From Idea to Working AI in 15 Days: Your Framework to Run an Agentic AI POC

[SylvainHauserN2]

Most organisations have no shortage of AI ideas. The challenge is knowing which one to start with, how to scope it, and how to get to a result fast enough to maintain business momentum.

This framework consistently gets organisations from idea to working AI in 15 days. Not a demo. A working system, tested against criteria the business agreed upfront, on live data.

The 5-phase framework

 

Phase 1 -- Scope and align (Days 1-5)

Four things must be locked before build begins:

• Thin slice: one bounded business problem, one realistic agentic use case
• Clear ROI target: agree upfront on what success looks like -- risk reduction, faster processing, cost avoidance, or NPS improvement
• Measurable baseline: one metric the business already tracks
• Co-design: business owners in the room from day one, not brought in at the end

 

Phase 2 -- Design (Days 6-7)

• Lock data sources, process logic, and human-in-the-loop checkpoints before a single line is written
• Two days of design saves five days of rework

Phase 3 -- Build (Days 8-12)

• Build in an evaluation or sandbox instance
• Business validates weekly against real conditions

 

Phase 4 -- Validate (Days 13-14)

• Test against the baseline metric agreed in Phase 1
• Document edge cases honestly -- caveats build more trust than polish

 

Phase 5 -- Handover (Day 15)

Day 15 is not a demo day. It is a decision day. The business receives a working system, full documentation, a management decision walkthrough, and a clear path to production.

 

Before you scope anything: pick the right tool

Choosing between workflow automation, generative AI, and agentic AI before you build is a critical step -- and one covered in detail in a separate post: Stop reaching for AI first: your decision framework for picking the right tool for the right job 

 

What makes this a POC, not a demo

• Runs on your data, not on dummy content
• Business owners co-build throughout, not just at sign-off
• Validated against one metric you already track, not a curated scenario

Sylvain Hauser is an AI Architect at ServiceNow, based in Australia. He holds Certified Master Architect (CMA) and Certified Technical Architect (CTA) credentials with 18+ years of ServiceNow expertise.

[Shiny-Syslabo]

This is an excellent guideline. We are leading AI adoption initiatives in Japan, and in our experience there is often an additional Phase 0 before the phases you described.

Perhaps some of the activities are already included in your Phase 1, but we view it as a distinct stage:

Phase 0: Understanding what work truly requires AI

The reason this phase is necessary is that:

• Many users start with the desire to "use AI" without fully understanding which business processes are best suited for partnering with AI.
• They are often unclear whether a given requirement should be solved through traditional programming or through AI.

For example, if the goal is automatic incident assignment, it may be better to implement it using an out-of-the-box (OOTB) solution rather than relying on AI. In fact, this is exactly the kind of scenario where a Build Agent can add value.

Phase 0 exists to address these challenges and help organizations identify the right opportunities for AI.

That said, we find the sequencing of the subsequent phases very insightful, and we would certainly like to use it as a reference in our own approach as well.