Running a Context Discovery Workshop

A context graph should not begin with every available table, API, or document. It should begin with the business decisions that need better context. This guide shows Enterprise Architects how to facilitate a structured workshop that converts operational questions into a decision-domain map, entity model, evidence inventory, policy set, and practical delivery backlog.

The 60-second read

A context discovery workshop is a facilitated working session that identifies which business decisions matter, what questions must be answered before those decisions, which entities and relationships provide the evidence, what policies constrain action, and which systems hold the required signals. The workshop should produce concrete artifacts, not a broad data wish list: a decision inventory, entity and relationship map, critical-question catalogue, source and freshness matrix, policy register, and prioritized context backlog. The strongest workshops move through the Understand-Decide-Execute loop. They begin with decision outcomes, map the context required to understand each situation, define the governed choices available, and identify how actions and outcomes will return to the Enterprise Digital Twin.

Key takeaways

Definition

A context discovery workshop is a facilitated method for defining a bounded enterprise decision domain by identifying the decision moments, questions, entities, relationships, evidence, policies, source systems, freshness requirements, permitted actions, and measurable outcomes needed to run an Understand-Decide-Execute loop.

Why context discovery must begin with decisions #

Most enterprise modeling exercises begin with what the organization already has: application inventories, data dictionaries, integration catalogues, process maps, and reporting requirements. Those assets are useful, but they are poor starting points for context engineering. They describe the supply of information, not the demand for understanding. A context discovery workshop reverses that sequence. It begins with a consequential decision and works backward to the context required to make and execute it well.

Consider a telecommunications retention decision. The source-first question is, “Which customer tables should we connect?” The decision-first questions are different: Which customer is at risk? What changed recently? Which services and household relationships matter? What commitments are active? Which offers are permitted? What intervention is most likely to preserve value without creating margin leakage? These questions reveal that the domain spans customer, account, product, interaction, network experience, billing, consent, offer, and outcome entities. The workshop discovers this connected shape before the team commits to an ontology or integration plan.

The method is equally useful in banking, manufacturing, healthcare, insurance, retail, and workforce operations. The selected decision changes, but the discovery logic remains stable: define the situation, identify the evidence, express the relationships, govern the choices, and close the loop through executed outcomes.

CONTEXT DISCOVERY WORKSHOP: FROM DECISION TO DELIVERY BACKLOG 1. FRAMEOutcome and decisionScope and ownerSuccess measures 2. UNDERSTANDQuestionsEntities and linksEvidence and freshness 3. DECIDEOptions and rulesRisk and confidenceHuman authority 4. EXECUTEActions and channelsSystems of recordOutcome capture 5. PRIORITIZEContext APIsQuality gapsDelivery backlog Workshop outputs become the first decision-domain blueprintDecision inventory • entity map • relationship map • source/freshness matrixPolicy register • context API candidates • quality risks • outcome measures • backlog
Figure 1. A workshop should move from business framing through Understand, Decide, and Execute, then convert the findings into a prioritized delivery backlog.

Before the workshop: establish a bounded decision domain #

The facilitator’s first responsibility is scope discipline. “Customer 360,” “supply chain visibility,” and “workforce intelligence” are not workshop scopes. They are broad ambitions. A workable scope names a decision, an actor, a trigger, and an outcome. Examples include: approve or escalate a commercial credit request; choose the next retention intervention for a high-value customer; reroute production after a supplier disruption; determine whether a claim can proceed automatically; or recommend a redeployment path for an employee whose project is ending.

Prepare a one-page charter containing the decision statement, accountable owner, current process, known pain points, affected users, target outcomes, exclusions, and the systems believed to be relevant. Treat the systems list as a hypothesis, not a fixed boundary. Circulate the charter and ask attendees to bring representative cases: one routine case, one difficult case, and one case where the current process produced a poor or delayed outcome.

