Short version: After scoring 172 business ideas on opportunity, revenue potential, and difficulty, the thing that separates the top quarter from the bottom quarter is not novelty and it is not AI. It is structure. The best ideas are services, cost under 10k to start, reach revenue in under three months, and bill recurring. The worst ones are marketplaces, physical or logistics plays, and capital-heavy stores that take half a year to see a dollar. Adding AI helps a little at the margin, but it does not rescue a badly shaped business.

Here is the full breakdown, with the real numbers, and an honest note on what these scores are and are not.

How we scored them

Every idea in our bank carries three scores from 1 to 10: an opportunity score, a revenue potential score, and a difficulty score. To rank ideas by a single number I used a simple composite: net = opportunity + revenue potential - difficulty. Higher is a better bet, all else equal. Across 172 ideas the net score averaged 7.7 and ranged from 4 to 13.

I then split the set into a top quarter (43 ideas, mean net 9.5) and a bottom quarter (43 ideas, mean net 5.8) and looked at how the two groups differ on the things you actually decide when you pick an idea: the business model, the startup cost, the time to first revenue, whether the revenue is recurring, and whether the idea leans on AI.

Be clear-eyed about what this is. These are judgment scores, not realized outcomes. Nobody has run all 172 businesses and measured who made money. The scores encode a prior, a pattern-informed bet, not proof. There is also a selection effect: all 172 already passed a vetting filter before they entered the bank, so even the bottom quarter is the least-good of the already-decent, not a pile of obviously terrible ideas. And the composite weights are a choice. Weighting difficulty more heavily, or discounting revenue for competition, would reshuffle the list. Read the patterns below as strong tendencies in our data, not laws of nature.

Finding 1: services win, marketplaces lose

The single biggest difference between the top and bottom quarters is the business model.

In the top quarter, 74 percent were services businesses (32 of 43). Marketplaces were 2 percent.
In the bottom quarter, services were 16 percent. Marketplaces (28 percent), ecommerce (19 percent), and physical or logistics plays (14 percent) made up 61 percent of the group.

Mean net score by model, across all 172:

Services: 8.4
Content: 8.0
SaaS: 7.3
Ecommerce: 7.1
Marketplace: 6.2
Physical: 6.1

The reason is not that services are glamorous. It is that services score low on difficulty. You can start one with a laptop, a niche, and a repeatable process. Marketplaces have the cold-start problem: you need supply and demand at the same time, in the same place, or you have nothing. Physical and logistics businesses fight thin margins and real-world operations. Those constraints show up as high difficulty scores, and difficulty is what drags the net down.

The five lowest-scoring ideas in the whole bank tell the story plainly: a one-product dropshipping store, a single-cuisine ghost kitchen, a same-day local delivery marketplace, a niche meal-prep delivery service, and a gamified kids chores app. Four of the five involve moving atoms or aggregating two sides of a market. None of them is a stupid idea. They are just hard in ways that a solo founder rarely wins.

Finding 2: cheap and fast beats capital and patient

The top and bottom quarters diverge sharply on money and time.

Startup cost. In the top quarter, 91 percent cost under 10k to start (40 percent under 1k). In the bottom quarter, 53 percent needed 10 to 50k and another 16 percent needed 50k or more. Mean net by cost band: under 1k scored 8.5, the 1 to 10k band 7.9, the 10 to 50k band 6.9, and 50k-plus scored 6.7.

Time to revenue. In the top quarter, 93 percent reached revenue in under three months (56 percent in under a month). In the bottom quarter, 72 percent took three months or longer to earn a dollar.

This is the least surprising finding and the most useful one. Cost and time to revenue are the two variables most under your control when you choose an idea, and they are the two that most reliably track with a better bet. Every extra month before revenue and every extra dollar of upfront cost is risk you are taking on before you know anything. The ideas that let you charge a real customer in week three are structurally safer, not because they are bigger, but because they fail cheaper and teach you faster.

