Invisible

Business process outsourcing (BPO), the practice of contracting non-core business functions to third-party providers, has operated on the same fundamental model for decades. In the 1990s, major financial institutions relocated back-office functions to lower-cost labor markets such as Manila and New Delhi, giving rise to the first wave of the modern BPO industry, which was estimated at $328.4 billion in annual spend in 2025. As of 2025, this model, often dubbed "lift and shift," was ubiquitous across large industries like banking and financial services, healthcare, hospitality, logistics, and retail.

In 2005, Amazon's Mechanical Turk spurred a second wave, proving that discrete cognitive tasks could be decomposed and distributed across a global workforce at internet scale. Where the first wave moved work geographically, Mechanical Turk broke work down into Human Intelligence Tasks, like identifying textures in a photo (e.g., metal, water, hair), and routed them to whoever could complete them cheapest. The outsourcing industry's billing model did not change with it. Outsourcing vendors still bill by the hour, which gives them little incentive to automate. The less efficient the process, the more they earn.

Source: a16z

The third wave of BPO is breaking that pattern, as AI has created a clear "why now" moment. By 2025, foundation models had improved at data extraction, deep research, and complex reasoning, while voice AI agents had begun taking a share of high-volume call center work like customer service and scheduling. These capabilities cover work that software previously could not handle adequately. BPOs tend to be older incumbents that lack cutting-edge technology and operate within well-defined categories with existing budgets, making them prime disruption targets given the market need, available budget, and legacy competition.

Moving AI into production is the harder step. A March 2025 study found that 88% of AI proofs of concept never reached wide-scale deployment. The models themselves also create demand for the same kind of expert human work. Surge AI, which sells human-generated training data to AI labs, reportedly generated over $1 billion of revenue in 2024.

Invisible Technologies sells AI training data to the labs that build foundation models and deploys AI inside enterprises, pairing its own software with a network of human experts. Its enterprise platform decomposes workflows into discrete steps, automates what it can, and routes the remainder to domain specialists through Meridial, its marketplace of over 24K vetted experts spanning law, finance, coding, and STEM. Where legacy BPOs profit from inefficiency and pure software platforms sell the tool without executing the work, Invisible owns both layers, combining proprietary software for data readiness, workflow mapping, model evaluation, and agent execution with a human layer that handles what automation still cannot.

Francis Pedraza (Chairman) founded Invisible Technologies in 2015. Pedraza graduated from Cornell, and his first company, Everest, was a personal goal-tracking app backed by Peter Thiel that shut down in 2014 after its users stopped returning to it. He then built his next company around an observation that software could automate almost anything in theory, but every workflow still contained messy edge cases that required human judgment. Rather than build another pure software product, Pedraza set out to combine automation with on-demand human expertise.

In its early days, Invisible was an elite executive-support service for CEOs. When pitching investors, Pedraza found that VCs pushed back on the services label. Rather than reframe the model to fit investor preferences, he concluded that he had found a blind spot in the market. The model's enterprise potential became clear in March 2020, when DoorDash asked Invisible to help import restaurant menus and pricing during the COVID-19 pandemic. Invisible's case study on menu digitization for a delivery platform during the pandemic reports that it accelerated onboarding speed by 233%.

Invisible raised its initial funding rounds between 2018 and 2021, but from there took a different path from most venture-backed startups. Pedraza borrowed against the company to buy out its early venture investors, and as of February 2025, roughly 300 current and former employees held 55% of the company.

Invisible appointed Matthew Fitzpatrick as CEO in January 2025, succeeding Benjamin Plummer, while Pedraza remained chairman. Fitzpatrick was previously Global Head of QuantumBlack Labs at McKinsey, where he oversaw 1K engineers and product leaders across QuantumBlack Labs and McKinsey and helped organizations scale AI projects from proof of concept to production, the same work Invisible sells. Invisible described the appointment as part of its shift to a software-driven company anchored around its AI process platform.

