The AI Liability Database
Technical intelligence for evaluating AI companies
Understanding how AI companies fail — before those failures become visible to the market.
Why This Database Exists
The AI boom has created extraordinary innovation — and extraordinary exaggeration. Many companies present sophisticated AI narratives while relying on fragile architectures, incomplete machine learning systems, or hidden operational workarounds. By the time these weaknesses become visible, investors and acquirers have already committed significant capital. The AI Failure Pattern Database was created to study these breakdowns systematically and transform them into signals that can be recognized earlier.
Independent by Design
Technology assessments are often influenced by incentives. Many advisory firms audit companies while also hoping to win the contracts required to fix the problems they uncover. KairosInfra removes this conflict entirely. We do not build software, implement solutions, or sell engineering services. Our work focuses solely on analyzing the technical reality behind AI products and infrastructure, ensuring that investors receive an objective and uncompromised evaluation.
How Investigations Work
Each case in the AI Failure Pattern Database is examined through a structured investigation process. This includes analyzing system architecture, machine learning pipelines, engineering organization structures, and the operational processes behind product demonstrations. The objective is to identify patterns that reveal when a system marketed as advanced AI is actually dependent on manual operations, unstable infrastructure, or incomplete machine learning systems.
1
AI Failures Documented
$
1
B+
Estimated Value Destroyed
1
Industries Impacted
1
%
Failures Linked to AI Authenticity
Click any case to view investigation details.
| Case | Year | Sector | Deal Size | Failure | Verdict | Red Flags | Valuation Issue | Phase0 Catch |
|---|---|---|---|---|---|---|---|---|
| Enterprise HealthTech AI — Fortune 500 unit, sold to PE | 2022 | Healthcare AI | $4B invested | AI Authenticity | Do not proceed | AI recommendations based on synthetic training data, not real clinical workflows. No real ML pipeline — relied on structured rules. Org chart showed no ML engineers in clinical teams. | ~$3B written off | Product demo would show decision-tree logic, not ML inference. Org chart would reveal no ML engineers. Architecture diagram would show no real-time data pipeline. |
| Healthcare Revenue Cycle Unicorn, $4B valuation, shutdown | 2023 | Healthcare Admin | $902M raised | AI Authenticity | Do not proceed | Marketed as 'autonomous AI.' Humans were manually fixing errors behind the scenes. KLAS Research gave a 'C' rating citing misleading claims. Hospital clients reported delays and contract terminations. | ~$3.5B | Product demo of complex workflows would reveal manual intervention. Engineering org chart would show operations staff vastly outnumbering engineers. ML workflow description would reveal no real training pipeline. |
| UK Digital Health AI, SPAC IPO at $4.2B, bankrupt | 2023 | AI Telehealth | $1.2B raised | AI Authenticity | Do not proceed | CEO publicly claimed their chatbot scored 81% on a medical exam. A former engineer confirmed it was an "if/then decision tree in an Excel spreadsheet — not AI." UK regulator opened a review of the symptom checker. The Lancet published a study saying its diagnostic claims lacked convincing evidence. CTO left just before collapse. | ~$4.15B valuation overstatement | Product demo of chatbot would have exposed decision-tree logic. LinkedIn would show no ML engineers in clinical AI teams. Architecture diagram would show no real inference pipeline. |
| Autonomous Vehicle AI, $3.6B raised, Large car manufacturer-backed, shutdown | 2022 | Autonomous Driving | $3.6B raised | Architecture Risk | Do not proceed | Projected Level 4 autonomy by 2021 — publicly stated by Company CEO. No revenue after years of operation. Architecture required full sensor suite, no graceful degradation. 2,000 employees but no commercial product. Could not attract third-party investors when JV tried to exit. | $12.4B peak; zero recovery for most investors | ML workflow description would show Level 4 was still unsolved research, not deployable product. Cloud infrastructure review would show no production-scale serving. Engineering team check would show pure research org with no commercial path. |
| Autonomous Trucking AI, Nasdaq IPO, $8B peak, US operations shut | 2023 | Autonomous Trucking | $8B peak market cap | Team Risk | Do not proceed | Caught up with investigations into tech transfer to othercompany. Co-founder fired for poaching employees to his new venture. April 2022: autonomous truck crashed into barrier on I-10, Revenue was $2.6M/quarter despite $8B valuation. All key engineers outside country. | ~$7.7B from peak to near-zero | Engineering org chart would reveal core IP team was not available in country. Architecture diagram would show no commercial-scale deployment stack. Revenue vs. valuation math would immediately flag as unsustainable. |
