The Business Model Canvas

0:00 The Business Model Canvas

Can you explain your whole business on one page? Not a fifty-page plan that nobody reads, but one page where every part connects to every other part. That's what the Business Model Canvas does. By the end of this lecture, you'll be able to fill in all nine blocks, check them for coherence, spot the guesses that could break your model, and update the canvas for an AI-native business, where some of the costs and resources look very different from a traditional one.

0:36 Why a canvas?

Why use a canvas at all? Because a business model is a system, and systems fail at the joins. A brilliant product with the wrong channel fails. A loved service with the wrong revenue model fails. The canvas, created by Alexander Osterwalder and popularised in the book Business Model Generation, forces you to see the joins. It's also fast to change, which matters because early models change weekly. Here's the key idea. The canvas isn't a document you finish. It's a map of guesses you update as evidence arrives.

1:15 Nine blocks, one restaurant

Let's walk through the nine blocks with an analogy. Think of a restaurant. The right side of the canvas is the dining room: what customers see. Customer segments are who eats there. Value propositions are why they choose you. Channels are how they find you and get served. Customer relationships are how you treat them, from self-service to a personal waiter. Revenue streams are how they pay. The left side is the kitchen: what makes it possible. Key activities are the cooking. Key resources are the chefs, recipes and ovens. Key partners are your suppliers. And the cost structure is what all of that costs to run.

2:02 How to fill it

Now the order matters. Start with customer segments and value propositions, using real quotes from your interviews. Then channels, relationships and revenue. Then the kitchen. Put one idea per sticky note, and here's a habit I strongly recommend. Mark every note E for evidence or G for guess. On most early canvases, the G notes outnumber the E notes by a lot. That's fine. It shows you exactly what to test. Next, check coherence. Does each channel actually reach the segment you named? Does the revenue model match how that segment likes to buy? Do the key activities deliver the value proposition? Every no is a problem you've found for free.

2:50 Worked example 1: a cake business

A simple worked example. A home baker in Birmingham, let's call her Priya, sells celebration cakes. Segment: parents planning children's birthdays within five miles. Value: custom themed cakes, delivered on the day, no stress. Channels: Instagram and word of mouth at school gates. Relationship: personal, via direct messages. Revenue: per cake, with a deposit on ordering. Key activities: baking, decorating, delivery. Resources: her kitchen, her skill, her car. Partners: an ingredients wholesaler. Costs: ingredients, packaging, fuel, her time. Now the coherence check. Her Instagram posts get likes from other bakers, not from local parents. The channel doesn't reach the segment. Easy fix: school newsletters and local parent groups.

3:37 AI-native canvas changes

Now the AI-native version, because several blocks change. Key resources now include model access, your prompts and evaluation sets, proprietary data and human reviewers. Ask yourself which of these is truly yours, and which a competitor could rent tomorrow. Key partners include model providers and cloud hosts. What happens if one changes its price, terms or behaviour? Key activities include evaluation, monitoring and human review. Someone has to own quality every week. And the cost structure changes shape. Model usage is a variable cost, paid per task, often per token. It behaves like ingredients in a restaurant, not like rent. The more customers use you, the more it costs.

4:24 Worked example 2: two canvases (illustrative)

Here's the realistic scenario, with illustrative details. A team in Lahore wants to help small shops keep stock records. Canvas A is a classic SaaS app: a dashboard, sold by monthly subscription, marketed through ads. Canvas B is an AI-assisted WhatsApp service: the shopkeeper sends a photo of a delivery note or a voice message, and receives a weekly stock summary. Channel: wholesaler partners who already visit the shops. Revenue: a low monthly fee, paid by mobile wallet or bank transfer. Canvas B fits how owner-run shops actually behave. But look at the cost block. Every photo and voice note costs money to process, and a human checks unusual entries. So the next experiment is obvious: measure messages per shop per week and the true cost per summary.

5:20 Watch me: AI as canvas critic

Watch me do it. I've typed canvas B into a simple text list, block by block, with E or G next to each note. Now I paste it into an AI assistant and ask it to act as a critical reviewer. Check coherence between blocks. Flag missing costs, especially variable costs per customer. List the three guesses that would break the model if wrong. And suggest one alternative canvas for the same segment. I also tell it not to invent facts about my market, and to ask me questions instead. It flags that I haven't costed human review time, and asks whether wholesalers expect a commission. Both fair. I add them as G notes and turn the top three into experiment cards.

6:13 Using the canvas weekly

How do you actually use the canvas week to week? Keep it somewhere visible, like a shared board or a pinned page in Notion, and review it in a short weekly meeting. Three questions each time. What did we learn this week, and which notes move from G to E? Which note turned out to be wrong, and what does that change elsewhere on the canvas? And what's the riskiest remaining guess, which becomes next week's experiment? Keep old versions, too. Save a copy each time a block changes materially. When you look back after three months, you'll see your pivots clearly, and you'll be able to explain them to a co-founder, an adviser or an investor with evidence rather than hindsight.

7:06 Common mistakes

You'll also meet the Lean Canvas, Ash Maurya's variant for early startups. It swaps some blocks for problem, solution, key metrics and unfair advantage. Use whichever makes you think harder. Now the common mistakes. Filling it in once and never updating it. Writing everyone as the customer segment. Listing activities that don't deliver the value proposition. Treating guesses as facts, which is why the E and G marks help. And for AI businesses, putting model costs under fixed IT costs, which hides the fact that your margin shrinks as usage grows. Finally, confusing a feature list with a value proposition. Customers buy outcomes, not features.

7:52 Recap and try this now

Let's recap. The canvas shows your whole business on one page, split into the dining room customers see and the kitchen that makes it work. Start with segments and value. Mark every note as evidence or guess. Check the joins between blocks. For AI-native businesses, update resources, partners, activities and especially costs, because model usage is variable. Use an AI assistant as a critic, not a source of facts. Your try this now: fill in a canvas for your idea in forty-five minutes, mark E and G on every note, and pick the three guesses that would break it. Next, we'll sharpen the value proposition itself.