The AI Canvas
The strategic framework for enterprise deep learning
Get an editable copy of the AI Canvas
Why enterprises need the AI Canvas
Every enterprise is wondering: “How should we use AI?”
We’ve talked to hundreds of companies about AI. Almost all recognize that AI represents both great opportunities and tremendous threats. Unfortunately most are still struggling to define specific use cases in their environment.
“We know we need machine learning but we’re not sure where to start.”
“We have a bunch of ideas but haven’t turned them into business plans.”
“How can we get started quickly?”
AI projects are hard to define
Unfortunately these questions have no easy answers.
Your business, your customers, and your data are totally unique. AI is still very new and we have only limited product patterns.
Additionally every enterprise has existing systems, business processes and policies — all which are difficult to change.
We need a 1-page plan
Alexander Osterwalder pioneered the 1-page canvas with the Business Model Canvas, a high-level strategy document designed to replace the business plan. Canvases have the following benefit:
- Easy to update
- Simple to understand
- Portable
After hundreds of conversations with CIOs, CEOs, and product teams we recognize the need for a canvas to plan enterprise AI projects.
How to use the AI Canvas
Use the canvas document and vet enterprise AI use cases. The canvas helps surface the key questions and feasibility challenges associated with building and deploying machine learning models in the enterprise.
The left half addresses business issues and the right addresses technical feasibility.
The AI Canvas supplements our book AI for Business Leaders. The next book release will be completed revised to include the canvas as the organizing strategy.
Business blocks
Opportunity — A high-level description of the business benefit for the AI models. Revenue growth, cost reduction, speed, etc.
Consumers — AI models produce results from input data sources. Consumers are the products, systems, and people who use the model results to deliver business value.
Strategy — Unique data assets provide the only ongoing sustainable advantage in AI products. Without a data differentiation maintaining a competitive moat will be challenging.
Policy & process — AI can present unique legal and policy questions. For instance, you may have to address model interpretability challenges or data rights issues.
Technical blocks
Solution — A high level description of the models, workflow and system architecture.
Data — Primary internal and external sources of data for model inputs. Consider accessibility, cleansing challenges, costs. The highest risk block on the canvas.
Transfer learning — The most technically challenging block of the canvas. Identify existing models, datasets or research papers the development team can use to accelerate deployment.
Success criteria — Model benchmarks (e.g. existing baseline performance) or necessary business metrics. Ideally quantified to compare with industry benchmarks.
Example — Image enhancement for e-commerce
HaoChi! is an “AirBnB for restaurants” in Chengdu, China. Amateur chefs can cook dinner for a few guests right out of their kitchen.
The problem low res and poor images limit sales
The chefs are great at cooking traditional Sichuan cuisine but not so great at marketing. Most use their xaiomi phones to take dish pictures. After cropping many are low-res versions which don’t meet consumer expectations.
Higher resolution pictures would boost sales:
HaoChi! explored the possibility of sending professional photographers to capture dish pictures but ultimately found it cost-prohibitive. The company now wants to explore the possibilities of using AI to automatically increase picture quality.
HaoChi!’s AI Canvas
HaoChi! spends several months exploring options and ultimately decides to explore a custom super-resolution algorithm to automatically boost dish quality.
Here is the company’s AI canvas:
Get an editable copy of the AI Canvas
