The problem

PR teams are flying blind.

Agencies spend 3 days drafting crisis responses by instinct. There is no way to test a statement before publishing it. One wrong phrase can cost billions — United Airlines lost $1.4 billion in market cap in a single afternoon.

3 days

Average time to draft a crisis response

0

Ways to test before publishing

$1.4B

Lost by United Airlines in one afternoon

The solution

Simulate the crowd before the crowd reacts.

Crowdglass runs a synthetic social network — 88 AI stakeholders with distinct personas, influence tiers, and emotional states. They debate, amplify, and react. The platform finds the response that works before you publish.

How it works

Three steps to a tested response

01

Describe the situation

Paste crisis context. Crowdglass seeds the simulation.

02

Test your response

88 AI stakeholders react across a simulated social network.

03

Get the briefing

Download a PDF your team can present in 30 minutes.

Validation

5 real-world crises backtested

Each crisis was simulated and compared against documented real-world outcomes. The match percentage shows how closely the simulation’s dominant narratives aligned with what actually happened.

Honesty section

Three kill criteria

Here’s what would make us stop. If any of these are true after our validation phase, we wind down the company and return remaining capital.

Simulation consensus diverges from reality

If backtested simulation outcomes diverge more than 40% from real-world crisis trajectories across our 5 validated scenarios, the model is not reliable enough to ship.

Expert panels flag outputs as unrealistic

If crisis communication professionals consistently rate agent behavior as formulaic, implausible, or missing key stakeholder dynamics, the simulation lacks fidelity.

No improvement over instinct

If blind A/B evaluations show no measurable improvement in response quality compared to instinct-based drafting, the product does not justify its cost.

The ask

$250K – $500K pre-seed

18 months of runway to validate the model, sign 3 pilot clients, and build the infrastructure for scale.

Use of fundsAmount
LLM compute + fine-tuning$80-120K
Agent calibration research$50-80K
Go-to-market (3 pilot clients)$40-60K
Team (2 engineers + 1 researcher)$60-120K
18-month operating runway$20-120K

The team

Built by

Founder

Background in crisis communications, computational social science, and product engineering. More details coming soon.

hello@crowdglass.com