# Verdict vs Evidence: What Can Desk Research Actually Prove? Canonical: https://signalcraftinsights.tech/blog/verdict-vs-evidence-what-can-desk-research-actually-prove Written by: D. Warnock Last reviewed: 2026-11-03 > Desk research can reveal problems, alternatives, demand behaviour and commercial constraints—but it cannot predict whether a startup will succeed. AI idea validators often end with a verdict: build, pivot, validate further or stop. The verdict is memorable. It is also the least durable part of the report. A founder can act responsibly only when the report shows what the evidence actually supports, where the inference begins and which commercial questions remain unanswered. That distinction matters whether the research is produced by a consultant, an analyst or an AI idea validation tool. This guide explains what desk research can prove, what it can only suggest and what still requires direct customer and market testing. ## The short answer Desk research can establish that relevant public evidence exists. It can show recurring complaints, common workarounds, active alternatives, competitor positioning, visible price points, purchasing language, regulatory constraints and gaps in current products. It can also show that evidence is thin, contradictory, old, promotional or concentrated in one community. Desk research cannot prove that your specific product will acquire customers at an acceptable cost, survive procurement, deliver the promised outcome, retain users or become profitable. Those claims require observed behaviour in your target segment. That is why the most useful question is not “Did the tool approve my idea?” It is: **Which claim is supported strongly enough to justify the next investment?** ## Evidence and verdict are different objects Evidence is an observation that another person can inspect. A customer describes a costly workaround. A buyer posts a request for proposals. A team pays for several overlapping tools. Reviews repeatedly mention the same missing capability. A procurement guide lists a mandatory security control. A verdict is an interpretation of those observations. “Attractive opportunity” and “weak idea” are not evidence. They combine assumptions about severity, frequency, market size, competition, acquisition and execution. Two founders can look at the same evidence and reach different commercial decisions. One may possess distribution, domain expertise or an integration that changes the economics. The other may face high switching costs and no credible route to market. A good market research report preserves that distinction. It links the observation, explains the interpretation and states the uncertainty. ## What desk research can establish with reasonable confidence ### 1. The problem is discussed in recognisable terms If people independently describe the same workflow failure, consequence or workaround, you have evidence that the problem exists. Stronger evidence includes dates, context, frequency, cost and the action taken. The claim should remain narrow. Ten recent complaints from finance teams about manual reconciliation support “some finance teams experience this workflow problem.” They do not prove that every small business needs a new platform. Use the [Evidence Quality Scorecard](/free-tools/evidence-quality-scorecard) to examine recency, independence, behavioural strength, commercial relevance and corroboration. ### 2. Customers already use alternatives Competitors, spreadsheets, agencies, internal staff and doing nothing are all alternatives. Their presence shows that customers have a current response to the problem. Reviews and implementation discussions can reveal why that response persists and where it fails. Competition is not automatically negative. It may show that buyers understand the category and allocate budget. The commercial question is whether your offer creates a meaningful reason to switch. ### 3. Buying constraints are visible Public material can expose security expectations, integration dependencies, legal requirements, procurement processes, budget language and implementation concerns. This is especially valuable when you need to [validate a B2B SaaS idea](/validate-b2b-saas-idea). You may discover that an apparently simple application touches sensitive data, requires single sign-on or must integrate with a system of record. Those findings do not prove the idea is bad. They change the minimum credible product and the sales motion. ### 4. Some demand behaviours are observable Active vendor searches, requests for recommendations, paid workarounds, deposits, purchases and renewals are stronger than likes or compliments. They involve time, money, switching or reputation. The [Willingness-to-Pay Evidence Ladder](/free-tools/willingness-to-pay-evidence-ladder) helps keep each claim proportional to the behaviour observed. A waitlist supports interest. A paid pilot supports a stronger claim. Renewal supports a stronger claim again. ### 5. Important evidence is missing Absence is not proof that demand does not exist. It is still useful when reported correctly. If direct complaints are scarce, sources are outdated or spending evidence cannot be found, the confidence in a broad market claim should fall. The appropriate response may be a targeted interview, concierge test or paid pilot—not an automatic rejection. ## What desk research cannot prove ### Product–market fit Product–market fit is an outcome of a specific product serving a specific segment