The Tool Desk
Outbyte Driver Updater FREEScan for outdated or missing drivers - takes under a minuteDriver Scan →Outbyte PC Repair FREEClear out junk files and repair common Windows errorsFree Scan →Short answer: Crunchbase did report “up to 95% precision” and 99% recall in internal backtesting of fundraising predictions after its February 19, 2025 predictive-intelligence relaunch. That is materially narrower than predicting which startups will become durable businesses or generate superior investment returns. The product may improve sourcing, timing and pipeline triage, but public evidence does not justify treating a Crunchbase score as an investment recommendation.
What Crunchbase actually announced
Crunchbase’s relaunch describes an AI-powered system that forecasts several private-company events and trajectories, including fundraising, growth, acquisitions, initial public offerings, closures, layoffs and remaining private. Its announcement says internal backtesting of fundraising predictions reached up to 95% precision and 99% recall. Read the company’s announcement at Crunchbase’s February 2025 release.
That statement does not establish “95% accuracy at predicting startup success.” It concerns particular, observable events—especially whether a company is likely to raise money—not one universally defined outcome called success.
The forecasts are different questions
- Fundraising: Is a company likely to raise capital, and over what time horizon?
- Growth: Is its future growth trajectory likely to improve?
- Exit: Could it be acquired or go public?
- Failure or contraction: Could it close or conduct layoffs?
- Investment-thesis insight: An LLM-generated summary of an investor’s historical patterns.
- Heat and growth scores: Signals intended to help users find momentum and prioritize companies.
Crunchbase also cites a case in which it assigned Coda a 93% acquisition probability less than 60 days before Grammarly acquired it. A well-timed case study is evidence that a prediction can be useful; it is not a representative success rate.
#1 Best Overall
Why “95% accuracy” is the wrong translation
Crunchbase’s headline number uses precision, not ordinary-language accuracy. The metrics answer different questions:
| Metric | Question answered | What a high value can still hide |
|---|---|---|
| Precision | Of the events the model predicted, how many happened? | A model can issue relatively few positive forecasts and miss many real events. |
| Recall | Of all events that happened, how many did the model identify? | High recall can come with many false positives. |
| Accuracy | Across positive and negative cases, how many classifications were correct? | With a rare event, a simple “nothing will happen” rule can look accurate. |
| Calibration | Does a 70% forecast occur about 70% of the time? | A ranking can be useful without its probabilities being reliable. |
Consequently, “95% precision and 99% recall” cannot be rewritten as “the AI identifies successful startups 95% of the time.” Precision and recall also depend on the event definition, forecast window and threshold used to call a prediction positive.
Crunchbase’s probability labels
The company’s documentation maps scores into five labels. They are probability ranges, not guarantees that every company in a tier will experience the event.
| Label | Probability range |
|---|---|
| Very Likely | 0.95–1.00 |
| Probable | 0.66–0.95 |
| Uncertain | 0.36–0.65 |
| Doubtful | 0.06–0.35 |
| Very Unlikely | 0.00–0.05 |
For the definitions and model-performance notes, see Crunchbase’s prediction-model documentation.
Quick wins for a faster PC:
Repair Windows errors before they cause bigger problemsFix Now →Scan for outdated or missing drivers - takes under a minuteDriver Scan →Clear out junk files and repair common Windows errorsFree Scan →What remains unknown about the test
The public materials reviewed do not provide enough information to reproduce or independently verify the 95% result. A serious evaluator should request:
- The number of companies and prediction cases.
- The positive-event rate and class balance.
- An exact definition of a “funding event.”
- The forecast horizon—such as three, six or 12 months.
- Whether evaluation was chronological and out-of-time rather than randomly sampled.
- How missing-data companies, undisclosed rounds and delayed announcements were labeled.
- Whether the test set was kept separate from model development.
- Whether repeated predictions for one company counted as independent cases.
- Performance against simple baselines, not only against no model.
- Calibration and results by geography, sector, company age, stage and market cycle.
- Independent audit, peer review or a reproducible evaluation set.
Crunchbase’s later product materials report that the platform correctly predicted 84% of real-world funding events, 72% of acquisition events and more than 16,000 predictions proven correct. Those figures are not directly comparable with the earlier “up to 95% precision and 99% recall” backtest without the samples, horizons, event definitions and evaluation procedure. See Crunchbase’s data page and its API page.
What signals feed the predictions?
Crunchbase says its models combine funding activity, leadership changes, user-engagement signals, news velocity, market momentum and historical company attributes with public web information, government filings, investor and employee contributions, data partnerships, analyst validation and aggregated activity from professionals evaluating private companies. The company says its pipeline processes more than 30 million verified updates each year and markets access to more than 80 million live signals from professionals evaluating private companies. Details are described at about.crunchbase.com/data.
These are often proxies rather than direct measures of business quality:
Free tools Windows power users keep installed
One-click scans. No signup required.
Rank #3
- More news can mean momentum, or simply a stronger public-relations operation.
- More engagement can reflect investor attention rather than customer demand.
- A leadership change can signal expansion, distress or routine hiring.
- Earlier fundraising can predict another round partly because well-funded companies have greater visibility and network access.
Is it predicting success—or investor behavior?
A funding forecast may primarily predict that a startup is already attracting attention, that existing investors may support it again, or that it has enough visibility to produce another observable financing event. That can be commercially valuable while saying little about margins, retention or eventual returns.
- Financing: Will the company raise?
- Visibility: Will investors and the market pay attention?
