How I built a product solo using AI research
From Zero to World-Class AI Manager - part 6
A few months ago, I decided to conduct an experiment.
Could I use AI to create a product on my own that would have previously required a whole team?
The problem
Understanding how and when to monetise content is tough.
Not that there’s a shortage of tools that can help you do it. In fact, media and entertainment businesses are drowning in monetisation tools. Hundreds of vendors but no real framework for making decisions.
I wanted to help solve this problem.
The method
Three types of AI research, looped until the right product emerged.
1. Synthesise interview insights
I sat down with 15 people in the industry for wide-ranging conversations on what’s happening right now, how their businesses have responded, and their perspectives on what was likely to happen in the near future.
I fed the interview transcripts into ChatGPT:
I’ve conducted 15 interviews about subscriptions and advertising. [Attach transcripts]
Synthesise key patterns:
- What challenges come up repeatedly
- Where people disagree or have different approaches
- Gaps between what they want and what tools provide
- Non-obvious insights about what’s actually working
Format: Themes with supporting quotes, then strategic implications.
The analysis picked up on something I was hearing.
When it comes to the technology driving content and monetisation decisions, people’s problems weren’t necessarily about how to use specific products. The challenge seemed to be connecting tool capabilities - whether analytics, identity and access, or subscription management - to their business model and audience behaviour.
In other words, it wasn’t a tools problem. It was a decision-making problem.
People had plenty of options but no framework for figuring out which ones actually fit how their business worked and how their audience behaved.
That told me what I needed to research next.
(Examples of other uses for synthesising call transcripts to generate insights: L&D teams synthesising training feedback to redesign programmes. Product managers finding patterns across user research. Marketing analysing customer feedback.)
2. Track industry trends
I now had a sense of what was going on, but fifteen people is hardly a significant sample size. So, the next step was to zoom out and use AI Deep Research to generate analysis of the wider market.
Research: Current trends in media monetisation for content businesses
Focus on:
- Shifts in business models (what’s gaining/losing ground)
- New approaches people are testing
- Failed experiments and why they didn’t work
- Emerging patterns in audience behaviour
Prioritise recent news, operator posts, case studies from 2024-2025
This surfaced patterns I wouldn’t have found doing manual research alone - mainly because of the breadth of sources.
AI bot traffic was inflating dashboards while revenue stayed flat - metrics that looked like growth but weren’t
The biggest blocker wasn’t content or pricing - it was the infrastructure. When simple changes require engineering time, you’re maintaining systems, not running a business
Everyone talking about combining revenue streams while the underlying systems can’t share data
Cross-referencing these findings with my interviews showed what people were saying versus what was actually happening.
I had the big picture. Now I needed the details.
(Examples of other uses for Deep Research to validate customer insights: Product marketing teams tracking competitor positioning shifts. Strategists researching how others handled transitions you’re planning. Agency teams mapping client pain points to shape their pitch.)
3. Deep dive vendors
I now knew the trends and the problems. Time to research the actual vendors.
Deep Research: [Vendor name] - comprehensive profile
Include:
- Core capabilities vs marketing claims
- Pricing structure and hidden costs
- Integration complexity
- Who it actually works for (use cases, company size)
- Common complaints from users
Format: TL;DR, strengths/limitations, implementation reality check
I ran this for every major vendor whose name kept coming up in my research - companies like Adobe Analytics, Auth0, Google Analytics and Paddle.
I ended up with detailed research reports on each company, which I knew I needed to validate.
First with multiple AI platforms, cross-referencing outputs.
Then, importantly, I got real people back in the loop - asking for their experiences implementing the systems, building on the platforms, assessing the support they’d received along the way.
Now I had everything I needed to start building.
(Examples of other uses for Deep Research on specific products: Procurement teams comparing SaaS vendors before negotiations. HR leads assessing applicant tracking systems. Operations comparing logistics providers.)
The result
This process took me a few months - mainly because of the human conversations. But I know from experience it would have been impossible in the past without a full team working flat out.
In fact, until I worked with a developer to build the Monetization Works website - which I’d designed myself using Lovable, by the way - the only support I had was from AI.
Monetization Works is now live here.
You can find vendor profiles on companies like Stripe. And you can read comparisons of products in the same category like Recurly vs Zuora.
The profiles don’t read like feature lists because they’re not based on marketing material. They’re based on what people actually experience. Where tools fail. Which trade-offs matter.
And the added bonus - I designed the content so it would rank for searches people were actually making using LLMs like ChatGPT and Gemini.
Someone searching for “Stripe alternatives” or “Recurly vs Zuora” finds Monetization Works because the content answers strategic questions, not surface comparisons.
That’s Answer Engine Optimisation (AEO) - I built it that way from the start.
The research creates the product. The product creates its own distribution.
The point
Strategic research used to require a team because one person couldn’t move fast enough through the cycles.
AI compresses those cycles from weeks to days. You can operate solo at a level of depth that used to need multiple people.
But here’s the trap - AI will confidently tell you everything sounds great. It’ll validate every insight, support every hypothesis. It’s easy to build an entire product on research that sounds smart but doesn’t reflect reality.
The discipline is validation.
Checking with people who actually use these tools. Who’ve lived these problems. Who know where theory and reality diverge.
AI accelerates research. Humans tell you if it’s real.
Thanks for reading and good luck on your projects.
Ollie