# Why Startup Viability Scores Create False Precision Canonical: https://signalcraftinsights.tech/blog/why-startup-viability-scores-create-false-precision Written by: D. Warnock Last reviewed: 2026-11-02 > A startup score can organise research, but it cannot express the probability of success unless it is calibrated against real outcomes. Here is how to use scored assessments without being misled. A viability score feels decisive. It compresses a messy market, uncertain buyer behaviour and dozens of assumptions into one number. That convenience is also the danger. A score can be a useful checklist summary. It is not automatically a probability, a forecast or proof that one idea is better than another. ## A precise number can rest on imprecise evidence Most early-stage evidence is incomplete. Search data can show interest without purchase intent. Reviews reveal pain among existing users but may not represent the wider market. Competitor revenue estimates can be wrong. Interviews contain memory and courtesy bias. Turning these inputs into 73/100 does not remove those limitations. It hides them behind arithmetic. The key question is not whether a tool uses a score. It is what the score means, how it was constructed and whether it has been tested against the outcome people infer from it. ## Ranking is different from prediction A scoring model may consistently rank one research brief above another. That can help triage opportunities. But a ranking is not a statement that an idea has a 73% chance of success. To support a predictive claim, a provider would need: - a clearly defined outcome, such as reaching a revenue threshold within a fixed period; - a representative set of ideas assessed before the outcome was known; - enough time for those outcomes to occur; - published calibration showing what happened at different score bands; - controls for selection, founder quality, funding, geography and survival bias; - independent testing on data not used to tune the model. Without that evidence, the number is an internal assessment scale. It may be thoughtfully designed, but it should not be read as an observed success rate. ## Weighted factors contain judgement A score typically combines market size, pain, competition, differentiation, timing and execution risk. The model must decide how much each factor matters. Those weights may be reasonable, but they are still choices. They can also interact in ways a simple weighted average misses. A large market does not repair destructive unit economics. Strong user pain does not guarantee budget. A distinctive product does not remove procurement friction. Ask whether a high total can average away a fatal weakness. In many businesses, one constraint—regulation, gross margin, access to distribution or an impossible integration—can dominate the result. ## Scores encourage threshold thinking A founder shown 69 may feel compelled to improve the pitch until it becomes 72. That can turn research into score optimisation. The better response is to inspect what would change the decision: - Which conclusion rests on the weakest source? - What evidence contradicts the attractive narrative? - Which assumption involves money, switching or repeated behaviour? - Who controls the budget? - What test could cause us to stop? A useful assessment should make these questions easier to answer, even if the headline score disappears. ## Benchmark distributions are not outcome calibration A provider may publish the distribution of scores across thousands of reports. This reveals how its model behaves: the median, spread and proportion above a threshold. That is valuable transparency. It still does not show whether high-scoring companies succeeded more often than low-scoring companies. It is a benchmark of model outputs, not necessarily a benchmark of business outcomes. Always read the definition beside the number. ## Replace the viability score with an evidence profile Instead of asking whether an idea is 78/100 viable, evaluate the important claims separately: 1. **Problem behaviour:** Are people already spending time, money or political capital on the problem? 2. **Buyer evidence:** Is there a funded priority, trigger and identifiable approval path? 3. **Alternatives:** What does the customer use now, and why is it tolerated? 4. **Economics:** Can the value, gross margin and acquisition path support the delivery model? 5. **Adoption:** What security, integration, switching or behaviour change is required? 6. **Uncertainty:** What is missing, contradictory or based on a weak source? Use the free [Evidence Quality Scorecard](/free-tools/evidence-quality-scorecard) to rate an important source for recency, independence, behaviour, commercial relevance and corroboration. ## When a score is still useful A score can help when it is used as: - a repeatable checklist; - a way to compare the same idea before and after new evidence; - a prompt for discussion; - a triage mechanism for deciding what to research next. Label it accurately. Keep the underlying dimensions visible. Do not imply a probability unless it has been calibrated as one. ## A better decision statement Replace “this idea scores 78” with: > Current evidence supports a frequent problem among a defined customer group, but buyer urgency and switching economics remain unproven. Run three buyer interviews and one paid pilot proposal before building the integration. That statement is less exciting. It is also more actionable and honest. SignalCraft follows this evidence-led approach: observations, interpretation, contradictions and gaps remain separate. Read the [methodology](/methodology), inspect the [claims register](/evidence), or compare the approach with [Preuve AI](/compare/signalcraft-vs-preuve-ai).