How to Use AI for Market Research (The Bottom-Up Method That Actually Convinces Investors)

The Tennis Ball Problem: Why Bottom-Up Beats Top-Down

There's a classic consulting interview question: how many tennis balls fit inside a Boeing 747?

The wrong approach is to Google "volume of a Boeing 747" and "volume of a tennis ball" and divide one by the other. You'll get a number. It'll be wrong. And you'll have no idea why it's wrong or how to check it.

The right approach is to think from the bottom up. How wide is a row of seats? About 6 metres. How many tennis balls fit across 6 metres? About 90. How many rows of seats? About 50. So one layer of the floor is roughly 90 x 50 = 4,500 balls. How tall is the cabin? About 2.5 metres. A tennis ball is about 7cm, so that's about 35 layers. 4,500 x 35 = roughly 157,000 tennis balls.

Is that exactly right? Probably not. But you can check every assumption. You can adjust each variable. You understand the logic. You can defend every step. That's the point.

Market sizing works exactly the same way.

Top-down approach (what bad founders do): "The global SaaS market is $200B. HR software is 8% of that = $16B. We target SMBs which is 30% = $4.8B. If we get 1% market share = $48M."

This tells an investor nothing. Where did 8% come from? Why 30% SMBs? Why would you get 1%? Every number is arbitrary. It's fiction disguised as math.

Bottom-up approach (what good founders do): "There are 2.1 million businesses in Australia with 5-50 employees. 340,000 of them have an HR function. Our product costs $200/month per business. 340,000 x $200 x 12 = $816M addressable market in Australia alone. We're starting with hospitality, which is 42,000 of those businesses = $100M serviceable market."

Every number is checkable. Every assumption is stated. An investor can challenge any variable and you can have an intelligent conversation about it. That's what convincing market research looks like.

How AI Changes the Market Research Game

Traditional market research meant one of two things: expensive reports from firms like Gartner and IBISWorld, or weeks of manual Googling and spreadsheet work. Both were slow. Both were limited. And both usually gave you top-down numbers that didn't help.

AI flips this completely.

Speed: hours instead of weeks

What used to take a junior analyst two weeks - collecting data, cross-referencing sources, building spreadsheets - AI can do in an afternoon. You can query AI tools with specific questions and get structured, sourced answers in minutes.

Depth: go wider and deeper simultaneously

The old trade-off was breadth vs depth. You could either look at many markets superficially or one market deeply. AI removes that trade-off. You can explore ten adjacent markets, drill into three promising ones, and build detailed bottom-up models for two - all in the same day.

The "teach me everything" approach

Here's my favourite AI market research workflow. Instead of asking AI a specific question, start broad:

1. "Teach me about [market] as if I know nothing." Let AI give you the full landscape. Players, trends, challenges, opportunities.
2. "What are the most interesting things about this market?" AI surfaces non-obvious insights that you wouldn't think to ask about.
3. "What are the blind spots I might have?" This is the gold question. AI will identify assumptions you didn't know you were making.
4. "Now help me size this market from the bottom up." With the context from steps 1-3, build a rigorous bottom-up model.

Each step builds on the previous one. By the time you're doing the sizing, you have deep context that makes your assumptions much more grounded.

The Bottom-Up Market Sizing Template

Here's the exact framework I use with founders. It works for any market, any product, any stage.

Step 1: Define your customer precisely

Not "small businesses." How small? What industry? What geography? What specific problem do they have that your product solves?

Example: "Hospitality businesses in Australia with 10-50 employees who currently manage staff rostering with spreadsheets or pen-and-paper."

Step 2: Count them

This is where AI is incredibly useful. Ask: "How many hospitality businesses are there in Australia with 10-50 employees?" Cross-reference with ABS data, industry reports, and business registries.

You're looking for a specific, defensible number. Not a range. Not an estimate from a report that aggregates 12 different industries. A count of actual businesses that match your customer definition.

Step 3: Determine your price

What will you charge? If you don't know yet, research competitors. If there are no competitors, talk to 20 potential customers and ask what they'd pay. Your price is not a guess - it's data from the market.

Step 4: Multiply

Number of customers x price per customer x 12 months = annual addressable market.

That's it. That's the entire model.

Step 5: Layer in the expansion

Now add the dimensions: - Geography: "Australia first, then expand to UK and US." Each region gets its own customer count and price point. - Product expansion: "We start with rostering, then add payroll, then add HR compliance." Each product adds revenue per customer. - Market growth: "The hospitality industry is growing at 3% annually, and the shift from paper to digital is accelerating."

Each layer is a separate, defensible multiplication. An investor can challenge any single assumption without the whole model collapsing.

What AI Tools to Use (And How)

You don't need expensive market research subscriptions to do rigorous bottom-up sizing. Here's what actually works.