Recommended participants

A strong group combines business authority with architectural and operational reality. Include the decision owner, an operational practitioner who makes or supports the decision, a domain expert, an Enterprise Architect, a data or integration architect, and a risk, policy, or compliance representative. Add security, legal, analytics, AI, or product specialists only where the domain requires them. Keep the active working group small enough for decisions to be made in the room.

Stage 1: frame the decision and its measurable outcome #

Open with cases, not definitions. Ask the operational participant to walk through a recent decision from trigger to outcome. Record what initiated it, who owned it, what information they sought, where they waited, what judgment they applied, what actions were available, and how the result was recorded. Then write a precise decision statement: “When [trigger] occurs, [decision owner] must choose [action] for [subject] to improve [outcome] while respecting [constraints].”

Separate the decision from the process around it. A process may include ten handoffs, but only two moments may require meaningful judgment. These moments are the best candidates for context engineering because they concentrate uncertainty, evidence, policy, and consequences.

Stage 2: discover the questions that define understanding #

For each decision moment, ask: “What must be true or known before a responsible choice can be made?” Capture questions in the language of the user. Do not translate them into fields too early. A procurement leader may ask, “How exposed are we if this supplier fails?” That question implies ownership, product dependency, geography, inventory, substitute supplier, contract, shipment, and revenue relationships. The question is richer than any single data attribute.

Classify questions into identity, state, history, relationship, causality, policy, prediction, and action categories. Mark which questions are currently answered reliably, partially, manually, or not at all. This becomes the critical-question catalogue and helps prioritize context gaps by decision impact.

Stage 3: map entities, relationships, and evidence #

Now identify the nouns inside the questions: customer, account, household, contract, supplier, facility, product, employee, role, claim, asset, policy, location, event. These are candidate entities. Then identify the verbs and prepositions: owns, supplies, depends on, serves, reports to, covered by, located at, purchased, complained about, approved by. These become typed relationships, often with effective dates, confidence, provenance, and status.

For every important fact or relationship, ask where the evidence originates, how frequently it changes, how quickly the decision needs the change, and which source has authority when systems disagree. This produces a source and freshness matrix. It also exposes where the Context Graph Engine will need entity resolution, temporal modeling, document extraction, or conflict handling.

Stage 4: define decision rules, permissions, and confidence #

Understanding does not automatically authorize action. The workshop must identify the choices available, the rules that eliminate or rank options, the evidence thresholds required, and the roles permitted to approve or override. Capture regulatory obligations, contractual restrictions, consent, segregation of duties, monetary thresholds, fairness requirements, and escalation conditions.

These constraints form the initial Context Harness requirements. They determine which context can be assembled for which purpose, what quality must be present, whether inferred relationships may be used, and which actions remain advisory, human-approved, or autonomous.

Stage 5: close the loop through execution and outcomes #

Ask how the chosen action reaches the operating system: a CRM task, workflow approval, pricing engine, case-management update, maintenance order, workforce marketplace, or customer channel. Then identify what outcome should return to the Enterprise Digital Twin. A recommendation without outcome capture cannot improve. The loop needs the action taken, actor, timestamp, override reason, immediate result, and delayed business outcome where relevant.

This is where the Execution Grid becomes concrete. Digital Workers and AI Agents may perform parts of the action, but the workshop should describe business responsibilities before selecting automation components.

Workshop artifact templates #

ArtifactMinimum fieldsWhy it matters
Decision inventoryTrigger, owner, subject, options, constraints, outcomeDefines the bounded decision domain
Critical-question catalogueQuestion, category, current answerability, impact, evidenceTranslates business judgment into context requirements
Entity and relationship mapEntity types, relationship types, dates, identity rulesProvides the first graph model
Source and freshness matrixFact, source, authority, update pattern, freshness SLOGuides ingestion and quality controls
Policy registerRule, purpose, owner, evidence threshold, permitted actorSeeds the Context Harness
Execution mapAction, target system, approval, write-back, outcomeConnects decisions to the Execution Grid
Context backlogCapability, dependency, value, risk, effort, acceptance testTurns discovery into delivery

A realistic enterprise scenario #

Enterprise scenario

A global manufacturer selects the decision “how to respond when a tier-one supplier disruption threatens a committed customer order.” The initial system list contains procurement, ERP, and logistics. During discovery, the team finds that a responsible decision also requires product bills of material, tier-two dependencies, inventory by location, alternate-part qualification, customer priority, contractual penalties, plant capacity, shipment status, and engineering approval.

