The Business Journey Catalog — Organizing Agents Around How Work Actually Happens

A registry organized by repository or team is a registry only its builders can use. The breakthrough in an agent marketplace is to organize it around business journeys — the end-to-end flows of how work actually happens — so a marketer finds marketing capabilities, an underwriter finds underwriting capabilities, and everyone searches in the language of their job.

Six business areas, documented end to end

The catalog covers the enterprise as a set of journeys, each recording its activities, personas, systems, mapped MCPs & agents, KPIs, and gaps:

Business area	Journey
R&D	Idea → Launch (literature scan → protocol → experiment/analysis → regulatory dossier → launch)
Commercial	Lead → Renewal (territory planning → lead gen → opportunity → pricing → order → renewal)
Marketing	Plan → ROI (planning → audience → content → waterfall → rollout → analytics → ROI)
Customer experience	Contact → Loyalty (omnichannel intake → triage → resolution → follow-up → VoC → proactive care)
Supply chain	Forecast → Deliver (demand forecast → inventory → procurement → manufacturing/QA → logistics → returns)
Underwriting	Submission → Bind (intake → enrichment → risk scoring → pricing → referrals → quote → bind → portfolio)

The powerful move: any journey stage with no registered capability surfaces as a gap heatmap — and that heatmap is the marketplace's build/buy pipeline. Governance and demand planning become the same artifact.

A worked journey — Marketing, end to end

Take the Marketing journey as seven governed stages, each served by a named agent and MCP, each with a stage-matched autonomy level and a governance risk tier:

Stage	Agent	Autonomy	Risk	MCP
M1 Planning & budgeting	Planner copilot	L2 · Silver	Low	BudgetHub · HistPerf
M2 Audience & targeting	AudienceBuilder	L2 · gated · Gold	High	CDP
M3 Content & review	ContentGen	L1 · suggest-only	Medium	DAM
M4 Waterfall design	Waterfall Designer	L2 · human activates	Medium	Orchestrator
M5 Rollout	LaunchSentinel	L3 · bounded pause	Medium	Deliverability
M6 Channel analytics	ChannelInsights	L2 · governed SQL	Low	ChannelAnalytics
M7 Performance & ROI	ROIAnalyst	L2 · exec readout	Low	FinanceActuals

The pattern to notice: autonomy is matched to the stage, not the enterprise. Audience targeting touches consent-sensitive (PHI-adjacent) data, so it is gated at High risk with 100% human approval on action tools; content generation only suggests (L1); channel analytics runs governed SQL at Low risk. Learnings from M7 write back to campaign history, so the Planner and Propensity agents improve every cycle.

Auditable counts at every gate

A governed journey produces auditable numbers. A contact waterfall in stage M4, for example, logs counts in and out at every step:

Eligible + consented        480K
  − 65K  global suppressions (opt-outs, frequency caps, exclusions)
After suppressions          415K
  − 13K  cross-campaign dedup (priority conflicts resolved)
After dedup                 402K
  − 40K  reserved for holdout
Contacted (treatment)       362K   →  Email 210K · SMS 80K · Call 32K · Direct mail 40K
Holdout                      40K   →  control group: measures incremental lift (22% vs 15% = +7pt)

Leakage target = 0, holdout integrity preserved end to end — because every stage is a governed asset, not a spreadsheet.

Discovery in the language of the business

On top of the journey catalog sits a conversational assistant that answers three kinds of question:

• Asset discovery — "Which agents can take actions vs only suggest?" · "Who owns the AudienceBuilder agent and when was it last recertified?"
• Journey discovery — "Show me the Marketing journey end to end for healthcare" · "Where are the gaps in the Supply Chain journey?"
• Governance queries — "What's pending my approval?" · "If I deprecate the CDP connector v2, which agents break?"

Answers return ranked assets with certification badges, owners, and journey mappings — plus one-click access requests and dependency-graph impact for governance questions.

Measuring the marketplace

Four KPI dimensions prove it is working:

Dimension	Metrics
Adoption	Monthly active users, registered assets & growth, reuse rate per asset, journey coverage
Governance	100% of assets with complete metadata & named owners, approval-SLA adherence, on-time recertification
Risk	Zero unregistered (shadow) agents in production, high/critical assets with current eval evidence, guardrail incidents & time-to-suspend
Value	Duplicate builds avoided, time-to-approve trend, cost per invocation & model-spend attribution

Roll it out in 12 months

Start narrow, automate, then federate:

Phase	Window	Focus
1	0–3 mo	Registry + metadata model, manual approval workflow, onboard two areas (e.g. Marketing & Underwriting) with documented journeys
2	3–6 mo	Policy-as-code validation, risk-tiering automation, smart search + dependency graph
3	6–9 mo	Conversational catalog assistant, telemetry & drift monitoring, recertification automation
4	9–12 mo	All business areas + gap heatmaps, chargeback & cost analytics, federate with the Data Marketplace — agents cite the data products they consume

The exit criterion for every phase is the same: everything published carries complete metadata, a named owner, and a current approval — no exceptions.

The path to implement

1. Model the enterprise as business journeys, and map every stage to the agents/MCPs that serve it.
2. Turn unserved stages into a gap heatmap — your build/buy pipeline.
3. Match autonomy and risk tier to each stage, not the whole enterprise.
4. Add a conversational catalog assistant for asset, journey, and governance discovery.
5. Federate with the Data Marketplace so agents cite the data products they depend on.

Incipient's Data Marketplace accelerator provides the journey catalog, semantic discovery, and dependency graph — extended from data products to agents and MCPs.

Talk to us about a journey catalog →