Every voice in the room. One decision the evidence can defend.
Consensus Intelligence reads the meeting, has independent AI reviewers score every option, stress-tests each one across 2,000 simulated futures, and tells your team, neutrally, whether the evidence backs what the room prefers.
Runs inside your own Claude workspace. No servers to install, no API keys, and your meetings never leave your account.
How a decision moves through the engine
Five stages, from the first word spoken to a decision record you can hand to an auditor.
- 1
Capture
Fireflies recordings, live notes or a pasted transcript. Speakers are identified and personal data is removed before analysis.
- 2
Evaluate
Four independent AI reviewers score every option for finance, operations, risk and people, each with its confidence.
- 3
Stress-test
2,000 simulated futures, budget, delay and risk shocks, and three strategy lenses show what holds up.
- 4
Converge
Participants, including those who missed the meeting, add positions and counters. An arbiter rules on each counter.
- 5
Record
A decision brief with the full rationale, signed with a SHA-256 audit trail, published when the room closes.
Agreement is not evidence.
When a room agrees quickly, nobody checks whether it agreed on the right thing. The independent check compares the team's position with the evidence and says plainly when they differ, and why.
- Headcount never moves a scoreVotes are shown beside the evidence, never inside it.
- Unproductive agreement is flaggedA unanimous choice that fails its topic checks is called out, with the gaps listed.
- What would have to be trueFor a weaker favourite, the engine states how much better its numbers would need to be to lead.
The team favours Option B. The evidence prefers Option A.
Diverged- A is ahead on downside protection: pessimistic NPV ₹1.1 Cr against −₹0.5 Cr.
- B is faster to value (8 weeks against 26), which may be what the team is weighing.
- For B to lead, its annual benefit would need to be about 38% higher than estimated.
Specifications
What the engine does with every decision, stated plainly.
| Topic classification | Investment, sourcing, technology, product, people, operations, safety and compliance, strategy, budget; weights and required views adapt to each |
|---|---|
| Independent reviewers | 4 AI reviewers (finance, operations, risk, people), scored separately |
| Simulation | 2,000-run Monte Carlo per option, with P10, P50 and P90 outcomes and the chance of loss |
| Stress tests | Budget cut, delay, cost overrun, adoption shortfall and risk shock, plus 3 strategy lenses |
| Topic checks | 2 to 5 decision-critical checks set from the subject itself, including ones the room overlooked |
| Participation | Live capture, offline inputs before the room closes, counters ruled on by an AI arbiter |
| Automation | Hourly Fireflies autopilot opens rooms and publishes the decision brief at close |
| Audit | SHA-256 trail for every analysis; reports and a structured JSON export |
| Data residency | Your own Claude account and connectors; the publisher never receives meeting content |
Engineered for teams that answer for their decisions
Your data stays yours
The engine runs in your Claude workspace. Meeting content, positions and decisions are stored there, under your access rules.
Neutral by construction
Same inputs, same result, every time. Scores come from estimates and simulation, never from who spoke loudest.
Ready for audit
Every brief carries its reasoning, the backup data and a hash of the analysis, so the decision can be reviewed years later.
Run your next decision through it.
Thirty days free. Setup takes about five minutes inside Claude.