of agentic AI projects are expected to be discontinued by 2027
GartnerAI Discovery Workshop
Start with the outcome, not the technology.
Find out if your use case really needs an AI agent, and get a measurable goal and a clear Go/No-Go before you commit budget to building one.
How the AI Discovery Workshop works
- Step 0 · OptionalUse-Case Ideation Capture90-minute live session. Ideas tested against four gates.
- Step 1Problem Statement & Success Metrics20 questions. Measurable targets, tied to a named owner.
- Step 2Workflow-vs-Agent Scorecard13 questions, scored on 6 checks.
- ResultGo/No-Go recommendationA defensible rationale. “Not an agent” is a good result.
The AI Discovery Workshop is a short, focused engagement that decides whether an AI use case deserves an AI agent before you invest in building one. It is Stage 1 of Moxie’s Agentic AI Discovery approach. With the people who own the process and the people who would own the risk, you write a problem statement with measurable success metrics and score the use case in a Workflow-vs-Agent Scorecard. The result is a defensible Go, Hold or Redesign decision. A good result can be “not an agent”.
Most agentic AI projects stall on decisions, not technology
AI moves faster than teams can learn and adopt it. Many IT leaders end up between urgency and uncertainty, waiting for the landscape to settle.
- Costs higher than expected
- Unclear business value
- Risk controls that were never designed in
None of these is a technology problem. They are decision problems, and an evidence-based discovery solves them.
How much freedom should an agent have?
Every agentic AI sits somewhere on a scale of human involvement. If you do not define it, autonomy is decided without accountability, usually higher than intended and not what the risk requires.
- 1
Assistant
The AI drafts a recommendation. A person carries out every action.
Real-world analogyGPS navigation - 2
Approve each step
The AI proposes one action at a time. A person approves before anything happens.
Real-world analogyAdaptive cruise control - 3
Act, then review
The AI acts within an agreed scope. A person checks the work afterwards.
Real-world analogyLane-keeping assist - 4
Autonomous with exceptions
The AI completes the task alone and only stops for defined high-risk cases.
Real-world analogyHands-free highway autopilot - 5
Fully autonomous
No human checkpoint at all. Rare in practice and high risk.
Real-world analogyDriverless robotaxi
Each action an agent can take needs its own autonomy level and a named approver, set deliberately during discovery. A workflow that reads and drafts is a different bet from one that acts and spends on its own, and the two should never be approved the same way.
Three steps to a defensible decision
A short, focused engagement with the people who own the process and the people who would own the risk. Bring one use case idea. If you do not have one yet, start with the optional ideation session.
Use-Case Ideation Capture
A 90-minute live session to surface and screen candidate ideas, tested against four gates. For teams that have not named a use case yet.
A short, gated list of candidate use cases.
Problem Statement & Success Metrics
Structured discovery questions that pin the idea to an outcome and an owner, so success is measurable before anything is built.
A clear, written problem statement with success metrics, tied to a named owner.
Workflow-vs-Agent Scorecard
Scores whether the use case meets key agentic AI requirements, such as how predictable the work is, whether results can be verified, whether actions can be reversed and whether the data is available.
A Go, Hold or Redesign call, with a defensible rationale.
“Not an agent” is a good result
Work that is fixed, predictable and easy to verify is usually cheaper and safer as conventional automation. Finding that out before you build is the point of the workshop.
AI use cases in your industry
Bring one use case to the workshop. In each of these, AI prepares and checks, and people approve the decisions that matter. Pick your industry to see what Moxie builds.
Intelligent Processes, Smart AI Agents
From customer requests, complaints and onboarding to KYC, AML, lending and compliance review, with governance and audit.
See use cases HealthcareFrom Administration to Clinical Intelligence
Less administrative work in billing, patient intake and auditable records, with health data protected.
See use cases InsuranceAgentic claims processing for insurers
Claims from first notice of loss to settlement, with people approving the decisions that matter.
See use casesThe AI Discovery Workshop, answered
A short, focused engagement that tests whether an AI use case deserves an AI agent before you invest in building one. With the people who own the process and the people who would own the risk, you write a problem statement with success metrics and score the use case in a Workflow-vs-Agent Scorecard. It is Stage 1 of Moxie’s Agentic AI Discovery approach.
Bring one use case idea. If you have not identified one yet, the workshop can start with an optional ideation session (Step 0), a 90-minute live session that surfaces and screens candidate ideas first.
The people who own the process and the people who would own the risk. Two required steps, run in sequence, produce a decision you can act on immediately.
A written problem statement with measurable success metrics, tied to a named owner. A Workflow-vs-Agent Scorecard with a Go, Hold or Redesign call and a defensible rationale. And, if you started with ideation, a short, gated list of candidate use cases.
That is a good result. Work that is fixed, predictable and easy to verify is usually cheaper and safer as conventional automation, so you do not commit budget to an agent you do not need.
Gartner expects more than 40% of agentic AI projects to be discontinued by 2027, citing costs higher than expected, unclear business value and risk controls that were never designed in. These are decision problems, not technology problems, so Moxie starts with evidence.
Only use cases approved in the workshop move to Detailed Discovery (Stage 2), a separate, bounded stage of six steps: workflow map, actions and data, autonomy, safeguards and governance, cost model and failure plan. It ends in a costed implementation plan with scope, cost and expected return.
Online, in English. After you get in touch, our team talks to you to agree the format and the next step.
Book your AI Workshop
Leave your work email and we will contact you to schedule your AI Discovery Workshop. Stage 1 is a short, focused commitment.