The business process outsourcing (BPO) market is massive. The industry’s market cap reached over $300 billion in 2024 and is forecasted to be over $525 billion by 2030.
Enterprises lean on BPOs because they provide a cost-effective way to perform necessary and high-volume, yet mundane and repetitive work – like customer support, outsourced IT, and financial claims processing – that they do not want to handle themselves. The work BPOs do is important, but the experience of working with them is far from seamless. BPOs can have prolonged turnaround times for their work output, can be prone to human error since their employees lack individual accountability, and can be incapable of completing certain tasks satisfactorily because they lack the context and authority to do so. All of these factors culminate in a less efficient and often frustrating experience for the end customer. With AI, startups can now give customers the best of both worlds and enable enterprises to in-house their own customer experience and back-office operations in a high-quality, scalable, and cost-efficient way.
We believe there is a clear opportunity with AI to productize and unbundle the BPO. This is exciting for several reasons. From a technical perspective, there’s a clear “why now”: modern AI has become exceptionally good at handling work that couldn’t previously be done adequately with software. Core foundation models are rapidly getting better at data extraction, deep research, and complex reasoning, while voice AI agents are mature enough for large-scale production with browser agents soon to follow. From a business perspective, BPOs tend to be older incumbents that lack cutting-edge tech and operate in a well-defined category of work with clear, existing budgets; this makes them prime disruption targets for startups given the proven market need, available budget, and legacy competition.
The BPO Business Model and How AI Agents Disrupt It
Running a large enterprise operation requires an enormous amount of repetitive, transactional work, whether it’s data entry, call center operations, revenue cycle management, invoice reconciliation, or payroll processing. The work is gritty and behind-the-scenes, operationally complex, and not the core competency of the business. The work can also be inconsistent and seasonal, such as with increased customer service needs around the holidays; annual employee turnover for some functions can be as high as 30-40%.
The pain of managing this complexity — paired with the financial and operational cost of recruiting, hiring, and training in-house employees for this work — is why BPOs are such enormous businesses today. BPOs like Cognizant, Infosys, and Wipro reported revenues of between $10-20 billion in their latest fiscal years, respectively. BPOs are also ubiquitous across large industries like banking and financial services, healthcare, hospitality, logistics, and retail. In fact, some industries have such specific needs that numerous vertical-specific BPOs have emerged, including freight audit pay businesses to manage transportation audits and payments, third-party administrators (TPAs) to process insurance claims, and revenue cycle management (RCM) firms to help healthcare providers manage medical billing and collections.
While these BPOs do important work and serve large and impressive businesses, most were founded decades ago — some date back to the 1940s — and rely on deep customer relationships and archaic systems integrations rather than cutting-edge technology. For many years, this was the best option for enterprises given the limitations of software.
But now, modern AI makes productizing the BPO and bringing this work back in-house possible, thanks to:
- Rapidly improving general model capabilities: LLMs are rapidly improving at the typical tasks BPOs specialize in, such as unstructured document processing, data reconciliation, knowledge search, reasoning capabilities, and tool use. We expect the models to continue to improve and further productize the work BPOs do.
- Meaningful voice AI capabilities: Voice AI has proven to be a key 0-to-1 capability for unlocking core use cases previously limited by traditional software. New software — often powered by AI infrastructure companies like ElevenLabs, OpenAI, or Cartesia — has improved dramatically in recent years and will soon make AI agents virtually indistinguishable from human ones.
- Emerging browser technology: Anthropic’s computer use model, OpenAI’s Operator, and Google DeepMind’s Project Mariner preview how AI agents will be able to handle desktop and browser-based tasks. This opens up the opportunity for AI to be adopted outside of a specific app and across desktop, browser, and local applications. We believe this will be a key emerging property that people will leverage to build differentiated new agent experiences across heterogenous software surfaces.
With these capabilities, AI agents can operate at the speed of software, work 24/7, adapt to any important cultural norms, communicate in any language, and scale infinitely across a full customer base — with limited human participation needed. And because it’s much more scalable for companies to deploy AI than hire in-house or outsourced labor, AI agents have actually grown the market. Companies can now scale agents across more products, customers, and needs instead of only serving cohorts that made sense on a unit economics basis.
The Impact of AI on BPOs and Where the Opportunity Lies
AI-native companies are already going after BPO spend, and many of them are growing at unprecedented rates. These companies know they need to build defensible products to stay ahead of both incumbent and startup competition, but the early evidence from customers — both the high demand and the evident customer love — mean that they’ll have a clear opportunity to build deeper product workflows and create enduring moats in the process.
