Choosing between two SaaS ideas without guessing
One idea researched alone always looks promising. Two researched identically produce a decision you can defend a month later.
The short answer: research both ideas to the same depth against the same five questions — evidence of spending, contentment with the current workaround, reachability of the audience, whether the problem is technical or behavioural, and how fast you could be proven wrong — then take the one that wins on evidence, breaking ties on time-to-disproof rather than build effort. The matching depth is the part that makes the comparison mean anything.
Sequential research is what most people do and it reliably produces the answer they started with. Idea A gets a fortnight of enthusiastic digging, looks promising, and gets built. Idea B is never examined, so there is no way to know it was the better one.
Why parallel beats sequential
A single idea has no reference point, and findings without a reference point are interpreted by whoever is holding them. “Some evidence of spending” reads as encouraging when it is the only number you have. Set beside an idea where the spending evidence is unmistakable, the same finding reads correctly: as weak.
Parallel research also surfaces something sequential research cannot — the problem that appears in both corpora without being either idea. That residue is often the best thing the exercise produces.
- ✗Idea A gets two weeks, idea B gets a paragraph
- ✗Findings judged against expectations, not alternatives
- ✗Sunk attention makes abandoning A expensive
- ✗Concludes: A is promising (it always does)
- ✓Both get the same source count and the same questions
- ✓Findings judged against each other
- ✓Dropping one is the expected outcome, not a failure
- ✓Concludes: B, for these three reasons
Matching the corpora is the whole discipline
A comparison is only as honest as its worst-matched half. If idea A got fifteen sources and idea B got six, you have compared research effort rather than ideas — and effort will track whichever one you already liked.
Practically, matching means the same number of sources for each, sampled the same way across channel sizes and dates, with the same independence checks applied. Twelve sources each is a reasonable working figure; the reasoning about saturation and when extra sources stop adding anything is in how many videos a research corpus actually needs, and it applies per idea, not across the pair.
The bias rarely shows up as a thumb on the scale during scoring. It shows up during collection — the favourite quietly gets the richer, longer, more on-point sources because you searched harder for it. Fix the source count and the search time per idea in advance.
The five dimensions worth scoring
Long scorecards create false precision. Five dimensions, each answered from the material rather than from judgement, do the work.
| Dimension | The question the corpus must answer | Strong looks like |
|---|---|---|
| Evidence of spending | Has anyone paid anything to relieve this? | Named tools, abandoned subscriptions, hired help |
| Workaround contentment | Do people resent their current method or accept it? | Visible frustration, not weary acceptance |
| Audience reachability | Do these people gather anywhere findable? | A named community, channel or event, not “small businesses” |
| Technical vs behavioural | Would software fix it, or would people have to change? | The fix is mechanical, not a new habit |
| Time-to-disproof | How fast could you learn you were wrong? | Weeks, with a test that does not need the full product |
The behavioural dimension is the one most often skipped and most often fatal. A problem that software could fix only if users adopted a new habit is not a software problem yet, and no amount of enthusiasm in the corpus changes that.
Score before you look at the totals
The order of operations matters more than the scale you use. Score each idea on each dimension while reading that dimension's evidence, and do not total anything until all ten cells are filled.
The reason is straightforward: once a running total exists, every subsequent judgement is made in relation to it. If idea A is ahead after three dimensions, the fourth gets read in the light of A already winning — not deliberately, but reliably. Filling the grid first and adding it up last costs nothing and removes the effect.
A three-point scale is enough. Strong, weak, or absent evidence, with the source count written beside each mark so the score can be interrogated later. Finer scales feel more rigorous and mostly encode confidence about distinctions the material does not support — the same false precision that makes conventional prioritization frameworks easy to bend, discussed in prioritizing a backlog with evidence instead of opinion.
Breaking a tie
Close scores are common and the instinct is to pick the easier build. Time-to-disproof is the better tiebreak: prefer the idea you could be proven wrong about fastest, even if it is more work to build, because being wrong quickly returns you to the decision with better information.
