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

# AI Suggestions

> How Alan suggests t-shirt sizes and time estimates for quotes

Alan automatically analyzes quote requirements and suggests appropriate t-shirt sizes based on the brief, client history, and performance scores.

## How AI Suggestions Work

When a quote is created (either by a client or internally), Alan analyzes each requirement and suggests:

1. **T-Shirt Size**: The recommended size (XS through XXL)
2. **Confidence Level**: How certain Alan is about the suggestion
3. **Reasoning**: Explanation of why this size was suggested

<Info>
  AI suggestions are generated automatically when requirements are created. They appear in the quote details for team members to review.
</Info>

## What Alan Considers

### Brief Content

Alan analyzes the requirement details:

* Goal and expected outcome
* Current behaviour (for development)
* Technical constraints
* Design types and target audience (for design)
* Timeline requirements

### Historical Data

Alan reviews similar past tasks for this client:

* Previous task sizes and outcomes
* Actual vs quoted time ratios
* Patterns in requirement types

### Client Scores

Alan factors in client-specific performance metrics:

<CardGroup cols={2}>
  <Card title="Deliverability Score" icon="chart-line">
    How development/design tasks perform against estimates. A score of 1.15 means work typically takes 15% longer.
  </Card>

  <Card title="QA Score" icon="check-double">
    How many QA cycles the client typically needs. A score of 1.8 means \~80% more QA overhead than average.
  </Card>
</CardGroup>

### QA Impact on Suggestions

For development requirements, high QA scores may influence sizing:

* **QA Score > 1.5**: Alan may recommend bumping up the size
* **Reasoning**: Explains if QA overhead influenced the suggestion
* **Context**: First-pass rate and average cycles shown in prompt

<Tip>
  If Alan suggests a larger size "due to high QA cycles", consider whether the requirement clarity could be improved to reduce QA overhead.
</Tip>

## Understanding Confidence Levels

| Confidence | Meaning                                           | Recommendation                        |
| ---------- | ------------------------------------------------- | ------------------------------------- |
| **Low**    | Limited information in brief or few similar tasks | Review carefully, may need adjustment |
| **Medium** | Reasonable information and some historical data   | Good starting point, verify details   |
| **High**   | Clear scope and good historical data              | High confidence in suggestion         |

## Viewing AI Suggestions

### In Quotes

AI suggestions appear in the requirement details:

* Suggested size badge next to the requirement
* Click to see reasoning and confidence
* Team can accept or override the suggestion

### Reasoning Examples

<Tabs>
  <Tab title="Standard Suggestion">
    **Suggested: Medium (3-6 hours)**

    *Based on the scope described (custom product filter), this aligns with similar filtering features we've built. Historical data shows Medium tasks for this client typically take 4.2 hours.*
  </Tab>

  <Tab title="QA-Influenced">
    **Suggested: Large (6-12 hours)**

    *While the scope suggests Medium, this client has high QA cycles (1.8x average). Recommending Large to accommodate expected QA overhead. First-pass rate is 40%.*
  </Tab>

  <Tab title="Low Confidence">
    **Suggested: Medium (3-6 hours) - Low Confidence**

    *Limited information in the brief. Suggestion based on work type (bug fix) and general patterns. Recommend clarifying scope before finalizing.*
  </Tab>
</Tabs>

## Scope Review (Check with Alan)

When you write scope manually (rather than using "Generate with Alan"), you can ask Alan to review it before finalising.

### When the button appears

The **Check with Alan** button appears in the scope editor header when:

* You are editing a requirement
* There is scope content in the editor
* The scope was written manually (not AI-generated)

If you used "Generate with Alan" to create the scope, the review button is hidden since the scope was already AI-authored.

### What the review checks

Alan evaluates your scope against the client brief on three dimensions:

| Dimension           | What it measures                                                                                 |
| ------------------- | ------------------------------------------------------------------------------------------------ |
| **Clarity**         | Is the scope clearly written? Would a client understand what they're getting?                    |
| **Completeness**    | Does the scope cover all aspects of the brief? Are deliverables and acceptance criteria present? |
| **Brief Alignment** | Does the proposed work match what the client actually asked for?                                 |

Each dimension is scored 1-10 with specific feedback.

