{ "$schema": "https://transformation.ai/schemas/roadmap-1.0.0.schema.json", "version": "1.0.0", "contentDate": "2026-11-08", "updatedAt": "2026-11-08", "canonicalURL": "https://transformation.ai/roadmap", "canonicalMarkdown": "https://transformation.ai/roadmap.md", "hub": { "name": "AI Transformation Hub", "publisher": "Transformation.ai", "publisherURL": "https://transformation.ai/", "editorialStandards": "https://transformation.ai/about#editorial-standards", "disclosures": "https://transformation.ai/about#disclosures" }, "model": { "name": "The AI Transformation Roadmap", "description": "Five cumulative stages for turning AI ambition into better ways of working, from instrumenting the first existing workflow to running deep autonomy inside guardrails.", "practice": "cumulative", "stages": [ { "id": "pilot", "position": 1, "name": "Pilot", "url": "https://transformation.ai/roadmap/pilot", "summary": "Instrument one existing workflow end to end — data in, transforms, integrations, outputs, metrics.", "decision": "Which existing workflow to pilot — and whether to automate it fully, extend it, or run it with human-in-the-loop review.", "artifacts": [ { "name": "Workflow map", "description": "Data in (start and mid-stage), transforms, integrations (in and out across stages), outputs (data out, finished products), downstream consumers." }, { "name": "Baseline", "description": "Before-state metric values for task time, quality, and cost." }, { "name": "Automation posture decision", "description": "Full automation, extended workflow, or human-in-the-loop — with rationale." }, { "name": "Instrumented pilot", "description": "The workflow running under real conditions, logging outputs and interventions." } ], "metrics": [ { "name": "Time per task", "unit": "minutes", "direction": "down" }, { "name": "Error rate", "unit": "percent", "direction": "down" }, { "name": "Quality score", "unit": "index", "direction": "up" }, { "name": "HITL intervention rate", "unit": "percent", "direction": "stable" }, { "name": "Cost per unit", "unit": "currency", "direction": "down" } ], "trap": "Automating without mapping the data. If you don't know what enters at each stage — start and mid-way — you can't instrument outputs, set HITL thresholds, or prove anything changed.", "timebox": { "minWeeks": 4, "maxWeeks": 8, "display": "4–8 weeks" }, "owner": "Function leader with an AI engineer; a named workflow owner", "advanceCondition": "The workflow beats its baseline on task time or quality, and the HITL intervention rate is stable and documented.", "extra": { "workflowMap": { "dataIn": ["Start-stage inputs", "Mid-stage inputs"], "transforms": "Steps that change the data — decisions, rules, handoffs", "integrations": "Systems, APIs, people, approvals — in and out, across stages", "outputs": "Data out, finished products, downstream consumers, feedback loop" } } }, { "id": "expand", "position": 2, "name": "Expand", "url": "https://transformation.ai/roadmap/expand", "summary": "Turn one pilot into a repeatable pattern across functions — and level up operators and strategists who run the workflows.", "decision": "How to run the next ten to twenty workflows without paying the first-pilot cost each time — and how to level up the people who run them.", "artifacts": [ { "name": "Pattern library", "description": "Workflow archetypes — reusable designs that reduce the cost of the next workflow." }, { "name": "Operating model", "description": "Who runs what, with what training and what authority." }, { "name": "Operator role definition", "description": "The person who runs the workflow in production, handles HITL, tunes thresholds, and knows what normal looks like." }, { "name": "Strategist role definition", "description": "The person who redesigns the workflow, sets the goal, and owns the automation-versus-risk tradeoff." }, { "name": "Reusable integrations and evals", "description": "Shared components that don't get rebuilt per workflow." } ], "metrics": [ { "name": "Workflows in production", "unit": "count", "direction": "up" }, { "name": "Launch velocity per new workflow", "unit": "weeks", "direction": "down" }, { "name": "Operator count", "unit": "count", "direction": "up" }, { "name": "HITL intervention trend", "unit": "percent", "direction": "down" }, { "name": "Unit