Pedraza himself is involved with numerous companies and organizations, despite his role as Chairman of Invisible. Since May 2026, Pedraza has also served as CEO of Exponential Capital, and serves as Chairman of a number of other companies and foundations.

Invisible describes itself as an applied AI research company with two halves that feed each other. Its Data Lab sells training data, evaluations, and reinforcement learning (RL) environments to the labs building foundation models, and its enterprise platform deploys AI inside customer operations. Pedraza calls the model "AI services."

Across both halves, Invisible breaks down customer workflows into discrete steps, automates as many as possible, and routes the rest to human experts through Meridial, its expert network. The enterprise platform consists of five surfaces, which the site lists as Neuron for data readiness, Atomic for workflow mapping, Meridial for human expertise, Synapse for model evaluation, and Axon for agent execution. Forward-deployed engineers connect these to a customer's systems, starting with one workflow and expanding from there.

Data Lab

Source: Invisible

The Data Lab is Invisible's AI training business and, as of October 2026, the first product listed on its site. It offers RL environments, agentic and tool-use training, custom evaluations, supervised fine-tuning, reinforcement learning from human feedback, and multimodal and multilingual training data. Its RL environments are built from real workflows in coding, accounting, banking, legal, and compliance, with rewards graded by rubrics that domain experts design.

Invisible also sells off-the-shelf datasets, including GDPval+, a set of multi-step professional tasks scored by expert rubrics, and SWE Bench Pro, a set of agentic coding environments. The company claims to have "trained over 80% of the world's leading AI models." For the Data Lab, the company cites access to over 22 million experts across 800 domains through its Meridial network, a reach figure that is distinct from the over 24K vetted experts its homepage lists for the enterprise platform.

Neuron

Source: Invisible

Neuron is Invisible's data platform. It ingests messy, multi-source data across enterprise systems and organizes it into a governed, AI-ready layer, so that models and agents know where to look without custom engineering for each deployment. The Orchestra of St. Luke's, a New York orchestra, used Neuron to unify ticketing, fundraising, and CRM data from three systems into a single patron view in under three weeks. For enterprises where fragmented or unstructured data is the primary barrier to AI adoption, Neuron functions as the prerequisite layer that makes everything else possible.

Atomic

Source: Invisible

Atomic is Invisible's process builder. It maps existing manual workflows into a visual, step-by-step architecture, giving enterprises a documented baseline before automation begins. Rather than ripping out legacy processes, Atomic allows organizations to start with their existing operations and incrementally replace manual steps with AI. Invisible reports that Atomic reduces process complexity, measured by the number of steps, by an average of 80% versus legacy workflows.

Meridial

Source: Invisible

Meridial is the expert contractor marketplace that Invisible deploys on demand to handle tasks that AI agents cannot execute autonomously. It draws from a pool of over 24K vetted domain specialists, organized by skill, language, and expertise, and matches them to projects, including AI training projects from frontier labs, across law, STEM, finance, coding, linguistics, and safety. This human layer integrates directly with the platform so that every AI system Invisible deploys is backed by experts who handle edge cases, label training data, and stress-test model outputs. In March 2026, Invisible agreed to acquire WeCP, a skills assessment platform that had conducted over 2 million technical interviews and built over 18K assessment frameworks, to strengthen expert validation for high-precision workflows.

Synapse

Source: Invisible

Synapse is Invisible's evaluation layer. It runs any AI system through realistic work scenarios using criteria defined by the client, including accuracy, safety, tone, or domain fit, and compares model performance side by side, surfacing where models pass, fail, and why. The result is ongoing evaluation tied to actual use cases rather than a static benchmark score based on someone else's test. Clients can also reuse these frameworks. For example, if a client is choosing between vendors or validating models built in-house, Synapse serves as a repeatable measurement framework so clients don't need to rebuild the same test each time.

Axon

Source: Invisible

Axon is Invisible's agent deployment engine. It deploys AI agents that execute complex, multi-step workflows, using tools and writing code as needed, while maintaining full auditability and governance controls. Every workflow is versioned, trackable, and separated by client, allowing enterprises to test safely and promote what works. Invisible reports that Axon has reduced workflow processing time across clients by over 70%.