Finding 3: recurring revenue helps, but less than you would think

Recurring revenue is genuinely better. In the top quarter, 84 percent of ideas bill recurring versus 70 percent in the bottom quarter, and across the full set recurring ideas out-net one-time ideas 7.8 to 7.3.

But notice the gap is small. Seventy percent of even the bottom-quarter ideas are recurring. Recurring revenue has become table stakes in how ideas get pitched, which means it no longer separates the good from the bad on its own. A recurring marketplace is still a marketplace. Do not let a subscription model talk you out of noticing that the underlying business is hard.

Finding 4: AI is a tailwind, not the dividing line

This is the finding I expected to go the other way. 36 percent of the ideas in the bank are AI-centric. AI ideas do score better on average: mean net 8.3 versus 7.4 for non-AI, driven mostly by a higher opportunity score (6.9 versus 6.4). AI barely moves the difficulty score at all (5.4 versus 5.6).

And yet the top quarter is 51 percent non-AI. Plenty of the best-shaped ideas, a productized local Google Business Profile service, a niche paid B2B newsletter, a guest-review response service for independent hotels, use AI as an ingredient at most. Meanwhile, in the bottom quarter, only 16 percent are AI ideas, which tells you the opposite of what the hype implies: sprinkling AI onto a structurally hard business (a delivery marketplace, a ghost kitchen) does not appear in our data as a rescue. It shows up as a hard business that also has to build AI.

The honest read: AI raises the ceiling on what a small team can deliver, so it nudges opportunity up. It does not change the physics of cold-start problems, capital intensity, or operational drag. If you are choosing between two ideas and one is AI, that is a mild point in its favor. If you are choosing an idea because it is AI, you are optimizing the wrong variable.

The profile of a good idea, from the data

Put the four findings together and the top-quarter idea has a recognizable shape:

It is a service, often productized so it sells like a product.
It costs under 10k to start, frequently under 1k.
It reaches revenue in under three months.
It bills recurring.
It may or may not use AI, and that is fine either way.

The highest-net idea in the entire bank fits it exactly: an AI-assisted WhatsApp sales and support service for small businesses in African markets. Opportunity 9, revenue 7, difficulty 3. It is a service, cheap to start, fast to revenue, recurring, with AI as leverage rather than the whole point. The point is not that idea specifically. The point is the shape.

What this means if you are picking an idea

You do not need our scores to use this. When you look at your own idea, ask the four structural questions before you fall in love with the concept:

Is it a service, or does it need a two-sided market or a warehouse? If it is the latter, your difficulty just went up a lot. Proceed with your eyes open.
Can you start it for under 10k? If not, what exactly does the extra capital buy, and can you validate before you spend it?
Can you charge a real customer within three months? If revenue is six months out, you are betting on your patience and your runway, not just your idea.
Does it bill again next month without a new sale? Nice to have, but do not let it paper over a bad answer to questions one through three.

If you want the fuller version of this, we turned the same logic into a free idea validation scorecard that scores your own idea on the dimensions above, and you can browse the full set of vetted ideas with their scores to see the pattern for yourself.

Methodology note

Numbers come from 172 business ideas in our bank as of July 2026. Each has opportunity, revenue potential, and difficulty scores on a 1 to 10 scale, plus categorical fields for business model, startup cost band, time to revenue, revenue type, and whether the idea is AI-centric. The composite net score is opportunity plus revenue minus difficulty. Top and bottom quarters are the 43 highest and 43 lowest by net. All percentages are computed directly from the data, not estimated. The scores themselves are our editorial judgment, informed by market patterns, not measured business outcomes, and every idea in the set had already passed an initial vetting filter, so the bottom quarter is relatively weak, not absolutely bad. Treat the findings as strong tendencies in a curated dataset, useful as a prior when you evaluate your own idea, not as a guarantee.

I analyzed 172 vetted business ideas. Here is what separates the good ones from the bad.