How to Build a World Model for a 10-Person Company

A world model for a 10-person company sounds like something a Fortune 500 CTO would greenlight after a two-day offsite. It isn't. A small company is actually the ideal place to build one — fewer moving parts, tighter feedback loops, and the people who own the data are usually in the same Slack workspace. Here's what that world model would actually look like: the specific entities, the relationships between them, and the data points worth capturing.

What a World Model Actually Is

A world model is a structured, queryable computational model of your organization. Not a org chart. Not a project management tool. A living graph of the things that make your business run — who does what, what assets exist, what processes connect them, what state everything is in right now. The goal is to make your company legible: to you, to AI systems operating on your behalf, and eventually to external systems that need to understand what you do and how you do it.

The Core Entity Types for a 10-Person Company

Start here. These are the objects your world model needs to know about before anything else.

• —People — employees, contractors, advisors. Attributes: role, skills, capacity (hours/week), active projects, decision authority
• —Customers — current, past, prospective. Attributes: industry, contract value, renewal date, health score, primary contact
• —Products or services — what you actually sell. Attributes: pricing model, delivery method, dependencies, responsible owner
• —Assets — software tools, domains, IP, physical equipment. Attributes: owner, cost, renewal date, access permissions
• —Processes — recurring activities that produce outputs. Attributes: trigger, steps, owner, SLA, last run, failure rate
• —Obligations — contracts, compliance requirements, SLAs, subscriptions. Attributes: counterparty, due date, value, status

The Relationships That Actually Matter

Entities without relationships are just a list. The world model gets useful when you map the edges — which person owns which process, which process produces which asset, which customer depends on which product. For a 10-person company, the most valuable relationships to capture are:

• —Person → Process (owns, executes, approves)
• —Process → Asset (produces, consumes, modifies)
• —Customer → Product (contracted for, in trial, churned)
• —Obligation → Person (responsible for fulfilling)
• —Asset → Asset (depends on, replaces, generates)
• —Person → Customer (relationship owner, support contact)

Here's the counterintuitive part: the relationships are more valuable than the entities themselves. You probably already know your customer list. What you don't have written down anywhere is that your onboarding process depends on three tools, two of which auto-renew in October, and the one person who knows how to run it is on parental leave in November. The world model surfaces that. A spreadsheet doesn't.

The Data Points Worth Actually Capturing

Don't capture everything. Capture what changes the answer to a question. For a 10-person company, the highest-signal data points are:

• —Capacity state — who has bandwidth, who is overallocated, right now
• —Revenue concentration — what percentage of ARR is in your top 3 customers
• —Process health — which recurring processes failed or ran late in the last 30 days
• —Obligation exposure — total value and count of obligations due in the next 90 days
• —Asset risk — tools or domains with no clear owner or expiring access
• —Skill coverage — which skills exist in only one person (single points of failure)

Why This Matters for AI Systems Right Now

ChatGPT, Claude, Perplexity, and Gemini are already being used to answer questions about businesses — by customers, by partners, by job candidates. The companies those AI systems recommend and describe accurately are the ones that have made themselves legible: structured information, consistent public signals, clear entity definitions. A world model is the internal version of that same discipline. It makes your company queryable by the AI tools your own team uses, and it produces the structured outputs that improve how AI systems represent you externally. These are not separate problems.

A Minimal Starting Point

You don't need a graph database on day one. A minimal world model for a 10-person company can start as three linked tables: one for entities (name, type, owner, status), one for relationships (entity A, relationship type, entity B), and one for state (entity, metric, value, timestamp). The discipline of maintaining those three tables consistently is more valuable than any software you could buy. Once the data is clean and current, you can query it, visualize it, or feed it to an AI system. Until then, you're just speculating.

Frequently Asked Questions

What's the difference between a world model and a digital twin?

A digital twin is a good entry-level concept for small businesses — a structured snapshot of how your company works. A world model goes further: it's designed to be queried, updated in real time, and used as the computational foundation for AI operating on your behalf. Think of a digital twin as the on-ramp and a world model as the destination.

How long does it take to build a basic world model for a small company?

A first version with 30–50 entities and their core relationships can be assembled in a focused week. The harder part is keeping it current. That's why the processes and ownership fields matter as much as the data itself — someone has to be responsible for each record.

Does a world model help with how AI systems describe my business?

Indirectly, yes. The same discipline that produces a clean world model — defined entities, consistent attributes, explicit relationships — also produces the structured public signals that ChatGPT, Perplexity, Claude, and Gemini use when generating answers about companies in your category. A company that is legible internally tends to be more legible externally.

What's the biggest mistake companies make when trying to build one?

Capturing too much and maintaining nothing. Start with the six entity types listed above, map only the relationships that change the answer to a real question, and pick one person to own the model. Breadth without maintenance produces expensive noise.

Is a world model only useful once a company gets bigger?

No — it's actually most useful at 5–20 people, before institutional knowledge becomes fully tribal. The 10-person company where one person knows everything about a client or process is the most fragile version of a company. A world model is how you make that knowledge structural instead of personal.

The Practical Next Step

Before you build the internal model, it's worth understanding how AI systems currently see your company from the outside. Your public-facing information — your website, your structured data, your entity definitions — is what AI systems like ChatGPT and Perplexity are already working with when someone asks about you. That's a simpler, faster place to start, and it gives you a baseline.