Customer

Source: Invisible

Source: Invisible

Invisible's customer base has shifted considerably since the company's early days. Initially serving CEOs only, the model generated immediate demand, but its costs were double what it earned in revenue. Over the following year, Invisible built 50 processes for a single client, which drove a pivot from executive support to running any business process.

Invisible's customer base spans three broad segments: foundation model builders, large enterprises, and the public sector. Invisible has trained foundation models for Microsoft, AWS, and Cohere, work that depends on human-in-the-loop (HITL) review by domain experts. For Cohere, Invisible's expert annotators ran blind human evaluations of its Command A model and supported its expansion into 10 languages. Thomson Reuters used Invisible to build the annotation schema and expert review workflow for a document layout model in under two weeks, after which task volume grew 2.5x.

Large enterprises deploy Invisible for data operations and back-office automation across industries. Nasdaq credits Invisible with saving over 10K developer hours through a data integration project, and Grubhub cut its menu extraction time by 75-80% with Invisible's agents. SwissGear used the platform to structure and unify its data for demand forecasting, and the Charlotte Hornets used Invisible's computer vision platform to extract player-movement data from game film for the team's 2025 draft analysis. Invisible's site also features American Family Insurance and Headway as notable clients. In the public sector, Invisible earned "Awardable" status on the Department of Defense's Tradewinds Solutions Marketplace in April 2025, and it supported SAIC and Vatn Systems in a joint US Navy exercise applying AI to unmanned underwater vehicles.

Market Size

Invisible competes for three pools of spend: AI training data, intelligent process automation, and the outsourced work BPOs perform. The first is data annotation and AI training services, where it goes up against Scale AI and Surge AI for foundation model builder contracts. The data collection and labeling market was estimated at $4.9 billion in 2025 and is expected to reach $13.8 billion by 2030 at a CAGR of 23% over the period, though annotation represents only one task within the broader workflow delivery layer Invisible is trying to own.

The second is intelligent process automation, the infrastructure layer that sits between enterprise AI investment and actual production output. In September 2025, Fitzpatrick noted that 70% of the world's software was over 20 years old and that most enterprises were struggling to get AI models into production and quantify ROI. This category was estimated at $14.6 billion in 2024 and is projected to reach $44.7 billion by 2030. The third is the BPO market itself, the spend Invisible's enterprise platform competes for. Global BPO spend was estimated at $328.4 billion in 2025 and is projected to reach $695.8 billion by 2033.

Invisible competes on two fronts. In AI training data, it sells to the same foundation model builders as Scale AI, Mercor, Turing, and Surge AI, a market that shifted in June 2025 when Meta's investment in Scale AI led Google to plan to cut ties with Scale AI. In enterprise AI operations, it competes for budgets held by annotation-focused HITL vendors and by BPO incumbents that bill by headcount. Invisible's expert network supplies the first market and its software platform the second.

AI Data Labeling and Model Training

Scale AI: Founded in 2016 by Alexandr Wang and Lucy Guo through Y Combinator, Scale AI offers data infrastructure and model evaluation tools for training generative AI, covering data labeling, reinforcement learning from human feedback, LLM fine-tuning, and safety testing for frontier model labs and enterprises. In June 2025, Meta reportedly invested $14.3 billion for a 49% non-voting stake in the company, valuing the company at $29 billion. As of October 2026, Scale AI had raised a total of $15.9 billion. Wang left to join Meta, and Francis deSouza became Scale AI's CEO in August 2026.

Meta's stake prompted customers to leave. Google, which had planned to spend about $200 million with Scale AI in 2025, reportedly planned to cut ties, and OpenAI said it was winding down its work with Scale AI because it needed more specialized data expertise, a shift it had begun before the Meta deal. Scale AI describes itself as running two core businesses, data for frontier labs and applications built on those models, the same split as Invisible's, with $15.9 billion raised to Invisible's $107.9 million as of October 2026.

