Founders ask customers what they want. Sales teams hear what they will not pay for.
The first answer is useful. The second may contain more information.
A customer interview can reveal frustrations, preferences, and imagined future behaviour. A real sales process introduces something interviews often lack: consequences.
There is a budget. There are competing priorities. There is an implementation burden. There are security reviews, political risks, procurement requirements, existing contracts, and people whose careers may be affected by the decision.
When a qualified buyer says no, delays a purchase, or reaches the end of a process without acting, the answer contains evidence about where the market’s practical boundaries lie.
Most companies throw much of that evidence away. The opportunity is marked as lost. A salesperson selects “budget”, “timing”, “competitor”, or “no decision” from a dropdown. The revenue forecast is updated, and the team moves on.
Repeated rejection is not merely a record of revenue that failed to materialise. It can be a map of problems the existing market has failed to solve.
The pipeline graveyard contains market information
Research based on more than 2.5 million recorded sales conversations found that between 40% and 60% of deals ended in no decision, even when buyers had expressed an intention to purchase.
That means a large share of sales opportunities are not lost to a clearly superior competitor. They are lost to inaction.
Inaction is often treated as the absence of information. In reality, it may contain several different signals.
Perhaps the problem is real, but not painful enough to overcome implementation risk. Perhaps the buyer wants the outcome, but the product requires too much organisational change. Perhaps the solution is sold to the wrong budget owner. Perhaps the product solves one part of the workflow while creating new work elsewhere.
Perhaps the customer already uses a manual workaround that is inefficient but politically safer. Or perhaps the market needs the same capability packaged as a service, an integration, an infrastructure layer, or a completely different product.
A lost deal does not tell you which explanation is correct. But hundreds or thousands of similar lost deals can reveal a pattern worth investigating.
From win-loss analysis to rejection intelligence
Traditional win-loss analysis asks: why did we win or lose this opportunity, and how can we improve the current offer? That is an important question.
But there is a larger one: do recurring reasons for losing reveal a product, business model, category, or company that does not yet exist?
I call the process of answering that question rejection intelligence.
Rejection intelligence is the structured analysis of the raw language, constraints, workarounds, and decision barriers found in lost, stalled, and rejected sales opportunities. Its purpose is not simply to improve the close rate of the current product. Its purpose is to generate new market hypotheses.
At Belkins, we have worked across more than 1,000 client engagements and more than 50 industries. That creates an unusual observation point. We hear how companies describe their growth problems, what they have already attempted, why existing solutions disappointed them, and what prevents them from choosing another provider.
The least useful summary of that information is often the CRM loss reason.
“Too expensive” may mean the buyer does not believe the problem is urgent. “Wrong timing” may mean nobody internally owns the project. “Missing feature” may mean the product does not fit the customer’s real workflow. “Already have a vendor” may mean the switching cost is greater than the value being offered.
The label records the objection. The underlying conversation reveals the constraint.
Prospects rarely describe the product you should build
Customers are usually good at describing friction. They are less reliable at designing the best solution.
That is why the objective is not to build every feature a prospect requests. Doing so produces bloated products assembled from unrelated sales objections.
The objective is to identify constraints that appear repeatedly, create material pain, and already force customers to spend time, money, or political capital on a workaround.
Folderly emerged from this kind of recurring operational friction. In 2019, Belkins initially developed Folderly for internal use after email deliverability problems began affecting outreach. Clients were also repeatedly concerned about messages failing to reach the inbox, turning deliverability from a technical nuisance into a business constraint.
Folderly was not created by asking an AI model to cluster thousands of lost deals. But it demonstrates the same mechanism. A problem appeared too frequently and affected the core customer outcome too directly to remain somebody else’s problem.
The market did not necessarily say, “Please build a deliverability platform.” It said, “We cannot achieve the result we are paying for unless this problem is solved consistently.”
Valuable products often first appear not as fully formed product requests, but as recurring obstacles that customers cannot remove.
AI makes the hidden patterns easier to see
Until recently, extracting those patterns was expensive. The useful information is scattered across sales notes, call transcripts, emails, loss interviews, proposals, procurement discussions, support tickets, and account-management conversations.
CRM categories compress that complexity into a few convenient labels. AI can help process the underlying material at a much larger scale.
A system can group semantically related objections, compare patterns across industries, identify newly emerging language, separate objections by buyer role, and monitor how reasons for rejection change over time.
