Products
The Model Platform MVP Pricing Blog Request demo
A Foundation Model for Organizational Intelligence

The Large Behavior Model knows
who actually owns the work. who has capacity right now. which document is the truth. when to escalate to a human.

The Large Behavior Model is pretrained on cross-company behavior, so it can read a company it has never seen. It finds actual owners, open capacity, source-of-truth documents, and hidden influence.

Connect metadata only. Try the model on your own org over MCP first, or book a call with us directly.

Get a demo Read the model overview

Integrates with your existing stack

Google
Microsoft 365
Slack
Jira
Salesforce
Box
ServiceNow
Workday
SAPSAP

The org chart tells you who reports to whom. Documents tell you what was written. Neither tells you how work actually moves.

The formal organization
The org chart, the titles, the official policy, the canonical document. This is the organization your systems can already see.
The behavioral organization
Who people actually go to, who they rely on, the document they actually open. This is where work really happens, and it is invisible to your AI.
The gap is the signal
The two usually agree. Where they diverge is where the useful information is: hidden owners, real escalation paths, forming risk. The model measures that gap.
Why it matters

Two reasons this is the layer you are missing.

Every company has two versions: the org chart, and the operating network where work actually moves. Here is why that second one decides everything.

For strategy

You found the path. Can your company walk it?

Every leadership team hunts for the one move that wins, the Doctor Strange path through millions of futures. Finding it is not the hard part. Walking it is. A strategy only becomes real when thousands of people carry it through the company, and two invisible things decide whether they can.

Is your best talent even connected, and does change travel through the people others actually trust? Most leaders can name only a few of those people. They cannot name enough.

BehaviorGraph surfaces the real influence network, the hidden experts, and the missing bridges, and turns your strategy into an execution map: who to involve, who to align first, where it will stall. You found the path. This is how you walk it.

For your AI

Your AI is blind to how your company actually runs.

You are putting AI everywhere because the alternative is falling behind, including to competitors you may not even see yet. But every enterprise AI effort hits the same wall.

Your real operating knowledge is not in one system. It lives in thousands of people's heads: who owns the workaround, which document people trust, who decides when the process breaks. An agent pointed at your company is brilliant and blind at once.

BehaviorGraph turns that operating reality into a map your agents can query, so they route to the right person, escalate to the right authority, and defer when a human must decide. It needs to know how your organization works.

How it works

Already trained to read how work moves.

The Large Behavior Model is an AI model pretrained across companies to read how work actually moves. You choose what metadata to share, even calendar data to start, and it can answer the same questions from different approved sources.

What teams use it for

Start with one of these four.

🔀
Agent Routing

Your agents ask who actually owns the work, who has capacity right now, and where to escalate, before they assign or approve.

Use case →
🔍
Enterprise Search Enrichment

Every result carries the real owner, a qualified backup, and the document people actually treat as the truth, not just the officially endorsed one.

Use case →
🌱
Change Management Insights

Who actually drives or blocks a rollout, which teams are adopting, and where the initiative is quietly stalling, read from behavior instead of status reports.

See the walkthrough →
🛡️
Enterprise Risk Management

Shadow authority and collusion patterns read from structure alone, plus attrition risk months before HR sees it. On the Enron corpus: 2.12x over chance.

See the evidence →
Validation results

Validated against real outcomes.

Every result is tested against simpler methods and checked to make sure the answer does not come from accidental clues in the data.

Case study 1: Enron, the test the model cannot game.

About 517,000 emails across roughly 170 mapped employees over three and a half years (1999 to 2002), where who held power and who was prosecuted is court record. Reading communication patterns alone, without being told who was later charged or convicted, the model:

  • Ranks the indicted insiders at more than twice the rate chance would produce.
  • Detects the trading desk hardening into an insular ring through the California energy crisis, then dissolving afterward, a collusion pattern visible in structure before it was public.
  • Surfaces internal code names tied to the fraud in internal subject lines about a year before public exposure, clustered around the employees later convicted.