cost trend", "unit": "currency", "direction": "down" } ], "trap": "Treating each new workflow as a fresh pilot. You rebuild the integrations, the evals, the HITL, the metrics — and you scale cost linearly with workflow count.", "timebox": { "minWeeks": 13, "maxWeeks": 26, "display": "1–2 quarters" }, "owner": "Transformation office, with each function running its own workflows", "advanceCondition": "The pattern library is reused without help. A function leader launches a new workflow on their own — no central team required.", "extra": { "newRoles": ["Operator", "Strategist"], "reuseArtifact": "Pattern library" } }, { "id": "executize", "position": 3, "name": "Executize", "url": "https://transformation.ai/roadmap/executize", "summary": "Build the coordination layer — an executive intelligence agent that self-extends and improves workflows, inside a registry and a decision log.", "decision": "Do you build the coordination layer — an executive intelligence agent that sees all workflows, proposes extensions, and improves the agents inside them?", "artifacts": [ { "name": "Workflow registry", "description": "Machine-readable inventory of every workflow and the agents inside it." }, { "name": "Coordination agent", "description": "The controller that sees all workflows and proposes extensions." }, { "name": "Decision log", "description": "Every extension and improvement — auditable, revertible, reviewable." }, { "name": "Evaluation harness", "description": "The test set that every proposed change must pass before it lands." }, { "name": "Permission model", "description": "What the coordination agent is allowed to do without approval." } ], "metrics": [ { "name": "Workflows self-extended vs. human-initiated", "unit": "ratio", "direction": "up" }, { "name": "Agent-proposed improvements accepted", "unit": "percent", "direction": "up" }, { "name": "Coordination latency", "unit": "milliseconds", "direction": "down" }, { "name": "Drift detection latency", "unit": "hours", "direction": "down" } ], "trap": "Building the brain before the substrate. A coordination agent without a registry, evaluations, and a decision log is a chatbot with grand ambitions.", "timebox": { "display": "1 quarter to design, 1–2 to trust" }, "owner": "Transformation office, a designated agent owner, and an executive sponsor", "advanceCondition": "The coordination agent's recommendations are acted on without re-derivation, and the decision log has survived an external audit.", "extra": { "components": ["Registry", "Controller", "Decision log"], "auditRequirement": "External audit passed" } }, { "id": "transform", "position": 4, "name": "Transform", "url": "https://transformation.ai/roadmap/transform", "summary": "Design new workflows, roles, and functions that weren't possible before AI — not the same workflows with a tool layered on.", "decision": "What could your organization do that was impossible before AI — and what roles and functions does doing it require?", "artifacts": [ { "name": "New workflow designs", "description": "End-to-end designs for work that couldn't run before AI." }, { "name": "New role charters", "description": "What a person does — and what decisions stay human." }, { "name": "New function definitions", "description": "What team, budget, and mandate the new workflow requires." }, { "name": "Business case", "description": "Value from new workflows, not from old ones running faster." } ], "metrics": [ { "name": "New capabilities live", "unit": "count", "direction": "up" }, { "name": "Value from new workflows", "unit": "currency", "direction": "up" }, { "name": "Roles created or retired", "unit": "count", "direction": "up" }, { "name": "Time-to-competence for new roles", "unit": "weeks", "direction": "down" } ], "trap": "Renaming old workflows 'transformed.' If the workflow still runs the same way with a tool layered on, it isn't transformation — it's a rebrand.", "timebox": { "minWeeks": 26, "maxWeeks": 52, "display": "2–4 quarters" }, "owner": "Executive sponsor, with function leaders and HR/people", "advanceCondition": "A role or function exists that didn't before — and produces value you can point to that isn't just the old workflow running faster.", "extra": { "designObjects": ["Workflow", "Role", "Function"], "test": "Not