What is a digital twin of an organization, and how does a small firm build one?

Gartner's definition is longer. It calls the DTO a dynamic software model that relies on operational and contextual data to understand how an organization:

• operationalizes its business model,
• connects with its current state,
• responds to changes,
• deploys resources,
• simulates future states,
• and delivers customer value.

The concept is credited to Gartner analyst Marc Kerremans. Gartner covered the category in market guides before it ranked vendors; GBTEC, for one, was listed in the January 2025 edition (GBTEC).

On 27 July 2026 Gartner published its first Magic Quadrant for DTO platforms, with Kerremans as lead author (SAP, 31 July 2026; ARIS, 30 July 2026; Celonis, 30 July 2026).

In plain words: an engineering twin copies a machine so you can stress it without breaking it. An organization twin copies the way your company turns demand into delivered work and cash, so you can stress a decision without paying for it. IBM sorts engineering twins into component, asset, system and process twins (IBM); the DTO sits one level above the process twin, because it includes people, clients and money.

Two questions sort the tools: does it describe the past or play the future, and does it model the machinery of the firm (work, capacity, cash) or the choices people make? A DTO is the only one that plays the future on both, tied to your own data.

What each neighbour lacks for a decision: a BI dashboard has no "what if" and no causes; process mining describes the past and does not play people's reactions; simulation software is not tied to your live data unless you build that link; synthetic panels have no operations, capacity or cash. A good DTO borrows from all four: the event log of process mining, the engine of simulation, the behaviour models of synthetic panels and the habit of measuring that BI built. Its own weak point is trust, so it must prove itself on the past before anyone acts on it.

Five questions, one per thing a twin cannot do without. Each answer scores 2, 1 or 0 points; the result names your weakest point and the next step. Nothing leaves your browser.

Phases and gates refer to the six-phase build method further down: Gate 1 is the founder signing the goal, Gate 2 the replay of the last 12 months.

It is a DTO in which AI agents stand in for the people whose decisions move the numbers: clients who sign or leave, partners who review or delegate, competitors who cut prices.

It is one of three kinds of AI digital twin, next to twins of machines and of single people.

83% vs 74%. In a 2024 Stanford-led study, agents built from two-hour interviews reproduced people's survey answers about 83% as well as the people reproduced their own answers two weeks later. Agents built from demographics alone reached 74% (Park et al., 2024).

Investors noticed: Simile, the start-up founded by the study's lead author, raised $200 million in July 2026 (TechCrunch).

The same research shows the trap: synthetic respondents squeeze variance, drift toward average answers and react to how a prompt is worded. A twin you can trust follows four rules:

1. Agents never produce the numbers. A flow engine computes volumes, delays and cash.
2. Each agent is calibrated on the firm's own past decisions, such as past proposals that were signed or lost.
3. Variance is restored on purpose, so the simulated clients disagree as much as real ones.
4. An agent that does not beat a simple rule on history (say, the average win rate) is switched off.

A service firm has fewer machines and more decisions than a factory, so its twin is built around objects and events: a client, a mission, a consultant role, an invoice.

From a year of such events, the twin rebuilds the real process. It is usually far from the one written in the quality manual.

Two laws then do most of the explaining. Little's law: work in progress equals throughput times lead time, so a firm that accepts more work without more capacity simply waits longer.

The theory of constraints: a system moves at the pace of its bottleneck, so speeding up any other step changes nothing.

In owner-led firms, the bottleneck is very often the founder's own review time. Classroom games such as Littlefield teach the same lesson on a fictional factory, as our guide to business simulation games shows. You can see it happen in our Twin Lite simulator and in the Maison Delaunay case.

Six phases and two gates. The founder signs the goal before anything is modelled, and nothing is decided before the twin passes its replay certificate.

Three phases need more than a label. Frame fixes the win condition: sell in three years, a revenue target at constant margin, a four-day week, or six months of cash after the worst shock. Connect reads the CRM, time or mission tracking and invoicing through read-only connectors, metadata only, with the compliance file ready before any staff data is touched. Play and steer logs every decision with its forecast, so the twin is checked against what really happened.

Phases 1 to 5 take six to ten weeks for a firm of up to 50 people. Our twins of service firms, built by AI Jungle, follow this method, with builds from $55,555; larger scopes, maintenance and partnership are priced with you.

The platforms are built for large enterprises. ARIS, Celonis, GBTEC and SAP each announced a Leader placement in Gartner's first DTO Magic Quadrant, published on 27 July 2026. The table lists them in alphabetical order, not as a ranking. The middle column is what each vendor stresses in its own announcement; the last column is our reading of who it suits.

Placements are as each vendor announced them; Gartner's report sits behind its paywall, and Gartner does not endorse vendors. Simulation toolkits such as AnyLogic and Simio let analysts build their own twin; our digital twin software ranking compares their prices. None of the vendors in our 2026 list of digital twin companies packages a twin for owner-led service firms of 5 to 50 people; that is the gap we work in.