Enron's email record is older and less complete than a modern digital workplace, which leaves a far richer signal. Court records make these results independently checkable. For risk, compliance, and diligence teams, this is an early warning system that reads collusion and shadow authority from structure alone, years before an investigation.

Case study 2: a real 900-person tech company.

Tested on a real technology company of about 900 people the model had never trained on: fast-moving, socially driven, project-based, under heavy delivery pressure. It read only the company's permissioned calendars and Slack. From those two signals alone, the model rebuilt the company's operating map:

  • About 70 go-to experts the company truly relies on, plus about 100 hidden experts, genuinely relied on, with titles that understate their real role, that a title chart alone would miss.
  • About 35 key-person risks whose exit would break a live workflow, concentrated in senior specialists rather than junior staff.
  • Backup depth: about 70% of people have a peer who could step in, but about 25 critical roles have no real backup, and roughly a quarter of the priority projects depend on a single person.
  • About 85 cross-team bridges holding otherwise separate groups together, who also carried the highest overload and burnout risk, and about 7 clusters at risk of becoming isolated silos.

Two checks compared the model's answers with facts it was never shown: job function, identified from work patterns alone, and departures, predicted from past calendar behavior and confirmed against independent external records. No surveys, no documents, no answer sheets handed in.

2.12xbetter than chance at finding hidden influence in the Enron email data
r ≈ 0.50applied to a company it never trained on, the model's reads still hold
0.94 AUCpredicts likely departures months ahead, on a scale where 0.90+ is excellent

Full validation results → the model brief

FAQ

Common questions

What is the Large Behavior Model?

A foundation model for organizational intelligence. Every company has two organizations: the formal one (the org chart, titles, official policy) and the behavioral one (who people actually go to, who they rely on, the document they actually open). The model reads the behavioral organization from ordinary work signals and measures the gap between the two. Where they diverge is where the useful information is.

Does BehaviorGraph read emails or message content?

No. BehaviorGraph operates on metadata only: who meets whom, how quickly people reply, and how tickets and documents are used, plus optional short surveys. It never reads message content, emails, or document text. People appear as anonymized codes. Privacy by design.

How do we start?

With the free evaluation. Connect calendar metadata (identities hashed), and the model returns one full diagnosis of your organization over MCP. Judge the results on your own org, then decide. No setup, no dashboards required.

How is it priced?

Two products. The model is an annual license plus usage, consumed over MCP and the REST API. The Platform MVP is priced per dashboard seat on top of a model license. Everything starts with the free evaluation. Details on the pricing page.

Is the model validated?

Yes. On the Enron email data the model finds hidden influence at 2.12x the chance rate; applied to a company it never trained on, its reads still hold (r about 0.50); and in controlled testing it flags likely departures months ahead at 0.94 AUC. Validation uses simulated companies with known right answers, real companies excluded from training, and outside records, with full methods in the model overview.

Do we need ONA or a behavior graph already?

No. The model builds a first map of how work moves from one or two data sources you already run, such as calendar and messaging metadata. More sources, and existing network analysis or peer feedback where available, make the results more precise, but they are not required to begin.

What else can the model be used for?

Teams usually start with one of four: agent routing, enterprise search enrichment, change management insights, or enterprise risk management. The same model also reads knowledge source-of-truth, succession and backfill cover, bottlenecks, and integration risk for M&A diligence.

How is this different from organizational network analysis tools?

Traditional ONA tools use surveys to show who is connected to whom at one point in time. The Large Behavior Model starts from signals that are always flowing, so it does not wait on a survey cycle; where surveys and peer feedback exist, it uses them to calibrate and validate. It tracks change over time, supports live routing, compares the org chart with how work actually happens, and uses one model across all use cases.

Request demo

Your AI already reads the docs. Give it the org.

The Large Behavior Model works alongside your existing enterprise search, agent platform, or governance stack. If your AI is accurate on content but still misroutes decisions, stalls on escalation, or misses the right person, we should talk.

Email hi@lead.app Read the blog