The most expensive sentence in e-commerce is "we'll just test it with a small ad budget." Small tests compound. Fifty for creative, a couple hundred for a starter ad set, a sample order, an app subscription — and an idea that could have been disqualified at a desk has quietly cost four hundred dollars before producing its first honest data point.
Most bad product ideas can be killed in about an hour of desk research, using nothing but free, public sources. That is the entire point of validation: not to prove an idea is good, but to disqualify it as cheaply as possible. The ideas that survive the hour earn the right to cost you money.
This is the zero-budget pass — four checks, no card, no subscriptions. Are competitors paying to sell this right now? Is search interest moving the right way? Does Amazon prove people already buy it? Do communities talk about it like buyers or like spectators? Here is how to run each pass, and how to read the four answers together.
The order of operations
Free evidence first, money later. Written down it sounds obvious, but the common failure mode runs the other way: sellers spend first — on samples, ads, or tools — and only go looking for demand evidence once the money is committed and the sunk-cost pull has set in. By then the research is no longer a filter; it is a search for permission.
Run the passes in order of how decisive they are per minute. Competitor ads come first because they are the strongest single free signal: someone with their own money at risk has already concluded this product sells, and the ad library shows you their conviction in public. Search direction comes second because it takes ten minutes and stops you from boarding a trend on its way down.
Make one rule with yourself before you start. Nothing gets a cent — not a sample, not a domain — until the idea has survived all four passes. Budget one hour per idea. Most will not make it through, which is the system working, not failing.
Pass 1: competitor ads, 15 minutes
Open the Meta ad library — free, public, no account required. Search your product terms, filter to active ads in your target country, and count three things.
- Advertiser count. How many distinct businesses are paying to sell this right now — not how many ads, how many advertisers. Two or three is a quiet market; a dozen is a proven one; dozens upon dozens is a crowded one.
- Ad age. Check the start dates on the longest-running ads. The working field judgment: an ad that has run thirty days or more is almost certainly paying for itself, because nobody funds a losing ad for a month. Old ads are the closest thing to public revenue data you will get.
- Cluster size. Click through to each advertiser's full list of active ads. One or two creatives suggests a test; fifteen variations of the same product suggests a winner being scaled.
Read the combination, not any single number. No advertisers at all means either no market or missionary work you would fund alone. A handful of advertisers with month-old ads means validated demand with room to enter. A wall of advertisers all running large clusters means the demand is real and the auction is expensive. The full reading method — including how to judge creative angles and landing pages — is in the Meta ad library field guide.
Pass 2: search direction, 10 minutes
Open Google Trends and check the head term on two horizons: twelve months and five years. You are not reading volume — Trends shows relative interest, not absolute numbers. You are reading direction and shape.
The picture you want is a twelve-month line that is rising or holding, sitting on a five-year base that is stable or growing. A term falling on both horizons means you are arriving after the party. A jagged seasonal shape is fine, but note where you are on the curve — entering a Christmas-shaped market in January means carrying costs for ten months before the payoff window.
Then check the transactional modifiers: the head term plus "buy," "best," and "review." When those rise alongside the head term, curiosity is converting into intent. When the head term climbs but the modifiers stay flat, people are talking about the thing, not shopping for it. The traps in this data — and there are several — are covered in the Google Trends compass piece.
Pass 3: retail proof, 15 minutes
Search the product on Amazon and read the first page as a purchase record, because that is close to what it is. Reviews are an imperfect ledger — unverified and planted ones exist — but the great bulk of them sit on top of completed transactions, which makes this the hardest evidence in the whole pass.
Look at review depth first: a first page where several listings carry hundreds or thousands of reviews proves people buy this category at volume. Then check velocity — sort a few listings by most recent reviews and see whether they landed in the past few weeks or dried up two years ago. Depth without velocity is a market that was.
Finally, read the price spread. A tight cluster at the bottom of the range signals a commodity race where the cheapest listing wins. A wide spread where premium-priced listings still hold serious review counts signals that buyers will pay for positioning — which is where a new entrant without a cost advantage actually has a chance. The deeper method, including what rank movements do and do not mean, is in the retail proof guide.
Pass 4: community chatter, 20 minutes
The last pass is the slowest and the richest. Search Reddit for the product and the problem it solves; do the same on TikTok and X. You are listening for three things.
First, recurring complaints about existing options. When multiple unrelated threads gripe about the same flaw — it breaks, it leaks, the sizing runs small — you have found an angle a new product can own. Second, purchase language. "Just ordered one," "where do you buy this," and "is it worth it" are the words of buyers; long threads debating the concept in the abstract are the words of spectators. A market of spectators produces engagement and no orders.
Third, harvest the vocabulary. Write down the exact phrases real people use to describe the problem and the product — not your words, theirs. Those phrases become your search terms, your ad hooks, and your product page copy if the idea survives. Check recency as you go: a thread storm from two years ago is an archive, not a signal. The full listening method is in the listening posts field guide.
Reading the combined picture
You now hold four verdicts: paid competition, search direction, retail proof, community pull. Resist the urge to average them into a mushy "maybe." Use a plain decision rule instead.
- Three or four signals agree: proceed toward a small paid test. The idea has earned a budget.
- Exactly two agree: do not average — investigate the specific disagreement, because each combination means something. Ads without search often indicates an impulse product that lives on interruption, not intent. Search and chatter without retail proof can mean unmet demand, or it can mean a fulfillment problem nobody has solved. Name the contradiction before you decide.
- One or zero: archive it and move on. No amount of enthusiasm outvotes the evidence.
Keep a research log — a plain spreadsheet with the date, the idea, the four verdicts, and a one-line reason for the call. Dead ideas stay dead instead of resurfacing every few months wearing a new coat of optimism, and near-misses get a recheck date next quarter, when a flat trend may have turned. Memory is a terrible archive; a log costs nothing.
The one-scan version
For what it is worth, this four-pass hour is exactly what Kestrel automates: one query, and it checks live competitor ads in the Meta ad library, Google search demand, Amazon retail proof, and community chatter, then returns a 0–100 score — Hot, Promising, or Weak — with the evidence behind each signal laid out. The free tier includes twenty market scans with no card, which is enough to triage a full idea list in an afternoon; the report format is public at /specimen if you want to see exactly what it shows before running one. The manual method above is still worth learning either way — you will read any report better once you have gathered the evidence by hand a few times.
What free validation cannot prove
Be honest about the limit. All four passes describe other people's transactions — their ads, their searches, their reviews, their conversations. None of them proves the one thing your business depends on: willingness to pay your price for your offer. Desk research can tell you a market exists; it cannot tell you the market wants your version of it.
That proof requires the first small spend, and the escalation should stay small. A preorder page, or a modest ad test pointed at a landing page where you measure clicks and add-to-carts against real money — not a bulk inventory order, not a twelve-month app stack. The threshold for that spend is the decision rule above: only an idea with three or four agreeing signals has shown enough to justify it.
The zero-budget pass is not a substitute for the paid test. It is the filter that decides which ideas deserve one — and run honestly, whether by hand over an hour or through a scan in a minute, it means the money you eventually spend goes to the one idea in ten that survived the desk, instead of being split across the nine that never should have left it.