Understand. The workshop maps supplier, part, product, plant, order, customer, contract, shipment, and qualification entities. It identifies critical questions such as which orders are exposed, how long inventory will last, and which substitutes are already approved.

Decide. The group defines permitted responses: expedite, reallocate stock, substitute a part, change production sequence, negotiate delivery, or escalate. Policy rules protect regulated products and strategic customers.

Execute. Approved actions create workflow tasks in procurement, planning, engineering, and customer operations. Actual recovery time, cost, delivery impact, and override reasons return to the twin. The first backlog focuses on one product family and three plants rather than the entire network.

Common mistakes to avoid #

Watch out for
  1. Starting with an enterprise-wide ontology instead of a bounded decision.
  2. Inviting only data and technology teams while excluding the people who make the decision.
  3. Capturing fields without preserving the business questions they answer.
  4. Modeling entities but not relationships, time, provenance, and confidence.
  5. Ignoring policy and permission until implementation.
  6. Ending with workshop notes rather than named artifacts, owners, and acceptance tests.
  7. Prioritizing by connector convenience instead of decision value and context risk.

Turning workshop outputs into an implementation plan #

Within forty-eight hours, publish a concise decision-domain blueprint. Validate disputed entities, relationships, source authority, and policies with named owners. Convert each context gap into a backlog item with a business question, required evidence, quality threshold, dependency, and acceptance test. Sequence delivery so the team can answer increasingly valuable questions, not merely connect increasingly many systems.

The first release should expose one or more governed context APIs to the Decision Layer, support a defined user or agent, and capture outcomes through the Execution Grid. This establishes a measurable loop and gives the next discovery workshop a working pattern to reuse.

How OpenKnowra approaches this #

OpenKnowra treats discovery as decision-domain design. The workshop outputs define what the Context Engine must ingest and resolve, what the Context Graph Engine must model, what the Decision Layer must evaluate, what the Context Harness must govern, and what the Execution Grid must perform and write back. The Enterprise Digital Twin grows through these bounded domains, connected by shared entities and policies rather than by a one-time attempt to model everything.

Frequently asked questions

What is a context discovery workshop?
A context discovery workshop is a structured session that identifies a high-value business decision, the questions needed to make it, the entities and relationships involved, the evidence and source systems required, the policies that constrain action, and the outcomes that should be written back. Its purpose is to define a bounded decision domain for context engineering.
Who should attend a context discovery workshop?
The core group should include a business decision owner, Enterprise Architect, domain subject-matter expert, data or integration architect, operational user, risk or policy representative, and a facilitator. Add security, legal, analytics, or AI specialists when the selected decision domain requires them.
How long should the workshop run?
A focused workshop commonly fits into one full day or two half-day sessions. Complex domains may require follow-up validation sessions, but the initial workshop should stay bounded to one decision domain rather than attempting to map the whole enterprise.
What artifacts should the workshop produce?
The minimum outputs are a decision inventory, critical-question catalogue, entity and relationship map, source and freshness matrix, policy and permission register, context quality risks, context API candidates, outcome measures, and a prioritized implementation backlog.
How do you choose the first decision domain?
Choose a decision that is frequent or economically important, currently suffers from fragmented context, has an identifiable owner, can be measured through outcomes, and can be bounded to a manageable set of entities and systems.

Keep exploring this cluster