James Bell
Product & GTMBuilt DoorDash's personalization platform, driving $1B in incremental annual GMV. Led product for DoorDash's AI data assistant.
Previously: Apple, Yelp, DoorDash
Company intelligence
The connective tissue of your business.
People and agents ask Stroma about the business from the tools they already use. Stroma gets the answer right from your data — and remembers what people learn along the way.
Monday, 8:04 — “Why are first orders down in the West?”
395▼ 11.3%first orders · West · last week
Signups rose 8.8%, but fewer of them ordered — the referral relaunch bought volume at lower intent.
ConfirmedMetric graph · checked against the order tablePhoenix on-time fell to 59% after the Aug 24 routing change; promo first orders sat in those late windows.
EmergingDispatch note from Dana · ops ticketA driver in Phoenix says the new zones double back across the I-10.
UnconfirmedMeeting note · one sourcePause the West first-order promo in Phoenix until on-time recovers; keep it running elsewhere.Projected: +38 first orders/wk · basis: last 4 weeks
Needs approval
The problem
As companies scale, knowledge fragments.
What a small team once knew together gets scattered across people, data, conversations, experiments and systems. Teams redo investigations, miss results that should change decisions, and rely on whoever happens to remember.
People become the connective tissue.
Friday
The deck goes out.
Weekly numbers, carefully built, sent to leadership.
Monday
The numbers moved.
Someone asks why. You message an analyst and wait — again.
Thursday
An answer, maybe.
The reason was in a thread someone half-remembered, and the next deck starts from zero.
Why now
Model intelligence is abundant. Company intelligence isn't.
Companies are rewriting how their business works for AI — in context files, semantic layers, instructions and rules. When people are missing context, they ask someone. When agents are missing it, they guess and keep going.
Now agents need the knowledge people already struggle to connect.
How it works
Two connected layers, alongside the tools you already use.
Stroma turns company understanding into infrastructure. People and agents keep working where they work today.
SlackClaude & MCPTableauYour inboxThe Stroma app
Company memory layer
Learns from how people and agents investigate, correct and act on the data. Connections strengthen or weaken as people confirm, contest and correct.
Deterministic data layer
A reliable, shared understanding of the company's data — your metrics, your definitions, your warehouse.
Why we win
Every approach today stops short.
| Today's answer | Where it stops | Stroma |
|---|---|---|
| More context | Assumes the agent knows what matters | Learns what matters from how people use data and build evidence |
| Semantic layers | Deterministic, but only for what you can afford to model | Builds on what you've modeled and learns what hasn't been |
| Better models | Better reasoning still extrapolates what the business hasn't captured | Supplies the business knowledge models can't infer |
| Human tuning | Every new agent must be taught separately and maintained | Learns once, then shares it with the agents you connect |
Stroma benefits from advances in the current stack. We tried each of these approaches at DoorDash. None was enough.
Expansion
Land on one outcome. Compound from there.
Land
Start with one measurable outcome. No company-wide data or semantic-layer initiative required.
Expand
Add adjacent workflows, teams and agents without rebuilding company understanding from scratch.
Embed
Company understanding becomes shared infrastructure for people and agents.
Team
We spent years solving this problem at DoorDash scale.
Ram Karunanidhi
EngineeringBuilt DoorDash's merchandising platform across 2.3M stores, driving $65M+ in incremental GMV.
Previously: Zillow, Airbnb, Lyft, DoorDash
Company intelligence is the next layer of the stack.
We're working with design partners and early investors who believe it too. Tell us the outcome you'd start with.