A useful reframing is to ask what test would settle each idea without building it — a landing page, a manual service run for three customers, a spreadsheet delivered by hand. The idea with the cheaper, faster such test wins ties, and the page for that test can be written straight out of the corpus you already built, as described in turning research into landing page copy.
What to do with the idea that lost
The losing corpus is the most commonly wasted asset in idea research. It gets closed, and six months later the same founder starts the same research again because the reasoning has faded to a feeling that the idea was somehow weaker.
Two things are worth keeping. The first is the decision record: which dimension the idea lost on, with the source counts that produced that verdict. Written down, it survives — and it means a later argument for reviving the idea has to engage with the specific finding rather than with a vague memory. The second is the expiry condition: the thing that, if it changed, would flip the decision. “Lost on workaround contentment — revisit if the free tier that covers this is withdrawn or priced” is a sentence that can actually be checked later.
That is what turns a rejected idea into a watchlist item rather than a dead end. Ideas lose on conditions, and conditions move — competitors withdraw free tiers, platforms change their terms, and a problem that was solved becomes unsolved again. Keeping the losing corpus as a standing project with a periodic re-run is cheap and occasionally very valuable, which is the argument in monitoring a niche with recurring research.
When both fail
Two failed passes is a cheap, good outcome, and the temptation to force a winner out of the wreckage is worth resisting. If neither idea shows spending evidence and both meet contented workarounds, promoting the less bad one is just a slower way to arrive at the same disappointment.
Read the residue instead. Twenty-four sources across two adjacent ideas usually contain a recurring frustration that belonged to neither — the thing people kept mentioning on the way to talking about something else. That pattern is the most common origin of a third idea worth its own pass, and it is the same signal that finding underserved niches through content gaps is built around.
What running two in parallel costs
The binding constraint is active projects rather than video volume, since each idea needs its own corpus kept separate — merging them destroys the comparison.
As of August 2026, two ideas fit the $19 Hobby plan, which carries 2 active projects and 25 videos a month; at twelve sources each that is a single full round with nothing spare. Pro at $59 carries 8 projects and 80 videos, which is the realistic shape for running two or three ideas at depth with room for follow-up passes. Studio at $199 carries 20 projects and 250 videos. The pricing page has the detail, and the manual-hours comparison is in what YouTube research actually costs.
Two projects, the same twelve-source depth on each, the same five questions — and a comparison you can still defend next quarter. 7-day free trial.
Closing thought
The value of comparing two ideas is not that you pick the better one, though you usually do. It is that the comparison makes your preference visible to you before it becomes a year of work — and a preference you can see is one you can argue with.
Frequently asked
How do you choose between two SaaS ideas objectively?
Research both to the same depth against the same questions, then compare on evidence rather than enthusiasm. The comparison only works if both corpora were built the same way — the same number of sources, the same independence checks, the same questions asked of each.
Why does researching one idea at a time produce bad decisions?
Because a single idea has no reference point. Every finding reads as encouraging or discouraging in isolation, and the founder's existing preference supplies the interpretation. Two ideas researched in parallel force a relative judgement, which is far harder to fool.
What are the comparison dimensions that actually matter?
Evidence of existing spending, how contented people are with the current workaround, how narrow and reachable the audience is, how much of the problem is technical rather than behavioural, and how quickly you could tell whether you were wrong.
Should the easier idea win when scores are close?
Usually yes — but the tiebreak worth using is time-to-disproof, not build effort. The idea you could be proven wrong about in three weeks is worth more than a marginally better idea that takes six months to test, because you get the next decision sooner.
Is it worth researching three or four ideas at once?
Three is often the practical maximum for one person. Beyond that the corpora get thin, and thin corpora produce comparisons that are really comparisons of which idea got the better sources. Depth per idea matters more than breadth across ideas.
What if both ideas fail the research?
That is a good outcome delivered cheaply. Two failed passes usually leave a residue — a problem that kept appearing in both corpora without being either idea — and that residue is frequently the better third idea.
What does running parallel research cost?
As of August 2026 the constraint is active projects, since each idea needs its own. Two ideas fit the $19 Hobby plan with 25 videos a month between them; the $59 Pro plan carries 8 projects and 80 videos, which is the comfortable shape for running two or three ideas at real depth.