Front-office customer experience
One of the clearest opportunities for AI startups is in customer support and customer experience, which makes up the largest subsegment — over $100 billion — of BPO spend.
Everyone has felt the pain of an automated customer service experience gone wrong. If your request doesn’t have an easy, pre-defined answer and resolution, you’re almost always shunted into a flow that involves calling an automated number and then demanding to speak to a human agent, futilely emailing a customer service email alias, angrily arguing with a chatbot that has no context, or some combination of the three.
With AI agents, companies can now provide first-class, in-house customer experiences across all modalities (text, email, voice) at the speed, quality, and scalability of software. These agents can work 24/7, resolve queries in any language, and respond instantaneously without queues or wait times — all without needing to hire, train, and retain in-house headcount against inconsistent demand. Companies like Decagon, for instance, have built AI support agents that have shown upwards of 80% resolution rates and improved CSAT scores for customers out of the gate. Decagon’s AI agents have also allowed customers to offer support across more users and products because they’ve made it cost-efficient to do so.
Industry-specific AI agents are also successfully productizing their industry’s core BPO use cases. In auto lending, Salient’s AI voice agents allow for high-volume customer intake and collections calls that can efficiently call a broad customer base while staying up-to-date on relevant compliance regulations. In home services, Avoca allows customers to productize the off-hours or overflow calls their own in-house staff can not handle (which they previously handed off to outsourced call centers). These vertical use cases are especially interesting because of the product complexity, platform integrations, and industry-specific regulatory requirements they address. These characteristics make it more difficult for horizontal players or foundation models to easily subsume these use cases, giving these startups a more enduring moat over time.
Back-office operations
AI startups are also meaningfully chipping away at BPO spend in back-office operations.
So many enterprise workflows are ultimately about taking messy, unstructured data from disparate systems, and then ingesting, normalizing, and reconciling that data. This is mundane and repetitive work, and companies have taken different approaches to handling it. Some hire operations employees, some use robotic process automation (RPA) solutions, and some outsource it to BPOs. We’ve written about how brittle RPA solutions will be productized through AI agents, and we think the same will happen for outsourced BPO spend.
We are already seeing this happen across different industries. The transportation industry, for instance, must manage and reconcile billions of dollars worth of invoices to reduce fraud and error and ensure that every party in the supply chain is paid correctly. While previously done manually by large freight audit pay firms like Cass and Green Mountain, Loop and others can now productize this invoice reconciliation, claims management, and cost allocation process using AI. And in healthcare (one of the largest industries to leverage BPOs), companies like Camber have applied gen AI to revenue cycle management and demonstrated clear efficiency gains; one customer even saw 80% fewer denials on first submissions and 50% time savings on claims processes and billing without an increase in cost.
Application development and generation
Lastly, many companies are indirectly eating into the BPO market, even if they don’t directly go after defined BPO spend. This is because a large BPO use case is building custom applications using outsourced engineering resources, in cases where the enterprise didn’t have the resources or capabilities to do so themselves.
Now, coding assistants like Cursor can allow enterprises to increase the output from their existing engineering workforce and dramatically improve the productivity of individual developers, which can empower enterprises to develop a lot more of these applications in-house. And with the rise of AI-powered web app builders, non-technical users will soon be able to build internal applications without needing to write code themselves. Both of these capabilities mean that in the future, there will be much less of a need to outsource application development to a BPO.
Distribution vs Innovation: What to Expect from the BPOs
Every startup building in this space has a legacy incumbent: the BPO itself. Unsurprisingly, these BPOs – like most enterprises – have taken note of the AI tidal wave and have announced initiatives of their own. Wipro COO Sanjeev Jain said that they’ve seen a 140% increase in AI adoption in existing projects, and Infosys announced that they have more than 100 new generative AI agents deployed within their client base. And Accenture, which specializes in both consulting and outsourced work, recently announced $1.2 billion in new bookings alone for generative AI projects.
There’s a natural battle between incumbents who have distribution and startups who have innovation. Incumbents will surely capture some of the value in this case, but we believe startups are advantaged over BPOs for a couple reasons:
There is a fundamental business model mismatch between building AI-native products and running a BPO. Most BPOs charge on a time-and-materials billing model and then charge a ~20-30% markup on that labor: their business model depends on employing people and selling the output of that labor to customers. Overhauling that model to become a product-first AI-native business would be a monumental undertaking that would dramatically compress their margins, kill their current cash cows, and distort the company culture. It would be an exceptionally difficult transformation for any company to make, let alone a public company that would be heavily scrutinized by the public markets.