Mercor: Founded in 2023, Mercor runs a marketplace that supplies domain experts to AI labs for training and evaluation work, the closest analog to Invisible's Meridial. In October 2025, Mercor raised a $350 million Series C led by Felicis at a $10 billion valuation, and as of October 2026, it had raised a total of $483.6 million. Mercor competes with Invisible's Data Lab for the same lab budgets.

Turing: Founded in 2018, Turing started as an AI-powered platform for hiring and vetting remote software engineers before expanding in 2022, when OpenAI engaged the company to generate training code for improving LLM reasoning. Turing offers LLM fine-tuning, reinforcement learning from human feedback, and enterprise AI deployment services alongside its talent marketplace. In March 2025, Turing raised a $111 million Series E led by Khazanah Nasional Berhad, valuing the company at $2.2 billion, and as of October 2026, it had raised a total of $270.2 million. Turing's trajectory most closely mirrors Invisible's services-to-platform evolution, but its go-to-market remains concentrated in model training and coding evaluation rather than the broader enterprise workflow automation space Invisible targets.

Surge AI: Founded in 2020, Surge AI sells human-generated training data to AI labs and grew without venture funding. It reportedly generated over $1 billion of revenue in 2024, and in July 2025 it was reported to be planning a first funding round of about $1 billion of funding at a valuation above $15 billion. Surge AI competes with Invisible's Data Lab for the same lab budgets.

HITL AI Services

Sama: Founded in 2008 as Samasource, Sama offers data annotation services covering image, video, lidar, and language labeling with a workforce based primarily in East Africa and India. Sama raised a $70 million Series B in November 2021, and as of October 2026, it had raised a total of $84.8 million. Unlike Invisible, Sama is focused narrowly on annotation and does not offer workflow automation, agent deployment, or embedded engineering services.

CloudFactory: Founded in 2010, CloudFactory offers HITL AI data services covering data labeling, annotation, and model monitoring, combined with a managed global workforce across four continents. CloudFactory raised a $65 million growth equity round led by FTV Capital in November 2019, and as of October 2026, it had raised a total of $78 million. Its valuation is not publicly disclosed. CloudFactory's WorkStreams subscription model creates predictable recurring revenue and embeds it into client operations, similar to Invisible's engagement model, but the company lacks a proprietary software platform and does not offer agent deployment or enterprise workflow automation.

BPO Incumbents

Cognizant: Founded in 1994 as an in-house technology unit of Dun & Bradstreet and taken public in 1998, Cognizant offers IT services, business process outsourcing, and AI-led automation spanning customer experience, back-office operations, and knowledge process outsourcing across financial services, healthcare, and enterprise clients. Cognizant had a market cap of $28.4 billion as of October 2026, and its full-year 2025 revenue was $21.1 billion, up 7% year-over-year. Cognizant competes with Invisible on knowledge process and back-office AI operations but relies on headcount-driven delivery and multi-year enterprise contracts, positioning it as a cost-based incumbent rather than a platform play.

TaskUs: Founded in 2008, TaskUs offers outsourced digital services spanning customer experience, content moderation, and AI operations, including data labeling and annotation. A take-private offer from Blackstone and TaskUs's co-founders, announced in May 2025, failed a stockholder vote in October 2025, and TaskUs remained public. It had a market cap of $730 million as of October 2026, and its full-year 2025 revenue was $1.2 billion. TaskUs competes with Invisible on back-office AI operations but operates primarily on multi-year, headcount-based contracts with no proprietary software layer, making it a cost-based competitor rather than a platform play.

Invisible operates as a managed AI operations layer, charging customers to run AI work for them rather than selling standalone software licenses. Invisible has not publicly disclosed its pricing model, but its revenue comes from the two halves of the business it describes on its site:

1. AI training for model builders, where labs pay Invisible for training data, evaluations, and RL environments produced with its expert network.