- Business performance: Will revenue, customers, margins or retention improve?
- Exit: Will it be acquired or go public?
- Returns: Will an investor earn an attractive risk-adjusted return after valuation, dilution and failure risk?
Crunchbase publicly documents the first four categories. The reviewed materials do not establish a validated model of investor returns. Its Predictions and Insights documentation should therefore be read as product guidance, not proof of alpha.
Where the tool could improve investing
Deal sourcing and timing
Investors can rank large company universes by likely fundraising, growth or attention, find candidates before a round is announced and receive alerts when a target’s trajectory changes. Crunchbase says its funding-prediction API supplies probability scores, time-horizon probabilities and supporting evidence; the technical description is at Funding Predictions in the API.
Pipeline prioritization
An analyst can combine model signals with stage, geography, sector and thesis filters to decide which companies deserve a first call. This is a triage function: it allocates scarce research time rather than deciding which security to buy.
Portfolio monitoring
Signals for potential layoffs, closures, financing needs or deteriorating momentum can add an early-warning layer between board updates and quarterly reporting.
Research productivity and integrations
Profile summaries, investor-pattern insights, alerts, exports and API access can reduce repetitive collection of fragmented market information and feed internal analytics or customer-facing products.
What it cannot replace
- Founder references and customer calls.
- Evidence of product quality, retention, pricing power and unit economics.
- Technical-defensibility and competitive analysis.
- Valuation, ownership, dilution and term negotiation.
- Portfolio construction, reserves and position sizing.
- Judgment about skewed venture outcomes, survivorship bias and information asymmetry.
A company can raise at an attractive headline valuation and still destroy value for new investors. Conversely, a bootstrapped or quiet company may be economically strong but generate too little public signal for a database model.
Technical and market failure modes
Leakage, look-ahead and survivorship
If a historical test includes information that became available only after an event, or applies later data corrections retrospectively, it can overstate live performance. Companies that quietly close or never disclose financing are also harder to label, creating survivorship and missing-data problems.
Recommended Free Tools
Coverage and base rates
Crunchbase coverage is not uniform across countries, sectors, stages or companies without prominent investors and public announcements. Fundraising and exits are selective events, so a high percentage is meaningful only alongside event prevalence and a simple baseline.
Drift, gaming and reflexivity
Interest rates, regulation, AI-driven company formation and investor preferences change private markets. Founders may increase news activity or profile updates to improve apparent momentum. If many investors act on the same ranked list, competition may rise, valuations may increase before the predicted round and the signal may lose value. Those effects are plausible market inferences, not outcomes demonstrated by Crunchbase.
False confidence and governance
A probability score can feel like a recommendation. Crunchbase says its AI may contain mistakes and is not legal, financial or investment advice; that warning appears in the product announcement. Buyers should also ask what usage data is collected, how it is anonymized, what contractual rights apply and whether outputs may be used in fiduciary or regulated decisions.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.How an investor should test it
- Define one use case: for example, finding seed companies likely to raise within six months.
- Freeze an evaluation date: save predictions before outcomes are known.
- Set a baseline: compare the model with a human shortlist and simple rules such as recent funding or employee growth.
- Track the right metrics: precision, recall, calibration, lift, analyst time saved and qualified-meeting conversion.
- Segment results: review stage, sector, geography, company age and market regime separately.
- Study errors: keep written examples of false positives and false negatives.
- Measure economics: test whether the signal improves diligence choices, follow-ons or risk-adjusted outcomes after fees and competition.
- Keep human controls: never change investment policy solely because of a model score.
Crunchbase versus other private-market tools
The right comparison is workflow fit, not a contest to declare one database “most accurate.”
| Product | Best fit | Distinctive use | Pricing signal |
|---|---|---|---|
| Crunchbase Pro | Individuals, scouts, founders and small analyst teams | Company discovery, alerts, growth signals and lightweight workflows | $99 monthly or a $588 introductory annual offer in current support materials; verify checkout because offers change. Official buying page |
| Crunchbase Business/API | Venture firms, corporate-development and data teams | Team workflows, integrations, bulk access and embedded predictions | Sales-led/contact-based. See Business comparison and API page. |
| PitchBook | Established VC, private-equity, banking and corporate-finance teams | Institutional private-capital coverage, fund and transaction analysis, benchmarking and analyst support | Request pricing; no public standard price shown. Pricing |
| CB Insights | Corporate strategy, innovation and competitive-intelligence teams | Market intelligence, predictive feeds, relationship data and research features | Enterprise/request pricing. Pricing |
For a solo investor primarily seeking startup discovery, a limited Crunchbase Pro trial may be the most practical starting point. Teams needing integrations should evaluate Business/API. Institutions requiring detailed fund analysis, benchmarking or primary research should compare PitchBook and CB Insights through demonstrations. None should be purchased on the assumption that a prediction score guarantees investment gains.
Verdict: useful screening layer, unproven return predictor
Crunchbase has a credible product case for helping investors find and prioritize opportunities, anticipate financing conversations and monitor private companies. Its “up to 95%” figure is a company-reported precision result for a narrower fundraising-prediction test, paired with 99% recall—not an independently validated 95% success rate for startups or investors.
Use the system to decide where to spend diligence time. Do not use it to skip customer calls, valuation work, references, portfolio construction or independent judgment. Before paying for an institutional deployment, demand an out-of-time validation sample, calibration data, stage and geography breakdowns, false-positive examples and evidence that the signal improves decisions rather than merely forecasting which companies already attract attention.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