The knowledge of how to leverage state-of-the-art AI models is not evenly distributed. A deluge of cutting-edge AI models, tools, and research is being released every day — to the point where even highly technical AI experts are overwhelmed. Companies need first-class, AI-native teams to stay on top of the latest research breakthroughs and understand how to leverage it for business users — a rare combination of talent not typically found in a BPO.
This opportunity for startups to disrupt the BPO has an expiration date, however. It’s naive to assume that legacy BPOs will not eventually act on these opportunities themselves. As the foundation model layer stabilizes and becomes more comprehensible to a broader audience, BPOs will be able to incorporate AI more easily into their business, and they will undoubtedly promote their own in-house AI products to their long-time customers. Startups looking to capitalize on this moment are thus best served by:
- Being able to prove an exceptionally clear ROI: Selecting a use case and industry with an exceptionally clear ROI case against the status quo will accelerate adoption in many industries. The reason we’ve seen such rapid adoption in AI voice and customer support in particular (such as in the case of Decagon) is because voice AI is an evident step-function advancement that buyers can clearly understand, and customer support has clear, quantifiable metrics of success to measure against (i.e., resolution rate and CSAT score). As a result, startups can easily prove their ROI and make the decision to purchase a straightforward one.
- Being heavily customer-first and forward-deployed for early customers: It’s always important to be customer obsessed. But because one distinct advantage BPOs sell is their custom services and bespoke systems integrations, we think it’s especially critical for AI startups looking to replace that work to emulate that level of customer-first deployments, especially at the beginning of an engagement. This is not only to deliver white glove-like outcomes akin to what BPOs offer, but it’s also to reassure customers nervous about using a fundamentally new technology. The Salient founders, for instance, actually moved cities for a few months to be closer to one of their early customers so they could deploy alongside them in person. This paid enormous dividends given the successful customer rollout and the learnings from the implementation process. While the goal long-term should be to productize and make the onboarding scalable, this level of initial upfront work is understandable and perhaps even necessary.
Building an AI-Native Full-Stack Firm
Many startups are productizing the BPO, but others are taking on the services themselves. Some businesses are taking a private equity-like roll-up strategy, where they buy assets and use AI to make them more efficient. Others are trying acquisition-as-distribution mechanisms to acquire customers that can prototype the software product they plan to make their long-term business. Yet other companies are building a new full-stack firm from the ground up to compete with existing BPOs on speed, cost and quality, embedding both AI and human judgment into their core DNA.
Tech-enabled businesses have historically been challenging because it is difficult to both build a first-class AI product and run a gritty operational business, but modern AI has made it easier to explore where a full-stack approach may be viable. These include:
Tackling industries that are reluctant to purchase software altogether: Some industries are simply late adopters and prefer buying outcomes instead of products; for these industries, it’s possible that delivering the outcome and leveraging AI internally is the optimal approach, as Foundation is trying to do for the insurance market and Long Lake is trying to do for HOAs. These companies will naturally face the traditional challenges of operating a tech-enabled service – namely, that it’s difficult to build first-class tech and run a first-class operational business – but if they can nail the motion, there are likely valuable businesses to be built.
Starting full-stack as a wedge into an eventual software play: Some companies plan to be a software-only business long-term, but start by making small service provider acquisitions to secure an initial customer base. They plan to use that base to observe the manual workflows and learn how to automate them with AI. While unconventional, this strategy could be an attractive way to get initial distribution and have a friendly prototyping environment for the software product, while still reaping the benefits of being a software business long-term.
Enabling the creation of frontier AI itself: Frontier AI is a category where human involvement is intrinsically important given the human judgment required to develop and train models. Scale — last reported to have a $13.8 billion valuation and one of the fastest-growing private companies today — in many ways started as the infrastructure for these novel AI needs. Its first major product was high-quality data for autonomous vehicle companies like Waymo, which required perfectly-annotated LiDAR and other sensor data to improve their self-driving capabilities. Scale sold outcomes (i.e., data that improved customer models) while building cutting-edge internal tools to do so efficiently and at high quality. With the advent of generative AI, they have since expanded to providing reinforcement learning with human feedback (RLHF) data — a critical component to training the most advanced foundation models — to the AI labs. In both of these cases, BPOs didn’t have existing offerings, failed to recognize the need early on, and lacked the technical underpinnings required to serve these needs, which only continued to grow in complexity. This left Scale well-positioned to capture the opportunity and obviated the need for a traditional outsourcing firm altogether. This path is rare since there are only so many frontier capabilities to serve (with Scale poised to capture many of them) — but it is an enormously valuable one if done correctly and at the right moment in time.
Conclusion
We think there are many massive companies to be created that subsume the work that BPOs do. If your main competition is a BPO, we’d love to chat.