| Autonomous Trucking Software, $5.2B SPAC, sold for $71M 18 months later | 2023 | Autonomous Trucking | $5.2B SPAC | AI Authenticity | Do not proceed | Zero revenue at time of SPAC. Short-seller Bear Cave report in January 2022 noted: "holds no patents, has only a dozen test trucks." Targeted "driver-out" by 2023 — missed completely. 14,200 "reservations" were non-binding. Traded below cash value within 12 months. | ~$5.1B | Product demo would show research-grade software, not production-ready system. ML workflow description would expose massive gap to commercial deployment. Cloud architecture review would show no production serving infrastructure. |
| AI Cybersecurity SPAC, ex-NSA Director, bankrupt — AWS cut service for $18K unpaid bill | 2023 | Cybersecurity | $400M raised | Team Risk | Do not proceed | Former employees said it was "like Theranos — a culture of deceit." Product was described as having "nothing special" technically — brand was entirely the founder's NSA reputation. Revenue flat at $27M/year despite $400M raised. AWS shut them down for an unpaid $18,000 bill. C5 Capital (major investor) was also their largest customer — circular revenue. | ~$3B+ | Tech stack review would show no proprietary ML — reliant on third-party threat feeds. Cloud spend review would show no scalable architecture. Engineering org chart would show no ML engineers — primarily former government officials. |
| Robo-Advisor, SEC-charged for AI-washing — no client data was ever used in any algorithm | 2024 | AI Investment Mgmt | SEC fine: $225K | AI Authenticity | Do not proceed | Claimed "AI uses collective client data to predict which companies will make it big." In reality: no client data was ever used in any algorithm. SEC found this in a routine exam in July 2021 — Company admitted it. Then continued making the same claims in marketing until 2023. This is the textbook Phase 0 pattern: pitch deck says AI, product demo reveals no inference pipeline. | Not an acquisition — regulatory case. But directly relevant: this is what PE targets are doing. | ML workflow description request would have revealed no training pipeline. Architecture diagram would show no ML inference layer. Product demo of "AI predictions" would expose simple rule-based logic. |
| AI Financial Advisory, claimed 'first regulated AI advisor,' SEC-fined | 2024 | AI Financial Advisory | SEC fine: $175K | AI Authenticity | Do not proceed | Claimed to be "first regulated AI financial advisor" — could not produce any documents to support this. Claimed "AI-driven forecasts outperform IMF by 34%" — no methodology disclosed. Listed no AUM while claiming $6B on website. Classic pattern: marketing claims 10x exceed actual product capability. | Not disclosed — but any PE deal valuing this based on "AI" claims would have overpaid massively | Tech stack documentation would show no proprietary AI model. Product demo would expose chatbot with no real forecasting engine. Engineering org chart: one or two developers, no ML team. |
| AI Primary Care Startup, $650M burned, hardware-dependent AI, shutdown | 2024 | AI Primary Healthcare | $650M raised | Architecture Risk | Do not proceed | Built $3,500 kiosks as the AI delivery mechanism — a hardware-dependent architecture with catastrophic unit economics. AI was marketed as replacing doctors. Reality: expensive kiosks with basic sensor readings and no reimbursement pathway. No ML training pipeline — sensors fed simple if/then logic. Could not achieve scale without manufacturing hundreds of thousands of units. | All $650M | Architecture diagram would expose hardware dependency as an unscalable single point of failure. Cloud infrastructure review would show no real ML inference serving. Unit economics on the product demo would immediately show why this cannot scale. |
| No-Code AI App Builder, Microsoft-backed, $1.5B valuation, bankrupt — revenue inflated 300% | 2025 | No-Code Development | $450M raised | AI Authenticity | Do not proceed | Claimed $220M in revenue for 2024 — real number was $55M, a 300% exaggeration. Crunchbase News A 2019 Wall Street Journal investigation had already found that the company's "AI" wasn't doing much — behind the curtain were hundreds of engineers in India and Ukraine, manually coding what was being advertised as automated magic. Crunchbase News Documents reviewed by Bloomberg showed Builder worked with VerSe, an India-based social media startup, to falsely increase its sales numbers, regularly billing each other for similar amounts between 2021 and 2024. Medium Despite all this, Microsoft and SoftBank continued investing. | ~$1.45B (raised $450M+, zero recovery) | Product demo of AI would have exposed human-assisted coding, not autonomous generation. ML workflow description would show no training pipeline. Revenue documentation request would have caught circular billing with other comapny. Engineering org chart would show operations headcount dwarfing ML engineers. |
| Consumer Wearable AI, ex-Apple founders, $240M raised, sold to HP for $116M | 2025 | Consumer AI Hardware | $241M raised | Architecture Risk | Caution | Due to overheating problems, Company executives would use ice packs to chill the AI Pin before previewing it to investors or partners. Once named a Time magazine best invention of the year, it hqw disappointed users who complained about malfunctions, its high price and overheating problems. Due to sluggish sales, the AI Pin had to cut its price from $699 to $499. Feedbck was "bad at almost everything it does." Returns outpaced sales at one point. | Rumored $1B valuation; Got paid $116M for IP only, not the product business | Architecture diagram would expose hardware dependency — all AI ran through cloud servers, meaning device was worthless without connectivity. Cloud infrastructure review would show no on-device ML inference. Unit economics on product demo would reveal $699 device + $24/month subscription was unsustainable at any realistic sales volume. |
