Industry Ontology Schema

Contents

Why an ontology

A vector database is not industry knowledge.

Loading an industry's documents into a retrieval index gives a model access to text. It does not give it the structure the industry actually runs on.

An ontology models that structure directly. What things exist, how they relate, what events change them, what rules constrain them, and what decisions the industry makes about them.

The test is simple. Can the system traverse from a territory to a brand to a prescriber to a compliance rule without guessing? If yes, there is an ontology. If no, there is a search index.

The ten declarations

What an industry declares about itself.

ENTITY: A thing the industry recognises. A brand, an asset, a patient, a batch, a loan.
RELATIONSHIP: How entities connect, and in which direction.
EVENT: Something that happens and changes state.
METRIC: A measure the industry agrees on, with its definition attached.
RULE: A constraint. Regulatory, contractual, or internal policy.
WORKFLOW: A sequence the industry runs, with its steps and owners.
DECISION: A choice the industry repeatedly makes, and what it is made from.
ACTION: Something that can be done to an entity, in a named system.
ROLE: A human position, with its scope and authority.
AGENT: A specialist bound to roles, decisions and actions.

Entity and relationship

Identity, attributes, and where a thing is mastered.

Entities carry identity, attributes, and the systems where they are mastered. Relationships carry direction and cardinality.

ENTITY  brand
  mastered_in:  veeva, sap
  attributes:   molecule, indication, therapy_area, market

RELATIONSHIP  brand -> prescribed_by -> hcp
  cardinality:  many_to_many
  evidenced_by: rx_data

Event, metric, rule

What changes, what is measured, what constrains.

Events change entity state. Metrics are defined once, centrally, so two agents cannot disagree about what a number means. Rules are enforced, not advisory.

EVENT   prescription_written
  changes:  hcp.rx_behaviour, brand.volume
  source:   rx_data

METRIC  territory_growth
  definition:  (current_period - prior_period) / prior_period
  grain:       territory, brand, month

RULE    off_label_promotion
  type:        regulatory
  enforcement: block
  applies_to:  recommend, execute

Workflow, decision, action

What the business does, chooses, and changes.

Workflows describe what the business already does. Decisions describe what it repeatedly chooses. Actions describe what can be changed, and where.

WORKFLOW  cycle_planning
  steps:   segment, target, allocate, approve, publish
  owner:   commercial_ops

DECISION  hcp_prioritisation
  made_from:  rx_behaviour, territory_growth, brand_plan
  made_by:    field_manager
  cadence:    per_cycle

ACTION    crm_task_create
  system:      veeva
  write_path:  declared
  reversible:  yes

Role and agent

An agent can never exceed the authority of the role it serves.

Roles come from the organisation, not from the software. Agents are bound to roles, so an agent can never exceed the authority of the position it serves.

ROLE   field_manager
  scope:      territory
  authority:  call_plan, task_assignment

AGENT  commercial_excellence
  serves:     field_manager, commercial_ops
  decisions:  hcp_prioritisation
  actions:    crm_task_create

Rules for writing one

Five that save the most time.

Model what the industry argues about: If two people in the sector would disagree on a definition, it belongs in the ontology.
Master every entity somewhere: An entity with no system of record is a guess.
Define metrics once: Duplicate definitions are the most common cause of agents contradicting each other.
Make rules enforceable: A rule with no enforcement type is a comment.
Keep company specifics out: Company brands, territories and policies belong in the Company Pack, not the industry ontology.

Where to go from here.

Plate data

Spec: Super Brain
Part: 02 of 03
Declarations: 10
Version: 1.0

Modelling your industry? Start with the ontology.

Most of the work in an Industry OS is here. We are happy to review a draft.

Talk to us

A conversation, not a demo script. We ask what your systems of record are, what they cannot answer, and which decisions actually matter. If an Industry OS is not the right answer for you, we will say so.