2. Enterprise deployments, where companies pay Invisible's forward-deployed engineers and platform to operate workflows covering data preparation, automation, agent execution, and human oversight.

The cost structure is designed around variable rather than fixed labor. Rather than carrying a large full-time workforce, Invisible deploys experts from Meridial on demand, allowing its human layer to scale up or down with client volume without fixed headcount costs. This model resembles how traditional BPOs operate, with the key difference that Invisible's automation layer is meant to absorb more of each workflow over time, so the marginal cost of handling additional volume should decrease as the platform matures. The company argues that each deployment shows where models break, and that fixing those gaps makes the next deployment better.

Invisible's revenue grew from $3 million in 2020 to $134 million in 2024. The company was profitable before its September 2025 growth round, with $15 million of EBITDA on that 2024 revenue. Invisible runs on a remote-first model with a small full-time team and a large contractor network. It had a team of 350 as of September 2025, and its About page listed a team of over 250 as of October 2026. Meridial, by contrast, lists over 24K vetted experts, which lets the workforce scale with client demand without adding fixed headcount.

In April 2026, Invisible raised an undisclosed amount from Golden Gate Ventures' MENA fund, aimed at enterprise deployments in the Gulf states. Before that, in September 2025, Invisible raised $100 million in a growth round led by Vanara Capital, a firm spun out of TPG. The September 2025 round valued the company at over $2 billion, with participation from Princeville Capital, HOF Capital, Freestyle VC, Acrew Capital, Greycroft, and Deepwater Asset Management. That valuation was four times the $500 million at which the company was valued in early 2024.

Prior to these rounds, Invisible raised $2.6 million and $5.3 million in seed funding in October 2018 and June 2019, respectively, plus an undisclosed seed round in May 2021. As of October 2026, it had raised a total of $107.9 million in disclosed rounds, although the company put its total capital raised at $144 million after the September 2025 round. Invisible reached $134 million of 2024 revenue on less than $8 million of disclosed seed funding before its first large round, an unusual funding history for a company of its size.

At its reported September 2025 valuation of over $2 billion, Invisible was valued at more than 14.9x its $134 million of 2024 revenue. Comparable public companies span both legacy BPO incumbents and enterprise software businesses, depending on whether Invisible successfully transitions from a services-led to a software-driven model. As of October 2026, legacy BPO incumbents traded at revenue multiples ranging from 1.3x for Cognizant to 2.5x for TCS, making Invisible's multiple a significant premium to services businesses.

Source: Koyfin

Against enterprise software comparables, multiples ranged from 3.2x for UiPath to 75.6x for Palantir as of October 2026, and Palantir's forward-deployed enterprise model is the closest public analog to Invisible's go-to-market. At 14.9x, Invisible sits near the low end of that range. If growth slows or the shift to software-driven revenue stalls, its September 2025 valuation of over $2 billion becomes difficult to defend against BPO comparables trading at 1-3x revenue.

Source: Koyfin

Packaging Repeat Deployments Into Solutions

As Invisible scales across verticals, it has begun turning its most repeatable deployments into packaged products rather than bespoke builds. As of October 2026, its site listed packaged solutions for back-office automation, contact centers, computer vision, demand forecasting, and AI training, sold across ten industries. The back-office product combines document ingestion, process mapping, and custom agents with human review of uncertain decisions, and the contact center product runs automated quality assurance on 100% of interactions.

Invisible's engagements start with one workflow and expand from there, and converting proven delivery workflows into reusable modules reduces the cost and time to stand up new client engagements, which in turn lowers deployment friction and increases the number of accounts Invisible can serve simultaneously. Each deployment that gets standardized raises the ceiling on how many accounts the team can manage without proportional headcount growth, improving average contract value and unit economics over time. The forward-deployed engineering model was built for customization, and the opportunity is to identify which parts of that customization are universal and systematize them.