| AI Freight Brokerage, Bezos + Gates-backed, $3.8B valuation, zero recovery | 2023 | AI Freight Logistics | $1B+ raised | AI Authenticity | Do not proceed | The business showed clear identity confusion, with even internal teams unsure whether it operated as a technology platform or a traditional freight brokerage, indicating weak product differentiation. The company commanded an extreme valuation multiple of roughly 28× revenue despite having no credible path to profitability. At the same time, it went through four rounds of layoffs within an 18-month period while continuing to raise capital at inflated valuations, signaling underlying operational stress. Additionally, while digital pricing and tendering were positioned as scalable innovations, the model failed to account for the inherently human elements of freight logistics—relationship management, exception handling, and problem resolution—which do not scale cleanly through APIs. | The valuation was largely driven by the narrative of being an “AI-powered logistics platform,” rather than by a defensible technological advantage or sustainable business fundamentals. With approximately $136M in revenue, the company’s scale did not justify a multi-billion-dollar valuation in a structurally low-margin brokerage industry. The underlying unit economics were weak, as cloud infrastructure and operational costs eroded already thin margins, leaving little room for profitability or resilience during market downturns. This resulted in near-total value destruction, with an estimated ~$3.6B in overstatement as the company’s assets were ultimately sold for a fraction of its prior valuation. | A basic cloud cost audit during early diligence would have revealed that the matching algorithm’s infrastructure costs were scaling faster than the revenue generated per transaction. A review of the ML workflow would have shown that the core algorithmic matching capability lacked true defensibility and could be replicated by incumbents. Additionally, a unit economics stress test would have exposed the inability of the business to sustain margins under realistic freight market conditions. Finally, mapping customer operations against system architecture would have highlighted the heavy reliance on non-automatable human processes, contradicting the scalability assumptions embedded in the investment thesis. |
| Autonomous Delivery Robot, ex-Google founders, $8.6B peak valuation, delivery business abandoned | 2024 | Delivery Robotics | $2.1B raised | Architecture Risk | Caution | The company pursued a capital-intensive strategy of building its own autonomous delivery robots, rapidly burning cash without demonstrating scalable economics. It invested in a $40M manufacturing facility intended for mass production, only to later abandon those plans, signaling a lack of execution clarity and shifting strategy. Public statements framed AI advancements as extending runway, but in reality reflected underlying financial distress masked as optimism. This was reinforced by two major rounds of layoffs within a year, cutting 20% and then 30% of the workforce. Despite high-profile partnerships with major retailers, revenue contribution remained negligible relative to the $2B+ invested, indicating a disconnect between commercial traction and capital deployment. | The valuation was driven by the narrative of autonomous delivery powered by advanced AI, rather than by viable unit economics or proven scalability. At its peak, the business was valued at $8.6B despite lacking meaningful revenue and operating in a hardware-heavy model with inherently high costs. The dependency on manufacturing and deployment of physical robots introduced structural inefficiencies that could not support sustainable margins. As the delivery business failed to materialize commercially, its value effectively dropped to zero, leading to a down-round at $6B and an estimated ~$2.6B overstatement, with remaining value tied only to a pivot toward software licensing. | An early architecture review would have revealed that the entire model depended on hardware manufacturing, creating a single point of failure and limiting scalability. A cloud and systems analysis would have shown that the AI required specialized on-device processing and custom silicon, with no realistic path to achieving mass production economics. Additionally, a unit economics assessment would have exposed that the cost per delivery was orders of magnitude higher than commercially viable thresholds. These diligence steps would have clearly indicated that the business model could not achieve sustainable profitability in its intended form. |