### Understanding the assessment levels

| Assessment            | Meaning                                                                   |
| --------------------- | ------------------------------------------------------------------------- |
| **Good**              | All scores 7+ with no high-priority suggestions                           |
| **Needs Improvement** | Some scores 5-6, or has high-priority suggestions to address              |
| **Major Issues**      | Critical gaps — scores below 5 or significant misalignment with the brief |

### Applying suggestions

If the review finds areas for improvement, an **Apply Suggestions** button appears. Clicking it rewrites the scope using AI, incorporating the review feedback while preserving your original intent.

After applying, an **Undo** button appears so you can revert to your original scope if needed.

<Tip>
  "Check with Alan" is a useful learning tool for less experienced team members. It helps build familiarity with what makes a strong scope document — clear deliverables, acceptance criteria, and alignment with the client brief.
</Tip>

## Pre-Calculated Subtask Suggestions

When a quote is approved and converted to a task, AI suggestions are pre-calculated for each subtask:

### What Gets Calculated

* **Development subtasks**: Adjusted by client's deliverability score
* **Design subtasks**: Adjusted by client's design score
* **QA subtasks**: Adjusted by client's QA score

### How It Works

1. Quote is approved and task is created
2. Subtasks are generated with base estimates (from t-shirt split)
3. AI suggestions are calculated for each subtask
4. Suggestions are stored and displayed in schedule dialog

### In the Schedule Dialog

When scheduling a subtask:

* **"Alan suggests: Xh Ym"** appears below the time input
* **Apply button** to accept the suggestion
* **Basis explanation**: Why this time was suggested

<Info>
  Suggestions are pre-calculated at task creation, so there's no loading delay when opening the schedule dialog.
</Info>

## Score-Based Adjustments

### Development/Design Work

Adjusted estimate = Base estimate x Deliverability Score

Example for Medium task (4.5 hours base), client score 1.15:

* Suggested: 5h 15m (4.5 x 1.15 = 5.175, rounded to 15 min)
* Basis: "Based on client history (+15% over quoted)"

### QA Subtasks

Adjusted estimate = QA allocation x QA Score

Example for QA review (30 min base), QA score 1.8:

* Suggested: 54m (30 x 1.8 = 54)
* Basis: "Adjusted for high QA cycles (1.8x avg)"

## Admin-Created Quotes

AI suggestions are now generated for both:

* **Client-submitted quotes**: Via client portal
* **Admin-created quotes**: Via internal admin panel

This ensures all quotes get consistent AI analysis regardless of origin.

## Best Practices

<CardGroup cols={2}>
  <Card title="Review Reasoning" icon="magnifying-glass">
    Always read the reasoning - it explains what influenced the suggestion
  </Card>

  <Card title="Check Confidence" icon="gauge">
    Low confidence suggestions may need manual adjustment
  </Card>

  <Card title="Consider Context" icon="brain">
    AI can't see everything - use judgment for unique situations
  </Card>

  <Card title="Improve Briefs" icon="pen">
    Better briefs lead to better suggestions - add detail when possible
  </Card>
</CardGroup>

## Related Documentation

<CardGroup cols={2}>
  <Card title="Client Intelligence" icon="brain" href="/user-guide/clients/client-intelligence">
    Understanding deliverability scores
  </Card>

  <Card title="QA Scoring" icon="check-double" href="/user-guide/clients/qa-scoring">
    How QA scores affect suggestions
  </Card>

  <Card title="T-Shirt Sizing" icon="ruler" href="/concepts/estimation/t-shirt-sizing">
    Understanding size ranges and estimates
  </Card>

  <Card title="Subtasks" icon="list-check" href="/user-guide/tasks/subtasks">
    Subtask scheduling and time estimates
  </Card>
</CardGroup>