The first thing Clearspeed tried to fix was not an insurance claim. It was a failure of trust with mortal consequences. Alex Martin, then a Marine officer, lost a close friend on a joint patrol when a vetted infiltrator attacked the unit. The screening machinery had produced a green light. Martin came home asking why a system so heavy with records could still be blind at the instant a decision mattered.
That question became Clearspeed, a San Diego company founded in 2016. Its product is deliberately brief. A person answers a handful of tailored yes-or-no questions by phone or web. The platform says it analyzes changes in the voice rather than the meaning of the answer, then returns one of two routing signals: minimal risk or elevated risk. The first group can move. The second gets attention from a trained human.
This is not conversation intelligence, transcription or a chatbot looking for suspicious phrases. It is also not supposed to be a robotic verdict. Clearspeed describes its output as one new data point placed beside an organization's existing evidence. That limited role is the hinge of the pitch: it promises speed without asking a risk team to surrender the consequential decision.
The costly thing was the old process
Clearspeed's first important customer was U.S. Special Operations Command. In 2018, the company says it screened 715 Afghan commando recruits in less than 20 hours. A workflow that had taken months was compressed into a day, and a later company case study put the cycle-time reduction at 95 percent - from roughly 90 minutes to five per person. Clearspeed also cites greater than 97 percent accuracy in a formal Department of Defense study.
That was the proof that changed the company's range of ambition. A multilingual screen could survive a setting where records were incomplete, time was short and errors carried unusual weight. The lesson traveled well. Insurers also have queues full of people who are probably genuine, a smaller number of questionable cases, and expensive specialists who should not examine everything with equal intensity.
“If you can trust people faster, you can pay them faster.”Alex Martin, co-founder and CEO
The cost of buying Clearspeed is less transparent. The company does not publish a price list, and enterprise deals are likely to vary with volume, use case, delivery channel and integration. The economic case is therefore told through customer outcomes. Seguros El Roble, a Guatemalan insurer, reported 31 times return on investment in the first year. The more general calculation is easy to understand: fewer routine files waiting for investigators, faster payment for genuine customers, and more focused scrutiny where the signal is elevated.
Insurance found the second act
Clearspeed launched its commercial product in 2021 and found a wedge in claims. Express handles the front of the funnel, where a short assessment can help a legitimate claim move toward straight-through payment. Precision goes deeper with supplementary questions, narrowing what may require investigation. Surge, introduced in 2022, is designed for catastrophes and weather events, when call volumes can jump and adjusters are already strained.
Zurich UK provides the most revealing account of how a buyer's mind changed. It first looked at the tool as a way to detect fraud earlier. Working with Allied Universal, the insurer put Clearspeed into property-and-casualty and warranty or device claims. Zurich reports that 58 percent of assessed claims were accelerated for payment, fraud referrals rose 20 percent, and withdrawals increased from about one percent to five percent. The company began describing the system less as a fraud detector and more as a claims-validation layer.
Allianz UK reached a similar conclusion. Its personal-lines deployment, also involving Allied Universal, produced a reported 59 percent lift in claims settled immediately and a 36 percent lift in average fraud-loss savings per claim. The results matter because they dissolve an old trade-off. A risk control is easier to defend internally when it improves the honest customer's trip instead of making everyone wait.
The three-move workflow
A layer, not a renovation
Many enterprise AI products arrive with a renovation project attached. Clearspeed's commercial advantage is that the voice channel often already exists. The assessment can run on the web, over the telephone or through a call center. A secure REST API can send the result into the customer's stack in real time. No interviewer needs to be on the other end, which also makes the questions consistent from one person to the next.
That helps explain the company's place in the market. Rules engines, identity checks, device intelligence, transaction models and manual investigations look at different pieces of risk. Voice-biometrics systems may ask who is speaking; conversational tools may parse what was said. Clearspeed says it extracts a present-tense, non-semantic signal from how an answer is voiced. It sells that signal as another layer, not a replacement for the layers already working.
Where it sits
Upstream of the expensive decision. The product belongs between intake and investigation - after a claimant, recruit or applicant responds, but before every case consumes the same expert time. Its competition is as much the undifferentiated queue as any single software vendor.
The company sells to government, defense, insurance, banking and other high-volume risk operations. It says the technology works in more than 60 languages and is used in 37 countries. PwC UK offers it to clients facing fraud exposure. Carahsoft became its principal public-sector distributor in 2025. Named commercial users include Allianz, Zurich, Seguros El Roble and Pinnacol Assurance.
That customer list also explains the company's expertise. Its leadership combines military operations, intelligence, enterprise software, insurance and risk. Chief technology officer Francesco Crippa sits beside executives responsible for product strategy, government and defense, legal, revenue and people. Public head-count figures vary by date, but recent estimates place the team at roughly 110 to 120. For a company selling confidence into regulated institutions, the mix is practical: model-building alone does not navigate procurement, claims policy or the consequences of an alert.
The business model is enterprise software, delivered through the Clearspeed application, APIs and existing voice channels, then extended by distributors and consulting partners. That makes culture commercially relevant. The company talks about consistent assessments and starts with the presumption that most people are genuine. It is a less adversarial stance than traditional fraud language, but also a useful product discipline. If the low-risk path is not meaningfully faster, the customer has simply purchased another checkpoint.
What another operator can copy
The reusable idea is not the signal itself; Clearspeed has spent years developing its proprietary dataset and models. The reusable idea is the shape of the workflow. Start with the majority outcome. Find the smallest intervention that can safely route obvious cases. Keep the expert in the exception path. Add the new capability through channels customers already use. Then measure time released, not merely alerts produced.
There is a second lesson in the company's slow beginning. Martin has called Clearspeed a “six-year overnight success.” The interaction may last seconds, but acceptance in defense, insurance and banking depends on validation, governance and real-world evidence. A clever demonstration could not substitute for a deployment with 715 recruits or a claims pilot that showed both faster payment and more useful referrals.
Capital followed the evidence. Clearspeed raised a $27 million Series C in 2022, taking funding to $50 million. In June 2025 it added a $60 million Series D led by Align Private Capital, with IronGate Capital Advisors, Bravo Victor Venture Capital and KBW Ventures participating. Retired general and former CIA director David Petraeus joined as a multi-round investor. Total funding reached $110 million; the company has not publicly disclosed a valuation.
Where the method should stop
A voice layer is not universal infrastructure merely because it is multilingual. It needs a usable recording, a voice-enabled moment in the workflow, clear consent and a process that can accommodate people with speech, hearing or connectivity constraints. It will not help when the task requires understanding the content of an answer. And an elevated-risk indicator should not become a shortcut to punishment, denial or exclusion without context and review.
High case volume, an existing phone or web voice step, costly manual review, a large low-risk majority and a defined human escalation path.
No reliable voice channel, inaccessible participation, unclear consent, a need for semantic analysis or an expectation that one score makes the final decision.
Those boundaries are not side notes. They are part of the product. Clearspeed's promise is credible only if “minimal risk” means faster passage and “elevated risk” means look closer - not guilty. The company has built its pitch around that restraint, even as the language of AI risk scoring invites larger claims.
The result is a business positioned between the balance sheet and the battlefield, two places where delay and misplaced suspicion both have costs. Clearspeed's wager is that institutions do not need another mountain of historical data as much as they need one timely signal and the discipline to use it well. The product may be complex. The operating instruction is not: clear the hay, then look for the needle.