| AI-First Insurance Platform, $4.4B peak cap, AI caught denying claims using protected characteristics | 2023 | AI Insurance | $1B+ raised | ML Model Risk | Caution | The company’s revenue quality was fundamentally compromised by circular dynamics, where a major investor also acted as its largest customer, artificially inflating ARR and masking true market demand. The business leaned heavily on the credibility of a high-profile founder with government credentials, rather than demonstrating independent technical strength or validated ML capabilities. Former employees drawing comparisons to Theranos further signaled internal concerns about authenticity and substance. Operational discipline was also critically weak, highlighted by the company being shut down by its cloud provider over an unpaid $18K bill—an extreme indicator of financial mismanagement and lack of basic controls. | The valuation was driven more by narrative and perceived credibility—positioning as an “AI-powered cybersecurity platform” led by a prominent figure—than by proven technology or sustainable revenue. Inflated ARR figures, bolstered by circular customer relationships, created a misleading picture of traction and scalability. The absence of a genuine ML-driven product and reliance on reputation over execution meant that the underlying business lacked defensibility. This resulted in a rapid collapse, with investor capital effectively wiped out as the company entered bankruptcy. | A customer concentration and revenue quality audit would have immediately revealed circular revenue dependencies between investors and customers, invalidating ARR as a reliable metric. A technical diligence review of the ML stack would have exposed the absence of meaningful AI capability, highlighting that the product lacked substantive differentiation. Basic financial and operational checks, including cloud billing and payment discipline, would have flagged severe governance failures early. Finally, an engineering team assessment would have shown over-reliance on founder credibility without a supporting technical organization capable of delivering on the AI claims. |
| Enterprise AI Platform, Nasdaq-listed, $1.9B peak cap, collapsed 94% | 2024 | Enterprise AI | $1.9B peak | AI Authenticity | Caution | The company publicly revealed that its AI model analyzed voice and behavioral cues from claim videos to detect fraud, which immediately raised concerns about illegal profiling and led to rapid backlash and deletion of the statement. Loss ratios consistently exceeded 75%, indicating that the AI underwriting model was underperforming compared to traditional human-driven approaches, a critical signal of flawed model effectiveness. Despite claims that AI could process 30% of claims in seconds, the company never disclosed the underlying data inputs, raising transparency concerns. Additionally, its asserted data advantage—based on proprietary behavioral data—was never independently validated, suggesting the core differentiation was unproven. | The valuation was driven by the narrative of being an “AI-first insurance platform,” positioning technology as a superior replacement for traditional underwriting without demonstrating improved outcomes. At its peak, the company reached a ~$14B market cap despite persistent underwriting losses and questionable model performance. The reliance on unvalidated behavioral data and lack of regulatory clarity introduced significant hidden risks that were not reflected in valuation. This resulted in a substantial correction, with approximately ~$13.4B in value erosion as the market reassessed the viability and compliance risks of the AI-driven model. | A detailed ML workflow review would have revealed that the use of behavioral and voice-based signals lacked regulatory validation and posed high compliance risk. Examination of data sources would have shown reliance on proxy variables that could inadvertently encode protected characteristics, creating exposure to discrimination claims. An architecture-level audit would have identified the absence of bias testing and governance frameworks within the ML pipeline. Together, these diligence steps would have surfaced both the technical and regulatory fragility of the model before investment. |
| AI-Powered Real Estate Platform, $250M raised, shutdown 2022 | 2022 | PropTech | $250M raised | AI Authenticity | Do not proceed | The platform was positioned as orchestrating a broad ecosystem of machine learning models, yet it never disclosed revenue contribution per model, obscuring where real value was being created. Despite claims of scalable AI automation, gross margins fluctuated significantly quarter to quarter, indicating a lack of true operating leverage. The business experienced multiple CEO transitions, suggesting strategic instability, while a secondary AI hiring product was shut down amid broader scrutiny of inflated AI claims. A prolonged 94% stock decline further reflected the market’s loss of confidence in the underlying business fundamentals. | The valuation was driven by the narrative of being a comprehensive enterprise AI platform, despite limited evidence of proprietary technology or sustainable revenue streams. At its peak, the company reached a ~$1.9B market cap without demonstrating consistent margins or scalable economics tied to its AI offerings. Much of the perceived value was based on aggregation and orchestration of external AI capabilities rather than defensible innovation. As the market recognized the lack of differentiation and weak unit economics, approximately ~$1.79B in value was erased. | A technical stack review would have revealed that the core platform functioned primarily as an orchestration layer over third-party APIs rather than hosting proprietary models. A cloud infrastructure audit would have shown that compute and GPU costs were not scaling efficiently relative to revenue growth, undermining claims of automation-driven margins. Additionally, an ML workflow analysis would have exposed that many “AI features” were externally licensed models wrapped in a proprietary interface, rather than internally developed capabilities with defensible advantages. |