through a specific go-to-market system. Public conversations about the problem cannot establish retention, expansion or durable customer value. ### Willingness to pay your price Competitor pricing and reported budgets create a reference point. They do not prove that your buyer will pay your price for your implementation, brand, risk profile and outcome. Test price through a real offer. ### Acquisition cost and sales velocity Search volume and community activity do not reveal whether you can reach qualified buyers efficiently. A market can be real and still be inaccessible to a founder without trusted distribution. ### Procurement approval A champion may want the product while security, legal, finance or procurement blocks it. For B2B products, use the [Buying Committee Mapper](/free-tools/b2b-buying-committee-mapper) and speak to the people who control approval. ### Retention and expansion Only observed use over time can show whether the product becomes part of the workflow. Even a successful paid pilot does not guarantee renewal. ### Startup success Success depends on execution, timing, cash, distribution, pricing, competition and learning speed. No desk-research process can convert evidence into a reliable probability that a startup will succeed. ## A useful evidence hierarchy A practical hierarchy moves from statements toward costly behaviour: 1. Compliments and hypothetical interest. 2. Repeated, specific descriptions of a problem. 3. Existing workarounds and time spent. 4. Active search, stakeholder involvement or data sharing. 5. Budget confirmation and an approval path. 6. Deposit, paid discovery or paid pilot. 7. Purchase and implementation. 8. Renewal, expansion or repeat purchase. Each step supports a stronger claim, but none removes the need to inspect source quality and context. A paid pilot funded as an innovation experiment may not represent normal procurement. A renewal caused by a contractual lock-in may not demonstrate satisfaction. ## How to assess AI idea validator accuracy “AI idea validator accuracy” is often framed as whether the tool chose the eventual winners. That is usually the wrong test. A more defensible evaluation asks: - Did the tool retrieve relevant, current and independent sources? - Can the founder inspect those sources? - Did it distinguish direct observations from model-generated interpretation? - Did it reveal contradictory and missing evidence? - Did it avoid turning confidence into a probability? - Did the recommended next test address the riskiest assumption? - Would another analyst be able to reproduce the reasoning? Outcome prediction would require a defined cohort, a time window, a clear success outcome and follow-up data. Platform interactions, generated landing pages or high idea scores are not substitutes for customer outcomes. The [SignalCraft Research Data Standard](/research-data-standard) explains the minimum disclosure needed for aggregate validation claims. ## Turn research into a decision, not a permission slip A useful report should end with a decision boundary. For example: **Supported:** finance teams publicly describe costly reconciliation work and pay for fragmented alternatives. **Not established:** whether mid-market controllers will replace an existing system with this product. **Next test:** offer five controllers a fixed-scope paid concierge service that reproduces the proposed outcome, with data-access and security requirements agreed in advance. This format is more actionable than “promising idea.” It tells the founder what can be believed now and what must happen before the next investment. ## Questions to ask before trusting a verdict Before accepting any build, pivot or stop recommendation, ask: - What sources materially changed the conclusion? - Are they independent of the seller? - Do they show behaviour or only opinion? - Are they current for the market and buying process? - Does the report identify the target segment precisely? - Which assumptions remain untested? - What evidence would reverse the verdict? - Is the next experiment capable of producing that evidence? If the report cannot answer those questions, the verdict is probably more confident than the research. ## Where SignalCraft draws the line SignalCraft gathers source-linked public evidence, organises it around demand, alternatives, customer behaviour and commercial risk, and recommends the next test. It does not claim that desk research validates a startup in the final sense. The [SignalCraft methodology](/methodology) gives behavioural evidence more weight than stated enthusiasm and keeps uncertainty visible. The [startup validation report example](/sample-report) shows how the evidence, confidence drivers, risks and unanswered questions appear in practice. The objective is not to outsource judgement. It is to make the founder’s judgement better informed and easier to challenge. ## The principle to keep Desk research is strongest when it narrows uncertainty before an expensive decision. It can prove that evidence exists. It can show what people currently do. It can reveal constraints and contradictions. It can help select a better experiment. It cannot prove that your product will win. Treat every verdict as a hypothesis about the evidence, not a fact about the future. ## Apply this guide Test one important claim for recency, independence, behavioural strength, commercial relevance and corroboration. [Score the evidence behind a verdict](/free-tools/evidence-quality-scorecard)