For example, “security will not approve this”, “we cannot expose customer records”, and “legal rejected external data processing” may all point to the same underlying constraint, even though they use different words.
But an AI-generated cluster is not automatically a market opportunity. Large language models are very good at producing organised explanations from messy information. They can also make weak patterns look more coherent than they really are.
The output must therefore be treated as a hypothesis, not a verdict.
Six tests for a genuine venture signal
1. Repetition
Does the same underlying constraint appear across many qualified opportunities, or is the idea based on a handful of loud prospects? Frequency alone does not prove demand, but an issue that repeatedly blocks purchases deserves attention.
2. Intensity
Is the problem mildly inconvenient, or does it stop the customer from reaching an important outcome? The strongest opportunities usually involve revenue, risk, compliance, cost, speed, or organisational credibility.
3. Existing workaround
What does the customer do today? A spreadsheet, internal employee, collection of disconnected tools, manual review process, consultant, or informal workaround is evidence that the problem already consumes resources. The more effort customers invest in compensating for a problem, the more credible the opportunity becomes.
4. Cross-segment recurrence
Does the same underlying job appear in several industries or company sizes? A repeated problem across multiple segments may support a larger category. A problem isolated to one narrow segment may still support a valuable vertical product, but the company must be designed accordingly.
5. Timing and trigger
Why has the problem become urgent now? A regulatory change, new technology, rising labour cost, platform policy, security requirement, or shift in customer behaviour can convert an old inconvenience into a new market.
6. Right to win
Does your company possess an unfair advantage in solving the problem? That advantage may be distribution, customer access, domain knowledge, proprietary workflow data, a trusted brand, existing infrastructure, or the ability to test the product inside a live service business.
A good problem is not automatically a good startup for you.
Test commitment, not enthusiasm
The final validation cannot be another survey asking whether people like the idea. Buyers frequently express enthusiasm for products they will never implement or purchase.
The test should require some form of commitment. That may be a paid pilot, a deposit, access to operational data, an integration, a design partnership, internal stakeholder time, or permission to replace part of the existing workflow manually.
The objective is not to prove that people find the idea interesting. It is to prove that the problem is important enough for them to change their behaviour.
This is where service businesses have an overlooked advantage. An agency or consultancy already operates inside real customer workflows. It can test a potential product manually, observe where it fails, and measure whether customers will pay for the outcome before committing to a large software build.
The service is not simply a source of cash flow. It can function as a live market-sensing system.
The database is a hypothesis engine, not a truth machine
Rejection data contains serious biases. A company only hears from the market it chose to target. A weak sales pitch can manufacture objections that have little to do with the product. A buyer may provide a polite answer rather than the real political reason for rejecting the offer. Sales representatives may record the explanation that is easiest to enter into the CRM.
Silence is even more ambiguous. An unanswered email is not a product brief.
Rejection intelligence must distinguish explicit buyer statements from inferred causes, qualified opportunities from poor-fit prospects, stalled deals from simple non-response, product limitations from sales-execution failures, and recurring patterns from isolated anecdotes.
The analysis also needs strict privacy controls. The valuable unit is not one prospect’s confidential sentence. It is an anonymised, recurring pattern.
Names and identifiers should be removed. Sensitive customer information should not be repurposed. Results should be aggregated, minimum cluster sizes should be applied, and human reviewers should validate AI-generated categories.
The goal is not to monetise confidential conversations. It is to recognise common market constraints responsibly.
From a sales function to a company-building system
At Equinox Ventures, we are trying to formalise a related operating loop. The portfolio is organised around utility, shared resources, and founder support. The underlying idea is that distribution, operating expertise, and customer proximity can do more than support existing companies. They can reveal which company should be created next.
This does not mean every agency should become a venture studio. Most rejection clusters will lead nowhere. Some will point to features rather than companies. Others will describe problems that are real but too small, too difficult, or too poorly timed to support a business.
But even a small number of strong signals can change how a company approaches innovation.
A CRM should not only tell you how much revenue was lost. It should help explain which constraints the market repeats too often to ignore.
The pipeline graveyard is often where tomorrow’s products are buried.
A lost deal is only a dead end when the company records “budget” and learns nothing.
Get the TNW newsletter
Get the most important tech news in your inbox each week.