| Generative AI Image Platform, $1B valuation, owed AWS $99M — CEO résumé fraud | 2024 | Generative AI | $101M raised | Cloud Cost | Do not proceed | The company claimed to use AI to eliminate buyer agent commissions and streamline real estate transactions, but in practice relied heavily on human agents and coordinators performing the same functions as traditional brokers. The “AI” positioning was largely marketing-driven, with no evidence of automation at the core operational level. This created a cost structure nearly identical to legacy brokerages, but without the pricing power or margins to sustain it. When the real estate market cooled in 2022, the business had no buffer to absorb the downturn and shut down abruptly within weeks, indicating a complete lack of resilience and overreliance on favorable market conditions. | The valuation was built on the narrative of AI-driven efficiency and disruption in real estate transactions, despite the absence of true technological differentiation. Investors effectively priced the company as a scalable software platform, while it operated more like a traditional, labor-intensive brokerage. Without a margin advantage or defensible automation, the business could not justify its capital raised or growth expectations. This led to a total loss of invested capital, with over $250M in value erased as the company collapsed with no recovery. | A product demonstration would have immediately revealed that transaction workflows were dependent on human coordination, with no meaningful AI decision-making layer. Reviewing the engineering organization would have shown a heavy imbalance, with operations and agent staff vastly outnumbering technical personnel, contradicting the AI-first narrative. An ML workflow analysis would have exposed the absence of a proprietary training pipeline, with so-called “AI pricing” relying on basic market comparable lookups rather than predictive modeling. These checks would have made clear that the business lacked the technological foundation required to support its claims. |
| Consumer AI Companion, $1.3B raised — entire value was two people, hollowed out by Microsoft acqui-hire | 2024 | Consumer AI | $1.3B raised | Team Risk | Caution | The company showed severe governance and credibility issues, with the CEO resigning amid allegations of inflating academic and professional credentials—undermining trust in leadership from the outset. Financial discipline was critically weak, highlighted by accumulating approximately $99M in unpaid cloud bills while generating only ~$11M in annual revenue, signaling a massive imbalance between compute costs and monetization. At the same time, multiple senior researchers, including the core team behind its flagship model, departed to form competing ventures, effectively eroding the company’s primary technical moat. These combined signals pointed to both leadership instability and a rapidly deteriorating engineering foundation. | The valuation was driven by the hype surrounding generative AI and the perceived strength of its flagship model, rather than by sustainable economics or organizational stability. At a $1B valuation, the company lacked the revenue base and cost controls to justify its scale, with cloud infrastructure expenses alone far exceeding income. The departure of the core research team further undermined any defensible advantage, reducing the company to a weakened shell of its original promise. As a result, nearly the entire ~$989M in implied value was eroded as investors reassessed both the financial viability and technical continuity of the business. | A basic cloud cost and provider audit would have immediately revealed the disproportionate $99M compute liability against ~$11M in revenue, exposing unsustainable unit economics. An engineering team review, cross-referenced with public profiles, would have shown that key researchers had already exited, signaling loss of core intellectual capital. Additionally, a founder background check would have surfaced inconsistencies in stated credentials, raising early concerns about leadership credibility. Together, these diligence steps would have clearly identified both the financial and organizational fragility of the company before investment. |
