Recommendation Canvas

What Is This?

Overview

The Recommendation Canvas is a structured evaluation framework designed to help product managers and technical leads assess AI product ideas before committing resources or presenting proposals to stakeholders. It organizes the analysis into distinct dimensions: business outcomes, customer outcomes, problem framing, solution hypotheses, positioning, risks, and value justification. By working through each dimension systematically, teams avoid the common trap of championing a solution before fully understanding the problem it is meant to solve.

This canvas is particularly valuable in the context of AI products, where the gap between technical capability and genuine business value is often wider than it appears. Many AI initiatives fail not because the technology is flawed, but because the recommendation lacked rigor. The canvas forces the evaluator to articulate hypotheses explicitly, identify assumptions that need validation, and surface risks before they become costly surprises.

The output of a completed canvas is a defensible, stakeholder-ready recommendation. Rather than a loose collection of slides or notes, the canvas produces a coherent argument that connects customer pain to business value, and solution design to measurable outcomes.

Who Should Use This

• Product managers preparing investment proposals for AI features or standalone AI products
• Technical leads who need to communicate the strategic rationale behind a proposed solution
• Product designers evaluating whether an AI-driven interaction model genuinely improves user experience
• Business analysts tasked with comparing multiple AI solution options against organizational goals
• Startup founders building a pitch that must demonstrate both technical and commercial viability
• Innovation teams inside enterprises who need to justify exploratory AI projects to budget holders

Why Use It?

Problems It Solves

• Prevents premature solution commitment by requiring explicit problem framing before solution design begins
• Eliminates vague value claims by demanding specific, measurable business and customer outcomes
• Reduces stakeholder rejection by producing a structured, evidence-backed recommendation rather than an opinion
• Surfaces hidden risks early, including data availability issues, regulatory concerns, and adoption barriers
• Aligns cross-functional teams around a shared understanding of what success looks like

Core Highlights

• Covers seven evaluation dimensions in a single structured document
• Separates business outcomes from customer outcomes to avoid conflating organizational goals with user needs
• Requires explicit hypothesis statements that can be tested and validated
• Includes a dedicated positioning section to clarify competitive differentiation
• Prompts risk identification across technical, ethical, and market dimensions
• Supports iterative refinement as new information becomes available
• Produces a recommendation artifact that can be version-controlled and reviewed over time
• Scales from quick evaluations to deep due-diligence exercises depending on the stakes involved

How to Use It?

Basic Usage

Start by filling in each section of the canvas as a plain text or markdown document. A minimal canvas entry might look like this:

## Recommendation Canvas:

AI-Powered Support Triage

### Business Outcome
Reduce support ticket resolution time by 30% within two quarters.

### Customer Outcome
Users receive accurate first-response guidance without waiting for a human agent.

### Problem Framing
Support agents spend 60% of their time on Tier 1 queries that follow predictable patterns.

### Solution Hypothesis
An LLM-based triage layer can classify and respond to Tier 1 queries with sufficient accuracy
to deflect them before human review.

### Positioning
Unlike rule-based chatbots, this solution handles natural language variation without
requiring manual rule updates.

### Risks
- Model hallucination on edge-case queries
- User distrust if responses are perceived as generic
- Data privacy constraints on ticket content

### Value Justification
Deflecting 40% of Tier 1 tickets saves approximately 1,200 agent-hours per quarter.

Specific Scenarios

Scenario 1: Evaluating a build-versus-buy decision. Use the canvas to compare an internally built model against a third-party API. Fill in separate hypothesis and risk sections for each option, then use the value justification section to quantify the cost difference.

Scenario 2: Preparing for a product review meeting. Complete the canvas in advance and share it as a pre-read. Stakeholders arrive with context, and the meeting focuses on decisions rather than explanations.

Real-World Examples

A fintech team used the canvas to evaluate an AI-driven fraud detection feature. The risk section revealed a regulatory gap that would have delayed launch by three months if discovered later. A SaaS company used it to reject an AI recommendation engine after the customer outcome section exposed that users had not actually requested personalization.

Important Notes

Requirements

• A clearly defined AI product idea or initiative to evaluate
• Access to basic business metrics to populate the value justification section
• Stakeholder input or customer research to inform the customer outcome and problem framing sections
• A markdown editor or document tool capable of rendering structured text