Keep business context connected
What could a digital twin of an organisation actually do?
A digital twin of an organisation is a digital representation of how that organisation operates, connected to information about its current state. The term covers different ambitions, from understanding processes to exploring possible changes. For a service business, a useful starting point is context that helps people understand a client, a decision and the work around it.
Define the useful decision and how its supporting context stays current before promising a complete model of the organisation.
Be clear about what the twin represents
Research on organisational twins connects enterprise models with operational information. A DFKI paper describes using structured relationships and context to bring those models closer to the work happening in the organisation.
Stronger claims include simulation and reliable prediction of organisational changes. Those require additional evidence. A Journal of Organization Design paper examines the limits of representing organisations with their human choices and conflicts.
For our current build, Said uses “digital twin” to describe connected business context that supports users and agents. We do not present it as a complete real-time simulation of the organisation or a validated predictor of every decision.
What Said has reported about the current build
The build brings context together periodically from internal sources including SharePoint, email and Slack. It keeps information connected to the work so applications and agents can use the relevant context.
One reported use helped someone unfamiliar with an account obtain enough context to cover a colleague’s meeting, in as little as 30 minutes. That is one firsthand report with no previous preparation baseline or independently inspected meeting outcome.
Said has also described an agent capability for investigating a falling metric. That is a reported capability, separate from an observed investigation with a checked diagnosis. Our delivered analytics case establishes connected reporting information, not proof of autonomous causal analysis.
Carry provenance into the decision
| Part | Question it answers |
|---|---|
| Source origin | Where did this information come from? |
| Validity dates | When did this information apply? |
| Collection date | When was it recorded in the system? |
| Relationship | Which client, agreement, person or task does it concern? |
| Responsibility | Who can resolve uncertainty or approve a change? |
| Permission | Who is allowed to use the information? |
Said describes confidence declining if information is not seen again by its validity date. Missing or contradictory evidence must remain visible. If a required trace cannot be queried or quoted for human review, the system must say it does not know.
The agent’s role in this description is to give people the information they need to decide. Reviewed decisions return to the system and update its understanding of the business. That does not imply that every interaction automatically retrains a model’s weights.
Begin with a useful job and a bounded model
Pick a question such as “what should a colleague know before this account meeting?” Identify the records, relationships and update rules needed to answer it. Test normal, missing and contradictory information with the people who understand the account.
Define read access and action authority separately. A system that can prepare a brief does not need authority to change the agreement it describes. Add actions only after their rules, review and failure behaviour have been tested.
Keep future opportunities in a separate part of the plan. Client-operated services using shared business context are an opportunity Said is exploring, with customer demand and delivery economics still to be tested.
Plan the context and its limits
Worked example
A bounded account-context model
Invented teaching example. No client outcome is claimed.
JOB
Prepare an invented account-meeting brief.
CONTEXT
Connect the client’s agreement, goals, recent decisions and open issues to their sources and owners.
UPDATE
Reviewed meeting decisions update the decision records; validity and collection dates remain separate.
HUMAN ROLE
The account owner resolves an issue where two current records disagree.
LIMIT
The model supports preparation. It does not predict the meeting outcome or simulate the entire organisation.
Further reading
Written by Said Lopez, founder of LeverageAI and author of Lean AI. Our guides draw on the method in the book and our work with clients.