| AI Model Fine-Tuning Platform — hyperscalers shipped identical features free, shutdown 2025 | 2025 | AI Infrastructure | Undisclosed VC | Architecture Risk | Do not proceed | The company exhibited extreme concentration risk, with the vast majority of its technical capability and strategic direction tied to just two key individuals. Despite achieving millions of users for its chatbot product, it generated near-zero revenue, indicating a disconnect between user growth and monetization. When a major partner effectively hired away the founders and much of the research team, the remaining organization was left without meaningful product differentiation or technical depth. Regulatory scrutiny further signaled structural concerns, as authorities investigated whether the transaction functioned as a de facto acquisition rather than a standard hiring arrangement. | The valuation was driven primarily by the perceived strength of the founding team rather than by a defensible product or sustainable business model. Investors priced the company as a leading consumer AI platform, despite its lack of revenue and dependence on a small group of researchers for ongoing innovation. Once the core team departed, the underlying value of the business collapsed, revealing that little standalone IP or operational capability remained. Investors ultimately recovered only a fraction of their capital through a licensing arrangement, with no traditional equity exit, highlighting the extent of overvaluation. | An engineering organization analysis would have revealed that over 80% of the company’s technical capability was concentrated in a handful of individuals, creating unacceptable key-person risk. A review of the ML workflow would have shown that continued model development depended almost entirely on the founding research team, with no distributed ownership of knowledge. Additionally, the absence of a documented succession plan or institutionalized research processes would have made clear that the company lacked resilience to team departures, exposing the fragility of its long-term viability. |
| AI Genomics Platform, $6B peak valuation, bankrupt March 2025 — 15M customers' DNA data at risk | 2025 | Genomics / Consumer Health | $900M raised | Data Collapse | Do not proceed | The company operated in a rapidly commoditizing space where major cloud providers and open-source ecosystems were already offering similar fine-tuning capabilities, often at lower cost or bundled into broader platforms. Its user base consisted largely of free-tier developers experimenting with the product rather than enterprise customers generating meaningful revenue. Despite this, infrastructure costs remained high, creating a growing mismatch between usage and monetization. The absence of a clear differentiation point or defensible wedge against hyperscalers signaled that the business was vulnerable to being outcompeted and ultimately displaced. | Any implied valuation was rooted in the narrative of enabling scalable LLM fine-tuning, but without proprietary infrastructure or defensible technology, the company functioned more as an interface layer than a core platform. As hyperscalers and open-source providers introduced similar or superior capabilities, often at minimal or no cost, the perceived value of the offering eroded rapidly. With no enterprise conversion and shrinking margins, the business lacked a sustainable path to profitability, leading to a complete collapse in value upon shutdown. | A technical stack review would have revealed that the core product was primarily an interface built on top of open-source models, with no proprietary training infrastructure or unique capabilities. An architecture analysis would have shown the absence of any defensible moat against cloud providers who could replicate and distribute similar features at scale. Additionally, examining customer metrics would have highlighted a heavy reliance on free-tier usage with negligible enterprise adoption, clearly indicating the classic “API wrapper with no moat” failure pattern |
| AI LiDAR Sensing, $500M raised via SPAC, stock fell 99% | 2024 | AV Sensing Hardware | $2B SPAC valuation | ML Model Risk | Do not proceed | The company’s entire thesis centered on a “proprietary genomic database,” yet customers retained ownership of their data, severely limiting the company’s ability to monetize it. A major data breach exposing millions of genetic records highlighted both security vulnerabilities and reputational risk, undermining trust in the core asset. Leadership behavior also raised concerns, with repeated attempts to take the company private suggesting internal recognition that public market expectations were unsustainable. Meanwhile, the primary monetization strategy—drug discovery partnerships—generated negligible revenue, and the consumer subscription model ($299/year) was fundamentally misaligned with a one-time-use product, leading to weak retention dynamics. | The valuation was driven by the perceived long-term value of a large-scale genomic dataset and its potential for AI-driven drug discovery, rather than by actual revenue generation or near-term commercial viability. At its peak, the company was valued at ~$6B despite limited recurring revenue and an unproven path to monetizing its data asset. Structural constraints, including customer data ownership and low subscription renewal rates, meant the core asset could not generate sustainable cash flow. This resulted in approximately ~$5.9B in value erosion as the gap between narrative and reality became clear. | A data rights and architecture review would have revealed that customer consent agreements restricted the company’s ability to freely monetize its genomic database, undermining the core investment thesis. An ML workflow and commercialization analysis would have shown that the AI-driven drug discovery use case was years away from producing meaningful revenue. Additionally, a customer revenue model assessment would have highlighted the flaw in a subscription approach for a one-time service, with low expected renewal rates. These diligence steps would have exposed that the supposed data moat was not only non-monetizable but also structurally misaligned with the business model. |
