> ## Documentation Index
> Fetch the complete documentation index at: https://docs.within.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Build your AI roadmap

> Turn your Work Brain into a plan: set the objective, find where the gap sits, choose the solution set, and decide where to deploy agents first.

Define is the second stage of the journey. Discover gives you a Work Brain (the **Process Index** in Within) built from observed work. Define turns it into a plan: the outcome you're going after, where the gap to that outcome sits, which changes close it, and in what order.

Most teams arrive with one question: *where should we deploy agents first?* Define answers it with evidence, and the answer usually includes changes that aren't agents at all.

> [!NOTE]
>
> **Before you start:** a populated Work Brain for the value stream in scope. If your analysis feels thin, there's usually more to capture first. See [Capturing current state](/user-docs/discover/capturing-current-state).

## The four steps of Define

**1. Set the objective**

Name the outcome in business terms before you pick any technology. A useful test is one sentence: **\[outcome and metric] for \[scope], by \[when], within \[constraints], owned by \[executive]**.

* **Vaguer:** "Deploy AI agents in finance this year."
* **Sharper:** "Cut month-end close from 8 days to 5 within two quarters, without adding headcount or net-new engineering. Owned by the Controller."

The sharper version leaves room for a process change, a rule-based automation, an agent, or no build at all. Record the baseline, the target and the constraints (regulatory, data residency, budget, timeline, human approval, technology) up front, because they decide which solutions can move forward.

**2. Find where the gap sits**

Point Advisor at the value stream and ask where time, cost, risk or friction concentrate. Work back from the outcome: which causes matter most, which sit outside your scope, and which work is high-volume versus on the critical path. Size the causes together rather than adding them up, because removing one bottleneck exposes the next.

See [Strategic analysis with Advisor](/user-docs/define/strategic-analysis-with-advisor).

**3. Define your solution set**

Choose the mix of changes that closes the gap: process redesign, automation, and enterprise transformation, with an owner for each. Simplify the work before you automate it.

See [Define your solution set](/user-docs/define/define-your-solution-set).

**4. Sequence the roadmap**

Decide what goes first and why. Separate quick wins your teams can deliver on their own from larger initiatives that need a central team, and keep the order the solution set implies: simplify, then automate, then change the foundation underneath.

## Where should we deploy agents first?

Start with candidates that are easy to test and safe to get slightly wrong. A good first agent usually passes all of these:

| Test | The question to ask |
| - | - |
| **Evidence** | Is it grounded in observed work, not assumption? |
| **Frequency** | Does the work happen often enough to learn from and to matter? |
| **Inputs** | Do the inputs already appear in your Work Brain? |
| **Catchable errors** | Can a person catch likely errors before they cause harm? |
| **Deployment readiness** | Can it ship without new procurement, IT or vendor review? |
| **Ownership** | Is there a durable, named owner for the outcome? |
| **Materiality** | Does it clear a meaningful value threshold for a team, value stream or system of record? |

Set a candidate aside when the work depends on judgment a person has to make, when regulation requires human review, or when the benefit can't justify the cost to build and run it. Work that needs near-perfect accuracy needs strong rule-based controls, and may not suit an agent.

> [!TIP]
>
> Carry more than one candidate into testing. A use case that worked elsewhere isn't proof that a similar one will work here.

## What you leave Define with

* **The objective**, with its baseline, target, constraints and owner.
* **The solution set**: the agreed future state, the mix of changes, and an owner for each.
* **The sequence**: what goes first, and why.
* **Approved investment** for the first wave.

That's the hand-off into Deliver. See [Execute your improvements](/user-docs/deliver/execute-your-improvements).

## Where to go next

> **[Strategic analysis with Advisor](/user-docs/define/strategic-analysis-with-advisor)**
>
> Reason across your whole Work Brain to find where the gap sits.

> **[Define your solution set](/user-docs/define/define-your-solution-set)**
>
> Choose the mix of changes that closes the gap, and who owns each.

> **[Advisor prompt library](/user-docs/define/advisor-prompt-library)**
>
> Ready-made prompts, including a set for defining your solution set.

> **[Execute your improvements](/user-docs/deliver/execute-your-improvements)**
>
> Take the solution set into delivery.

> [!NOTE]
>
> **Need help?** Use the in-app chat — click the chat bubble in the bottom-right corner of Within (staffed 24×5) — or email [support@within.ai](mailto:support@within.ai).


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.