trusted for repeatable investment decisions. The 500+ characteristic framework is presented as the solution to this, but it receives only a single line of attention:

"A 500+ characteristic scoring framework built to stop AI from giving a different answer to the same question twice."

This is a significant infrastructure investment signal — suggesting that AI reliability in VC requires far more structured scaffolding than most firms currently build.


Insight 2: The DDVC Landscape Report Shows 345 VC Firms Already Operationalizing AI

The article briefly references a just-launched report that captures the current state of AI adoption across the VC industry. The scale — 345 firms surveyed — suggests AI adoption in VC has crossed from early-adopter into mainstream territory, which raises the urgency of differentiation through proprietary data layers:

"The DDVC Landscape Report 2026 just launched! Check out how 345 VC firms are using AI and automation to become more efficient & win against their peers."