WE TURN HUMAN DATA INTO SYNTHETIC SOCIETIES TO TEST DECISIONS BEFORE EXECUTION.
Video · 5 min
Granovetter in five minutes.
From the problem to the simulation: how the method turns a decision into testable scenarios before it reaches people.
Narrated in Portuguese.
A method grounded in research published in
- American Journal of Sociology
- Nature
- Political Analysis
- British Journal of Psychology
- ICML 2025
- Stanford · arXiv
- WTR 2026
- talk “IA que Trabalha”
- 1973 · 1978
- the two papers by Mark Granovetter
- Open source
- auditable simulation engine
Every decision sets off reactions in a human system before it becomes a result.
Between intention and execution there are people. There is behavior.
Granovetter simulates those behaviors to anticipate outcomes, showing which scenarios are most likely to deliver the best return and which risks could compromise time, money and trust.
Simulate critical scenarios before a decision costs too much.
Granovetter lowers the cost of deciding wrong.
We created the grano.protocol, a behavioral simulation method built on artificial intelligence and applied science to turn strategic decisions into testable scenarios.
Simulating behavior is only the means. The end is showing which paths produce the best outcome before the decision burns time, money and trust.
Decision risk
Which path is likely to work best before you implement it.
Buy-in risk
Who is likely to buy in, resist, distort or ignore the decision.
Execution risk
What could compromise productivity, focus, collaboration and speed.
Cultural risk
Whether the decision strengthens or weakens trust, climate, belonging and openness to change.
Reputational risk
Noise, cynicism, parallel narratives and erosion of leadership.
Input→Simulation→Output →Intervention
Granovetter’s method, grano.protocol, turns strategic decisions into behavioral simulations: it takes in real data from the organization, builds synthetic agents for each group and delivers inferences to support the decision before execution.
Input
Cultural data, group profiles, organizational context, external environment and the decision to be tested.
Simulation
Synthetic agents interact with scenarios, pressures, narratives and strategic variables.
Output
Likely scenarios, emerging risks, hidden tensions, patterns of buy-in and resistance, and recommendations for the decision.
Intervention
Actions tested before acting: communication, incentives and pace of change are simulated again to see which one reduces resistance.
The engine is open source.
Each synthetic agent represents a real group in the organization, with its own values, risk tolerance and narrative. In round 1, each agent reacts alone. From round 2 on, it is exposed to the group consensus and to opposing views. The analysis measures adoption, resistance, conformism and anxiety, and locates the tipping point.
Simulation is not prediction. It is decision support, and should be validated against the organization's own data.
Watch a tipping point happen.
Each person has a threshold: how many colleagues must join before they join. Change the thresholds and see whether the change spreads or stalls.
Granovetter’s teaching model (1978), with 100 people. It is not the Granovetter engine, which uses AI agents with real profiles.
People who decide about people.
Board and executives
Restructurings, mergers, changes in business model and leadership.
HR and People
Work model, policies, benefits, culture and climate.
Sales
Commission plans, targets, territories and sales incentives.
Internal communication
How, when and to whom to announce a change, with the right message for each group.
Where Granovetter fits.
| Granovetter | Engagement survey | Traditional consulting | |
|---|---|---|---|
| Tests the reaction before the decision is announced | Yes | Partly | Partly |
| Keeps the decision private while it is tested | Yes | No | Partly |
| Shows how one group influences another | Yes | No | Partly |
| Compares interventions side by side | Yes | No | Partly |
| Method and engine open to audit | Yes | No | No |
| Hears real people directly | No | Yes | Yes |
YesPartlyNo
Simulation does not replace listening to people. The best result comes from combining all three: simulate first, then validate with real data.
AI Behavioral Science is the convergence of behavioral science, complex systems and generative artificial intelligence.
My name is Dheiver Santos, PhD.
- 200+
- papers
- 550+
- citations
- 12
- h-index
- 1
- INPI patent
- 10
- AI products
I am the founder of Granovetter.
Machine learning engineer and AI researcher, with a PhD in Chemical Engineering with an emphasis on Artificial Intelligence (UFBA/UFRJ) and postdoctoral research at UNICAMP/UNIT (CNPq) and UFAL (FUNDEPES). My career was shaped in senior industry roles, as Senior Staff Machine Learning Engineer at SX Negócios and Senior Scientist at Grupo Boticário, and in the classroom, teaching AI, distributed systems and data science at UPE and Estácio.
I created Mangaba.AI, an open-source framework for teams of autonomous AI agents, and founded Cognai, a healthtech applying AI to cardiology.
My academic work spans machine learning and neuro-symbolic computing, with AI products delivered in real clinical and operational workflows.
From open source to ongoing support.
Reference pricing. The final proposal depends on the number of groups involved and the depth of the analysis.
Open source
The simulation engine, open, to run on your own.
For technical teams who want to try the method.
- Full simulation engine
- Example scenarios
- Method documentation and bibliography
Diagnosis
One decision, tested before it reaches the table.
To validate a single decision.
- Mapping of the groups involved
- One simulated decision
- Report on scenarios, risks and the tipping point
SimulationRecommended
The decision and the ways to carry it out, compared.
To choose how to execute, not just whether.
- Everything in Diagnosis
- Intervention testing: messaging, incentives and pace
- Validation with the organization’s real data
- Results walkthrough with leadership
Ongoing
Recurring simulations while the change unfolds.
For long, multi-stage changes.
- Everything in Simulation
- New rounds at each stage of the change
- Recalibration with what actually happened
- Messaging tuned per group
Frequently asked questions.
What is a synthetic society?
A set of AI agents, each representing a real group in the organization, with its own values, risk tolerance and narrative. The agents react to the decision and then to each other.
Does simulation replace research with people?
No. Simulation is not prediction: it is decision support. It helps anticipate patterns, tensions and risks, and should be validated against the organization’s own data.
What kind of decision can be simulated?
Changes whose outcome depends on how people react: work model, pay and commission policies, restructurings, mergers and how each of them is communicated.
What data does Granovetter need?
Profiles of the groups involved, cultural data, the organizational and external context, and the decision to be tested.
What do I get at the end?
Likely scenarios, emerging risks, buy-in and resistance patterns by group, the tipping point and recommendations, plus testing interventions before acting.
Is the engine open?
Yes. The simulation engine is open source, and the research behind the method is on the research page.
A tool for deciding better.
We are a small team working at the frontier of multi-agent simulation, behavioral modeling and applied AI to solve the biggest problem in running a company: the lack of predictability.
