Launching an AI Initiative
Purpose
A step-by-step scenario for registering a new AI initiative, performing its initial review, and preparing it to move from the "New" state to the "Assessment" state.
The playbook relies on the functionality of the AI Conveyor platform: manual initiative creation, creation via the AI assistant, checking for similar initiatives, product selection, required fields, tasks, and the business funnel.
Core Ideas
- An initiative starts not with technology, but with a business problem and an expected impact.
- The system should identify duplicates, the absence of an owner, weak value, and an unsuitable product as early as possible.
- The AI assistant is used not as a chat for chat's sake, but as a way to assemble a brief and fill out the card faster.
- Movement through the funnel must be managed: through rules, required fields, and a decision history.
How It Works
When to Use
The playbook applies when:
- an employee or department proposes a new idea;
- the AI office itself forms a set of initiatives based on the company profile;
- a manager wants to turn a scattered idea into a managed card;
- the AI assistant assembles a brief through dialogue;
- you need to understand whether an initiative is worth further assessment.
Roles
| Role | Responsibility |
|---|---|
| Initiator | Describes the problem, process, and expected impact |
| Business owner | Confirms the significance of the problem and is responsible for the future result |
| AI office | Verifies card completeness, duplicates, priority, and routing |
| Project manager | If needed, takes the initiative into work and drives it forward |
| Product owner | Confirms that the initiative fits the product |
| Security | Provides an early signal if there are obvious constraints |
Input Data
Minimum required for registration:
- initiative name;
- business problem;
- initiator;
- department or process;
- expected type of impact;
- a brief description of the proposed solution.
It is advisable to add right away:
- baseline process metrics;
- an urgency assessment;
- a possible product;
- data requirements;
- security constraints;
- the prospective business owner.
Method 1. Manual Creation
- The user opens the initiatives registry.
- Clicks to create an initiative.
- Fills in the main fields: name, problem, description, initiator, expected impact, urgency.
- If there is sufficient understanding, selects a product.
- Saves the card.
Result:
- the initiative is created in the "New" state;
- the card enters the portfolio;
- the AI office sees it in the business funnel;
- further movement proceeds through transition rules.
Method 2. Creation via the AI Assistant
This method is suitable when the user has an idea but no ready-made structure.
- The user describes the idea in free text.
- The AI assistant extracts the known fields.
- The assistant asks only the missing questions.
- The system checks for similar initiatives.
- If a catalog is available, the assistant suggests a suitable product.
- After confirmation, the initiative card is created.
What matters:
- if the product catalog is empty, the assistant must not invent nonexistent products;
- if similar initiatives are found, the user must see them before creation or promotion;
- if there is little data, the assistant continues assembling the brief rather than creating a weak card.
Method 3. Batch Generation of Initiatives
The AI office can form a starter set of initiatives based on the company profile: industry, company description, AI adoption maturity, and the chosen starting scenario.
This mode is useful:
- during the first launch of the platform;
- for a demo tenant;
- for companies that do not yet have a mature flow of ideas;
- for preparing a workshop with business units.
Generated initiatives must not be automatically considered high quality. They need to be reviewed, owners clarified, duplicates checked, and the expected impact confirmed.
Step-by-Step Process
Step 1. Capture the Problem
A poor formulation:
We need an AI-based assistant.
A good formulation:
Support staff spend up to 30 minutes preparing a response to routine requests because the information is scattered across instructions and emails.
Readiness criteria:
- the process is clear;
- the pain point is clear;
- you can name a metric that should change.
Step 2. Formulate the Solution Hypothesis
The hypothesis must connect the solution and the change in a metric.
Template:
If [solution] is implemented, then [metric] will change from [baseline value] to [target value] within [period].
If there is no baseline value yet, the card must explicitly state who will clarify it and when.
Step 3. Check for Similar Initiatives
Before moving to assessment, you need to determine whether a similar initiative already exists.
Possible decisions:
- continue as a new initiative;
- merge with an existing one;
- link the initiatives as dependent;
- reject as a duplicate.
Checking for similar initiatives is especially important before moving to delivery, because that is where team resources begin to be consumed.
Step 4. Preliminarily Select a Product
If the product is clear, it is worth specifying it already at launch. If not, the product is selected during the assessment stage.
Routing examples:
- searching documents and answering questions → knowledge assistant;
- forecasting or classification based on data → machine learning platform;
- automating a chain of actions → automation product;
- a software code development task → code agent;
- document analysis → document processing product.
Product selection must not be decorative: an AI product has its own delivery track, team, and set of stages.
Step 5. Assign Owners
At a minimum, you need:
- the initiator;
- the business owner;
- a responsible person from the AI office;
- a project manager if needed.
An initiative without an owner must not go far through the funnel. This is almost always future portfolio debt.
Step 6. Decide on Moving to Assessment
The decision is made based on minimal criteria:
- the problem is clear;
- the expected impact is described;
- there is an owner or a clear candidate;
- there is no obvious duplicate;
- there is no immediate security prohibition;
- the card is filled in sufficiently for assessment.
Possible decisions:
- move to assessment;
- return for clarification;
- merge with a similar initiative;
- reject;
- keep in the new funnel until an owner or data appears.
Playbook Output
After successful execution of the playbook, the initiative should have:
- a completed card;
- the "New" or "Assessment" state;
- a clear business problem;
- an initial impact hypothesis;
- an initiator and an owner;
- the result of checking for similar initiatives;
- a preliminary product or a task to select a product;
- a set of near-term tasks.
Next, the initiative moves to the use case validation playbook.
Common Mistakes
- Starting with the name of a technology rather than with the problem.
- Creating an initiative without an impact owner.
- Not capturing the baseline metric.
- Ignoring similar initiatives.
- Assigning a product formally, without a link to delivery.
- Moving an initiative to delivery before the security and duplicate checks.
- Writing the expected impact in generic terms: "speed-up," "convenience," "quality."
Minimum Platform Configuration
For initiative launch to work in a managed way, the platform must have the following configured:
- the business funnel and allowed transitions;
- required fields for a new initiative and the assessment stage;
- a rejection reason rule;
- a check for similar initiatives;
- a product catalog;
- permissions to create and modify initiatives;
- field visibility by stage;
- the roles of the AI office, project managers, and business users.