There's a long list of ways to make money online. It is, in fact, almost comically long. Drop-shipping. Affiliate marketing. Online courses. Consulting. Freelancing. Software as a service. Stock photography. Newsletter sponsorships. Podcast advertising. Licensing. Royalties. NFTs, back when that was a conversation people had with straight faces.
The question I get asked most often, in different forms, is: why these particular streams? Why directories and POD and apps and this newsletter and whatever else we're testing? Why not YouTube? Why not e-commerce? Why not consulting?
The answer is a selection framework, and I think it's worth explaining in detail because the framework is as useful as any specific stream we identify.
Criterion One: The model has to already be proven.
This is the most important filter. I am not trying to invent a new category of business. I am not looking for the next big thing. I am looking for business models that have demonstrably produced income for people operating them — that the market has validated over time, ideally across multiple economic cycles.
Local directories exist because lead generation has always had value, and local businesses have always needed customers. Print-on-demand exists because branded merchandise has always sold and the production logistics problem is genuinely solved. Newsletters with paid subscriptions exist because people pay for information they trust and can't easily get elsewhere.
These are not new ideas. That is exactly the point. New ideas carry the risk of being wrong about the fundamental premise — that the market wants what you're building. Proven models have already cleared that bar. The question is whether you can execute them more efficiently than was previously possible.
Criterion Two: AI has to remove a meaningful bottleneck.
The second filter is identifying where, specifically, AI changes the economics. Because AI doesn't change the economics of every business model equally.
For local directories, the bottleneck was always content production and scale — you needed a team to build out hundreds of listings, produce enough SEO content to get traction, and keep everything current. AI handles all of that. The bottleneck shifts from production to relationship and sales, which is harder to automate but can be managed by a small team.
For print-on-demand, AI changes the design iteration speed. What used to require a freelance designer — or significant design skills of your own — can now be produced in hours with a good generative image workflow. Volume and variation are available at a fraction of the previous cost.
For content and newsletters, AI changes research speed, draft production, and editorial throughput. A single writer with good AI workflows can produce at a volume that previously required a small editorial team.
The key phrase is "meaningful bottleneck." If AI saves you twenty percent of the time on something that wasn't the limiting factor anyway, that's not a meaningful unlock. The unlock has to address the thing that was actually preventing a small operation from competing.
Criterion Three: The stream has to be able to trend near-passive.
I've written about this before but it bears repeating in the context of selection: not every proven business model trends near-passive. Consulting is proven. Consulting requires selling your time, which doesn't scale. Freelancing is proven. Same problem.
What I'm looking for is models where the work of building is genuinely front-loaded — where there's an asset being created (a content library, a domain authority, a product catalog, an audience) that continues to earn returns after the initial build is complete. The ongoing maintenance cost has to be low enough that one person or a small operation can maintain multiple streams without any single stream consuming full-time attention.
This is actually the hardest criterion to evaluate in advance, because the near-passive quality often doesn't reveal itself until you've been running the stream for a while. Early on, everything feels active. The question is whether the activity curve flattens as the asset matures.
What got filtered out.
Several models I considered didn't make it past these criteria. Consulting services are highly valuable and immediately income-generating, but they fail Criterion Three by design. Dropshipping physical products (beyond POD) has a proven model but the AI unlock is less definitive and the margin structure is harder. Freelance writing — yes, even with AI assistance — ultimately sells time.
The models we're testing have, in my assessment, cleared all three filters. I hold that assessment provisionally. We're testing precisely because I might be wrong about one or more of them.
The selection framework is in the free toolkit if you want to apply it to your own thinking.
This Week in AI: The concept of AI "fine-tuning" — training a model on specific data to improve its performance in a domain — has become increasingly accessible to non-engineers over the past year. What used to require a machine learning team is now achievable with off-the-shelf tools and a reasonably curated dataset. For businesses with proprietary knowledge, this is a meaningful capability to start thinking about.
The full stream selection framework — criteria, scoring rubric, and the models we ruled out — is in the free toolkit at start.tenstreamslab.com.