| AI Sales Intelligence Platform, $583M raised, 36x revenue multiple — active over-valuation case | 2024 | Sales AI / RevOps | $583M raised | Architecture Risk | Caution | The company is valued at an extreme ~36× ARR multiple, far exceeding typical high-growth SaaS benchmarks of 10–15×, indicating pricing driven more by market hype than fundamentals. Its AI capabilities are largely built on transcription and keyword analysis of sales calls, which are technically straightforward and easily replicable by major incumbents. The insights generated are based on pattern matching rather than proprietary machine learning, limiting differentiation. Additionally, the product integrates directly into existing CRM systems, resulting in low switching costs and making it vulnerable to bundled offerings from larger platforms that can deliver similar functionality at minimal incremental cost. | The valuation reflects the narrative of being a category-defining AI revenue intelligence platform, rather than a defensible or differentiated technology asset. With estimated revenue around $200M ARR, the ~$7.25B valuation implies a multiple that assumes sustained hypergrowth and strong competitive insulation, neither of which is clearly supported. The presence of large incumbents capable of bundling equivalent features into existing ecosystems compresses long-term pricing power and margins. As a result, a more defensible valuation would likely fall in the $3–4B range, implying significant overvaluation relative to realistic SaaS benchmarks and competitive dynamics. | A technical stack review would reveal that the NLP layer relies heavily on third-party speech-to-text providers, with limited proprietary infrastructure. An ML workflow analysis would show that the models are trained on generic sales call data, lacking unique or exclusive datasets that could create a durable advantage. An architecture and integration review would confirm that the product’s functionality is built on standard CRM connectors, offering minimal switching cost and making it straightforward for competitors to replicate or displace. These diligence steps would highlight the absence of a true moat and the reliance on positioning rather than defensible technology. |
| AI Facial Recognition Platform, $30M raised, banned in 5+ jurisdictions — data moat was illegally scraped | 2024 | Facial Recognition / Biometrics | $30M+ raised | Data Collapse | Caution | Entire training dataset scraped from public social media without consent. No consent mechanism, no opt-out, no data lineage documentation. Business model required commercial expansion which regulators blocked in every major market. | $100M+ commercial plans blocked | Data sources documentation would immediately reveal scraping-based training data with no consent framework. Any European or Australian PE/VC would have killed this deal on data sourcing alone. |
| AI LiDAR Sensing for Autonomous Vehicles, $500M raised via SPAC, stock fell 99% by 2024 | 2024 | AI Sensing Hardware / Autonomous Vehicle Perception | SPAC merger at ~$2B implied valuation (2021); stock fell 99%+ to near-zero by 2024; reverse stock splits executed | ML Model Risk + Architecture Risk | Do not proceed | The company projected ~$2.3B in revenue within five years of the SPAC merger, despite generating only single-digit millions at the time, indicating a massive gap between projections and reality. Its “intelligent LiDAR” positioning relied on claims of AI-enhanced perception superiority without providing independent benchmarks or production validation. Revenue assumptions were tied to anticipated automotive OEM design wins that never materialized, making the growth narrative highly speculative. Repeated reverse stock splits to maintain listing compliance, combined with prior leadership history involving underperforming acquisitions, further signaled weak execution credibility. | The ~$2B valuation was driven by the narrative of being a next-generation AI perception leader in autonomous vehicles, rather than by demonstrated commercial traction or validated technology. With minimal revenue and no confirmed large-scale deployments, the business lacked the fundamentals to support such a valuation. The dependence on future OEM adoption, combined with unproven AI performance claims, created a highly speculative valuation foundation. This resulted in near-total value erosion, with the stock declining over 99% as projected growth failed to materialize. | An ML workflow review would have revealed that the perception AI capabilities were largely theoretical and not validated in real-world production environments. Customer and revenue analysis would have shown zero meaningful income from the claimed OEM partnerships, undermining the credibility of forward projections. An architecture assessment would have exposed the heavy dependence on hardware manufacturing at scales far beyond current capacity, requiring exponential production growth to achieve viable economics. These diligence steps would have clearly highlighted the disconnect between technological claims, commercial readiness, and valuation. |
Access Restricted
Enter your email to unlock AI failure insights.