Selling to Defense Agencies Through Tradewinds

Defense and intelligence agencies have large, persistent operational workflows like logistics, intelligence processing, document analysis, and communications that map to what Invisible already does for commercial enterprises. Invisible holds "Awardable" status on the Department of Defense's Tradewinds Solutions Marketplace, which it earned in April 2025 and which lists solutions that defense buyers can procure through the marketplace. The company also sells through Carahsoft on the NASA SEWP V and NASPO ValuePoint contract vehicles. As of September 2025, Invisible had supported SAIC and Vatn Systems in a joint US Navy exercise applying AI to unmanned underwater vehicles, work that applies its data and evaluation tools to the kind of operational analysis defense agencies run at scale.

Valuation Compression

Venture investors tend to discount services firms and systems integrators because their revenue is less recurring than software revenue, even though, as of 2025, such firms were doing much of the work of getting AI adopted inside enterprises. Invisible sits squarely in this category. Its delivery model depends on a managed network of human operators and forward-deployed teams rather than a self-serve software product with predictable seat-based revenue.

Invisible's reported valuation of over $2 billion implies a multiple closer to software than to services. Holding that multiple requires demonstrating a credible path to recurring, software-driven revenue from its platform (Neuron, Atomic, Synapse, and Axon). If the market re-rates Invisible as a services business rather than a software company, the valuation gap closes quickly and unfavorably for existing investors.

Competition From Better-Capitalized Rivals

Invisible operates in markets that better-capitalized rivals are moving into aggressively. Scale AI has evolved from a data annotation provider into a broader AI infrastructure company, with services spanning model evaluation and enterprise software for building and deploying AI applications. Scale AI's enterprise business closed its strongest bookings quarter in Q4 2025, adding customers including Mayo Clinic, BP, and Allianz. On the lab side, Mercor was valued at $10 billion in October 2025, five times Invisible's valuation.

Invisible's differentiation rests on its forward-deployed engineering model and customization, but those advantages are difficult to communicate in a sales cycle and possible to replicate as competitors mature. Scale AI launched Scale Labs in March 2026 as a research hub whose remit includes how advanced systems behave in high-stakes production environments, the territory Invisible occupies. In a December 2025 survey, enterprise VCs broadly predicted that enterprises would spend more on AI in 2026 through fewer vendors. That dynamic favors incumbents with larger balance sheets and broader enterprise relationships over newer entrants.

Contractor Classification Across the Expert Network

Invisible's model depends on independent experts. Meridial lists over 24K vetted experts who work on demand rather than as employees, and rivals with the same structure have been sued over it. Scale AI faced three worker lawsuits between December 2024 and January 2025, including claims that workers were misclassified as contractors, and Surge AI was hit with a proposed class action alleging misclassification in May 2025. No such suit against Invisible had been reported as of October 2026, but a ruling that reclassifies expert contractors as employees would raise the cost of the variable labor that keeps Invisible's human layer flexible.

Enterprises have relied on the same outsourcing model for three decades, moving work to cheaper labor markets, billing by the hour, and collecting the spread. That model gave vendors little reason to automate, since inefficiency was the product. Foundation models can handle tasks that previously required human judgment at scale, yet a study reported in March 2025 found that 88% of AI proofs of concept never reached wide-scale deployment. The bottleneck lies in the infrastructure required to evaluate, train, and operationalize AI against real enterprise workflows, a gap that legacy BPOs cannot close and that pure software platforms were not built to fill.

Invisible Technologies works on both sides of that gap. Its Data Lab sells training data and evaluations to the labs building foundation models, and its enterprise platform combines software for data preparation, workflow mapping, model evaluation, and agent deployment with Meridial, a marketplace of over 24K vetted domain specialists who handle what agents still cannot. The company reached $134 million in revenue in 2024 and raised its first large outside round, a $100 million growth round, in September 2025 at a valuation of over $2 billion.

Invisible faces pressure from better-capitalized rivals, including Scale AI and Mercor, and from the gap between its services-heavy revenue mix and a software-level valuation. Its future success will be determined by its ability to demonstrate a credible services-to-software transition and convert a fragmented, legacy-dominated BPO market into a durable software revenue base.