Claude or ChatGPT for initial research

Start with broad questions, then go specific. AI is remarkable at synthesising information from multiple sources and giving you structured summaries. Use it for: - Understanding market dynamics and trends - Identifying competitor landscapes - Finding data sources you didn't know existed - Building initial sizing models - Stress-testing your assumptions

The "keep asking" technique

Don't stop at the first answer. The real insights come from going deeper: - "What's the source for that number?" - "What would make that estimate too high or too low?" - "What adjacent markets should I consider?" - "Who are the non-obvious competitors?" - "What happened to companies that tried this before?"

Each follow-up question peels back a layer. Five rounds of "tell me more" will give you more insight than any market report.

Your own knowledge base

If you've been in your industry for a while, you already have market intelligence trapped in your notes, emails, meeting transcripts, and conversations. Feed that context to AI. "Here are my notes from 30 customer conversations. What patterns do you see in terms of market size and willingness to pay?"

This is where having a personal knowledge base pays off massively. AI with context about your specific market, your specific customers, and your specific conversations will produce far better research than AI working from generic data.

Government data sources

In Australia, the ABS (Australian Bureau of Statistics) has incredibly detailed industry data that most founders never look at. In the US, it's the Census Bureau and BLS. AI can help you navigate these sources and extract the specific numbers you need for your bottom-up model.

LinkedIn and job boards

Want to know how many companies have a specific role? Search LinkedIn. Want to know what companies are hiring for? Check job boards. These are real-time market signals that traditional market reports miss. AI can help you structure and analyse this data at scale.

The Five Mistakes Founders Make With Market Research

After reviewing hundreds of pitch decks and market sizing exercises, the same mistakes come up constantly.

Mistake 1: The meaningless TAM

"The global market for X is $47 billion." Cool. You're a two-person startup in Melbourne. That number has nothing to do with your actual opportunity. Investors see through this instantly. It signals that you haven't done the real work.

Mistake 2: Over-researching

Some founders spend months on market research before talking to a single customer. This is procrastination disguised as diligence. At some point - and that point arrives earlier than most founders think - you need to stop researching and start selling. If you've done a solid bottom-up sizing and talked to 20 potential customers, you have enough. Go build.

Mistake 3: Paying for reports you don't need

IBISWorld reports are great for understanding industry structure. They're terrible for sizing your specific opportunity. Don't spend $500 on a report that tells you an industry is worth $12 billion when what you need to know is how many customers will pay $50/month for your product. AI can help you find that answer for free.

Mistake 4: Confusing interest with demand

"200 people signed up for our waitlist" is not market research. It's a signal. Market research is understanding how many of those 200 would actually pay, at what price, how frequently, and for how long. AI can help you design surveys and analyse responses that convert interest signals into demand data.

Mistake 5: Static sizing

Markets aren't static. They grow, shrink, shift, and fragment. Your market sizing should account for where the market is going, not just where it is today. AI is particularly good at identifying trends and projecting trajectories because it can process multiple data sources simultaneously.

A Real Example: Sizing a Market in 30 Minutes

Let me walk you through how I'd size a market for a hypothetical product, using AI, in under 30 minutes.

The product: An AI-powered rostering tool for cafes and restaurants.

Minute 0-5: Define the customer

Ask AI: "How many cafes and restaurants are there in Australia? Break it down by size - specifically, how many have 5-30 employees?" Cross-reference with ABS data. Let's say the answer is roughly 45,000 establishments.

Minute 5-10: Validate the problem

"What percentage of hospitality businesses still use manual or spreadsheet-based rostering?" AI will pull from industry surveys and reports. Let's say research suggests 60% of small hospitality businesses haven't adopted dedicated rostering software. That's 27,000 potential customers.

Minute 10-15: Price the solution

"What do existing rostering tools charge for businesses with 5-30 employees?" Research Deputy, Tanda, Humanforce. Average is roughly $4-8 per employee per month. For a 15-person business, that's $60-120/month. Let's use $80/month as our midpoint.

Minute 15-20: Build the model

27,000 businesses x $80/month x 12 = $25.9 million annual addressable market in Australia. For cafes and restaurants with 5-30 employees who don't currently use rostering software.

Minute 20-25: Layer in expansion

Add New Zealand (roughly 15% of Australia's market = $3.9M). Add upsell to payroll integration (doubles ARPU for 30% of customers = additional $6.2M). Total near-term opportunity: roughly $36M.

Minute 25-30: Stress test

"What would make this estimate too high?" Maybe adoption is harder than expected. "What would make it too low?" Maybe the product expands to retail, healthcare, aged care - each of which has similar sizing dynamics.

Total time: 30 minutes. Every number is defensible. Every assumption is stated. This is what you put in your pitch deck. Not "$47 billion global market." This.