a rebrand: the workflow runs differently, not just faster." } }, { "id": "autonomate", "position": 5, "name": "Autonomate", "url": "https://transformation.ai/roadmap/autonomate", "summary": "Run deep autonomy at tiered levels — with self-improvement ceilings, kill switches, audit trails, and a review cadence.", "decision": "How far do you let autonomy go — and what guardrails must already exist before you go further?", "artifacts": [ { "name": "Autonomy tier definitions", "description": "L0–L4 per workflow, with the guardrails required at each." }, { "name": "Guardrail spec", "description": "Written rules for what each tier can and cannot do." }, { "name": "Kill switches", "description": "A tested mechanism to stop, with a named human who can invoke it." }, { "name": "Evaluation cadence", "description": "How often each workflow is re-tested against its ceiling." }, { "name": "Self-improvement rules", "description": "The evaluation ceiling, what can change, and what cannot." } ], "metrics": [ { "name": "Autonomy tier distribution", "unit": "percent per tier", "direction": "stable" }, { "name": "Incident rate by tier", "unit": "ratio", "direction": "down" }, { "name": "Review burden per workflow", "unit": "hours/quarter", "direction": "down" }, { "name": "Drift detection latency", "unit": "hours", "direction": "down" } ], "trap": "Autonomy without a kill switch. Or self-improving workflows without an evaluation ceiling. Or autonomy without an audit trail. Each of these ends the same way — a workflow you can't stop.", "timebox": { "display": "Ongoing — reviewed quarterly", "cadence": "Quarterly autonomy review" }, "owner": "Transformation office with risk, security, and executive sponsor", "advanceCondition": "Every workflow has a kill switch, an audit trail, and an evaluation ceiling — or it does not move up a tier.", "extra": { "autonomyTiers": ["L0", "L1", "L2", "L3", "L4"], "governanceGate": "Quarterly autonomy review. No tier move without all three guardrails." } } ] }, "vocabulary": { "autonomyTiers": [ { "id": "L0", "name": "Suggest", "definition": "Agent proposes; a human acts. Every decision, action, and output is human-initiated.", "requires": ["Workflow map"], "url": "https://transformation.ai/glossary/autonomy-tier#L0" }, { "id": "L1", "name": "Approve", "definition": "Agent acts, but only on human approval. The agent does the work; the human approves or rejects the result.", "requires": ["Audit log of approvals and rejections", "Rejection-rate review"], "url": "https://transformation.ai/glossary/autonomy-tier#L1" }, { "id": "L2", "name": "Exception", "definition": "Agent acts autonomously on a defined set of cases; humans review exceptions only. The exception line is written before the workflow runs.", "requires": ["Written exception definition", "Audit log", "Volume-anomaly alerting"], "url": "https://transformation.ai/glossary/autonomy-tier#L2" }, { "id": "L3", "name": "Self-improve", "definition": "Agent acts autonomously and tunes itself within an evaluation ceiling the human sets. The agent cannot change the ceiling.", "requires": ["Evaluation ceiling", "Decision log", "Drift monitoring", "Kill switch"], "url": "https://transformation.ai/glossary/autonomy-tier#L3" }, { "id": "L4", "name": "Self-extend", "definition": "Agent proposes entirely new workflows that serve a defined goal. Proposed workflows run at L0 or L1 until they have proven themselves.", "requires": ["All L3 requirements", "Proposed-workflow review board", "Per-workflow kill switches", "New-workflow approval cadence"], "url": "https://transformation.ai/glossary/autonomy-tier#L4" } ], "workflowMapParts": [ { "id": "data-in", "name": "Data in", "definition": "Inputs at workflow start and at mid-workflow steps. Capture source, format, quality, and permissions for each." }, { "id": "transforms", "name": "Transforms", "definition": "Steps that change the data — decisions, rules, handoffs, and step boundaries." }, { "id": "integrations", "name": "Integrations", "definition": "Systems, APIs, people, and approvals that connect stages. Inbound and outbound." }, { "id": "outputs", "name": "Outputs", "definition": "Data out, finished products, downstream consumers, and the feedback loop." } ] } }