Framing AI Projects
for ML Success

A structured approach to translate business ideas into machine learning projects using the Framing Canvas methodology.

Framing Canvas Structure

Business Goal

Why is this project helpful?

KPI

How do we measure success?

ML Output

What will ML deliver?

ML Input (Features)

What data is needed?

Data Sources

Where will data come from?

Success Metrics

Acceptance thresholds

The Challenge: Bridging the Business-ML Gap

Common Pitfalls

Vague Ideas

"We'll use internet data to predict what people will buy."

Undefined Objectives

"We'll use customer data to find the best ones."

Lack of Measurable Metrics

No clear definition of what "success" looks like

Essential AI Project Characteristics

Objective & Defined Output

Clear specification of what the AI will produce

Linked Datasets

Data connecting inputs to desired outputs

Measurable Success Metric

Quantifiable criteria for evaluating performance

Successful AI projects require translating business needs into ML-specific terms

Case Study: Home Price Predictor

Business Conversation

AI Leader:

"We tested a house-price predictor where users input property info and get immediate price estimates. People loved it. Now we want an AI to do it in seconds."

Data Scientist (John):

"What variable are we predicting? What features should we use? Do we have past transaction data?"

AI Leader:

"We're predicting sale price. We have expert broker insights on features and 6 years of transaction data (hundreds of thousands of records). Users accept predictions within +3% of actual price."

Project Framing

Business Goal

Immediate home price estimates

KPI

Prediction within +3% of actual price

ML Output

Predicted home sale price

ML Input (Features)

Square meters, bedrooms, location factors, neighborhood reputation

Data Sources

Past transaction records (6 years)

The Framing Canvas Methodology

Five-Step Framework

The Framing Canvas is designed to guide non-technical stakeholders through the process of converting business projects into well-defined ML projects.

1

Business Goal

Define why this project is helpful for your organization. What problem does it solve?

2

KPI

Establish how you will define and measure success. What quantitative metric will track ROI?

3

ML Output

Determine what ML will deliver to achieve your goal. What exactly will the algorithm produce?

4

ML Input (Features)

Identify what data the model needs. What information would you need to do the algorithm's job?

5

Data Sources

Locate where you will get this data. What systems or datasets contain this information?

Critical Success Factors

Establishing Acceptance Thresholds

For each KPI, identify a value that represents the minimum accepted performance.

Starting Point:

Measure current solution's performance on your KPI

New Products:

Run financial benefit calculations to determine needed performance

Understanding Error Implications

False Positives

Cost of incorrect predictions (e.g., pursuing unnecessary retention actions)

False Negatives

Cost of missed opportunities (e.g., failing to retain at-risk customers)

Framing Canvas in Practice

Reduce Customer Churn

Identifying customers about to cancel subscription services

Business Goal

Reduce subscription cancellations

KPI

Customer churn rate (monthly)

Success Metric

Recall > 90%

ML Output

Churn probability per customer

ML Input (Features)

Product usage, demographics, subscription type

Data Sources

CRM data, previous unsubscribed customers

Data Center Cooling Efficiency

Optimizing cooling systems to reduce energy consumption

Business Goal

Reduce energy costs

KPI

PUE (Power Usage Effectiveness)

Success Metric

Prediction within +3% of real PUE 95% of time

ML Output

Future performance of cooling systems

ML Input (Features)

Outside temperature, humidity, water pressure, data center load

Data Sources

Data center appliance logs, weather data

Small Business Loans

Offering next-day loans to eligible small businesses

Business Goal

Expand into small business loan market

KPI

% of eligible businesses identified

Success Metric

Precision > 95%

ML Output

Loan eligibility per customer

ML Input (Features)

Cash flow, business size, location, industry

Data Sources

CRM data, transaction data, financial markets data

The Framing Canvas produces clearly defined ML specifications
that can be directly implemented by technical teams

Key Insights & Benefits

Core Principles

1

Start with Business Value

Clearly articulate why the project matters to the organization before diving into technical details.

2

Quantify Success

Define measurable KPIs and establish acceptance thresholds upfront.

3

Consider Error Implications

Understand the business impact of false positives vs. false negatives.

Implementation Benefits

Alignment

Ensures business and technical teams share the same understanding of project goals.

Feasibility Assessment

Helps identify data gaps and technical challenges early in the process.

Efficient Development

Provides clear specifications that accelerate the ML development process.

The Framing Canvas bridges the gap between business vision and ML implementation

By translating AI projects into ML-friendly terms, organizations can significantly increase their success rate with AI initiatives.