The Best Way to Explain It Is to Show It
Most explanations of multi-agent simulation get lost in academic language. "Emergent behavior in complex adaptive systems" doesn't help a CMO decide whether to use this tool.
So here's the alternative: we'll walk through an actual simulation, step by step, using our Zomato fee hike case study. Every quote, every insight, every output is real.
The Setup: What We're Simulating
On March 24, 2026, Zomato raised its platform fee from Rs. 12.50 to Rs. 14.90. We wanted to predict: How will consumers, investors, competitors, restaurants, media, and regulators react over the next two weeks?
Step 1: The Seed Document
Everything starts with the seed document - a 13,000-byte brief containing only publicly available information from before the event. News articles about the fee change. Zomato's financial filings. Competitor pricing data. Regulatory frameworks. Industry reports.
The critical rule: nothing that wasn't published before March 24th. We're simulating what a smart analyst could have predicted on the eve of the announcement, not reverse-engineering the outcome.
Step 2: Ontology Extraction - Who Matters?
GPT-5.2 reads the seed document and builds a map of the ecosystem. It identifies:
- 53 distinct entities - individuals, companies, organizations, consumer segments
- Relationships between them - who competes with whom, who regulates whom, who influences whom
- Incentive structures - what each entity wants, what constrains them, what they fear
The whole system is interconnected. That's the point.
Step 3: Persona Generation - Making the Agents Real
Each entity gets a detailed persona. Not a flat archetype - a character with specifics.
Riya Sharma, 28, UX Designer, Bangalore: Orders lunch on Zomato 4x/week. Salary: Rs. 15 LPA. Splits orders with roommate to save on delivery. Price-conscious but hates cooking. Will complain on Twitter but won't uninstall.
Tanmay Bhat, Comedian/Content Creator: 3.2M Instagram followers. Makes observational comedy about everyday life. Will turn the fee hike into a meme within 48 hours. Doesn't care about the policy - cares about the joke.
Deepak Shenoy, Founder, Smallcase: Thinks in unit economics. Will calculate Rs. 2.40 × 100M monthly orders = Rs. 2.8-3B annualized incremental contribution profit. Investor-brain. Publicly bullish but privately notes demand elasticity risk.
These aren't hypothetical. The simulation creates dozens of these, each with distinct behavioral logic.
Step 4: The Simulation - 30 Rounds of Chaos
The 53 agents are placed on two simulated platforms - one resembling Twitter (short-form, viral), one resembling Reddit (long-form, threaded). Then the simulation runs for 30 rounds, each representing roughly half a day.
Round 1-3: The Initial Shockwave
Riya Sharma posts: "Just noticed Zomato charging Rs. 14.90 now. That's a 19% hike from last month. When did this happen? Feels like a checkout ambush - you don't see it until you're already committed to the order."
The phrase "checkout ambush" catches fire. Other consumer agents pick it up, remix it, amplify it. Within three rounds, it's a dominant narrative.
Tanmay Bhat responds: "Rs. 14.90 isn't a platform fee - it's a 'you're too lazy to go out' tax."
This gets shared widely. The humor reframes the conversation from outrage to resigned comedy - which, counterintuitively, diffuses boycott energy. People laugh instead of organizing.
Round 4-8: The Counter-Narrative Emerges
Deepak Shenoy posts his analysis: "+Rs. 2.40 per order times 100 million monthly orders equals roughly Rs. 280-300 crores annualized. This flows almost entirely to contribution profit. If I'm holding Zomato, I'm not mad - I'm impressed."
The investor agents cluster around this framing. Goldman's analyst agent writes a bullish note. ICICI's analyst agrees. The market narrative splits: consumers are angry, investors are optimistic.
Round 10-15: Competitors React
Swiggy's agent doesn't undercut - it follows with its own hike to Rs. 17.58. This is the non-obvious outcome that most analysts would have missed. Why would a competitor copy a move that's generating backlash?
Because both platforms face the same unit economics pressure, and neither can afford a price war in their path to profitability. The simulation captures this through the competitive dynamic between agents - they don't just react to Zomato's move, they react to each other's reactions.
Magicpin's agent sees the opening: "This is the biggest customer acquisition opportunity we've had in years." Zero platform fee promotion goes live.
Round 15-25: Behavioral Adaptation
The consumer agents don't boycott. They adapt. Riya starts pooling orders with her roommate. A student agent creates a "Zomato order group" to split fees. Restaurant agent Meera Nair starts a WhatsApp group for her regulars: "Why pay Rs. 14.90 to a platform when you can order directly from us?"
This is the "silent frequency collapse" - the insight that traditional analysis would miss. The headline metric (GMV) can look stable while order count quietly drops, degrading delivery density and rider efficiency.
Round 25-30: The New Normal
By round 30, the outrage has normalized. The "fee floor" is now just part of the ordering calculus. Consumers have adjusted their behavior. Investors are bullish. Competitors have responded. The regulatory agent is monitoring but not enforcing. The system has found its new equilibrium.
Step 5: The Report
The engine processes 1,572 interactions across both platforms and produces a 44,208-character report. The report maps:
- Stakeholder sentiment trajectories - who went from negative to neutral, who stayed bullish
- Dominant narratives - "checkout ambush," "lazy tax," "subsidy era is over"
- Behavioral shifts - basket inflation, frequency reduction, direct ordering
- Competitive dynamics - Swiggy follows, Magicpin attacks, restaurants defect
- Regulatory signals - monitoring, not enforcement; drip pricing is the attack surface
- Non-obvious risks - silent frequency collapse, receipts culture, transparency as the new battleground
The Scorecard: 11 for 11
We scored this simulation against what actually happened after the fee hike. Every single dimension was a HIT. 100% directional accuracy.
Consumer backlash: HIT. No boycott: HIT. Behavioral shifts: HIT. Investor bullishness: HIT. Swiggy follows: HIT. Magicpin attacks: HIT. Restaurant direct ordering: HIT. Media mixed framing: HIT. Regulatory monitoring: HIT. All 11 dimensions matched.
What This Means for You
The Zomato simulation took 41 minutes of compute time. The seed document took a day to prepare. The total cost was under $20 in API fees.
For that investment, a business leader would have known - before the announcement - exactly how the ecosystem would respond. Not guesses. Not opinions. Simulated behavior from 53 agents with distinct incentives, interacting over 30 rounds of real dynamics.
That's not science fiction. It's operational intelligence.
Read the full Zomato case study with accuracy scorecard at /case-studies/zomato, or email us to simulate your own scenario.