The most revealing thing about EXL's artificial intelligence business is where it begins. Not with a chatbot. Not with a moonshot model. It begins in an insurance claim, a bank transaction, a hospital payment or a customer call - the mundane, expensive places where a large company discovers whether technology can do more than perform onstage.
EXL has been standing in those places since 1999. Founded by Vikram Talwar and Rohit Kapoor, the company first made its name in business process services: taking on repetitive, rules-heavy work for corporations and running it from a growing international delivery network. The first Indian center opened in 2000. A UK office followed in 2004. By 2006, EXL had bought analytics firm Inductis and listed on Nasdaq. The company was already learning that merely doing a process more cheaply was less interesting than understanding why the process behaved the way it did.
That lesson now sits at the center of a 68,000-person company. EXL sells data engineering, analytics, AI platforms, digital operations and specialized software to more than 800 enterprise clients. Its customers include insurers, banks, healthcare payers and providers, retailers, media companies and energy suppliers. In 2025, revenue reached $2.09 billion. In the second quarter of 2026, revenue grew 15.6 percent from a year earlier to $594.8 million, and EXL added 17 clients.
The old business is the new advantage
The standard AI consultant arrives with models, engineers and a transformation plan. EXL arrives with something less glamorous and often more useful: memory. It knows how a claim moves from intake to adjudication, why an underwriting queue stalls, where a bank's customer data breaks apart and which healthcare payment exceptions need a human decision. That process knowledge gives its engineers a map of the work before they automate it.
Insurance shows the depth of the model. EXL says it serves more than 430 insurance clients, including nine of the 10 leading US insurers and three of the five leading UK insurers. It works across underwriting, policy administration, claims, customer experience, premium audit, finance and third-party administration. The company can provide services in all 50 states and Washington, DC as a licensed third-party administrator. In other words, it does not merely recommend a claims workflow. It can help build the system, operate the process and be measured against the result.
The model is rarely the whole problem. The stubborn part is the workflow around it.The operating thesis behind EXL's reinvention
The approach travels. In banking, EXL works on fraud, risk, lending, collections, payments and customer engagement. In healthcare, it applies analytics to payment integrity, care management and revenue optimization. Retailers use its forecasting and marketing work; energy companies use it for customer and operational analytics. These are environments where data is fragmented, decisions are auditable and small errors repeat at industrial scale.
For a client, the practical promise is less time spent reconciling systems and more decisions made with the same version of the facts. A bank can combine transaction signals for fraud and personalization instead of maintaining separate models. An insurer can extract evidence from incoming documents, route a routine claim automatically and send an ambiguous case to a specialist. A healthcare payer can find likely overpayments without treating every anomaly as proof. The useful output is not AI in isolation. It is a shorter queue, a more accurate decision and a record of how that decision happened.
A platform portfolio, not a magic button
EXL has wrapped that operating method in a growing set of platforms. EXLdata.ai, launched with Databricks in 2025, tackles the data layer: modernization, governance, lineage and unstructured information. By March 2026, EXL said the suite included more than 80 modular agents. EXLerate.ai sits above it as a development and orchestration environment, now including a no-code Agent Studio for building autonomous agents. EXLdecision.ai helps create analytical models and governed decisions. ClaimsAssist.ai applies the stack to an insurance workflow.
Claims, payments, underwriting, risk and customer operations.
Modernization, governance, lineage and AI-ready data.
Models, agents, controls and industry applications.
People and systems that keep the workflow moving.
There are older and narrower products too. XTRAKTO.AI extracts information from complicated documents. LifePRO and the Life Digital Suite support insurance policy administration, intake and underwriting. Digital Virtual Assist handles conversational customer service. The portfolio can feel less like one neat software suite than a well-stocked workshop, which reflects the company it came from: EXL tends to meet a specific operating problem, then assemble software, partner technology and people around it.
Its partners supply much of the underlying infrastructure. Databricks provides a lakehouse and governance ecosystem. NVIDIA contributes accelerated computing and, in 2026, a transaction foundation model workflow for fraud and risk. EXL also works across AWS, Microsoft and Google Cloud, and recently added formal relationships with OpenAI, Anthropic and Snowflake. The positioning is deliberately model-flexible. EXL wants to be the party that chooses, integrates and governs the components, not the party insisting that every problem has the same technical answer.
How the money arrives
EXL's business model is as hybrid as its product catalog. It bills for time and materials, transactions, fixed-price projects and long-term managed services. Software contributes license, maintenance, implementation and subscription revenue. Some healthcare payment-integrity contracts use contingent fees tied to identified and recovered overpayments. This mixture lets EXL earn from the transformation project and from the operation that follows.
The model also explains why the company still looks different from a pure software vendor. People are the largest cost. New long-term contracts require recruiting, training and infrastructure before margins improve. Large relationships matter: EXL's top 10 clients generated 34 percent of 2025 revenue. North America produced 82.7 percent. The opportunity is durable work and expanding accounts; the risk is concentration, wage pressure and customers deciding to build their own global capability centers.
Competition comes from every direction: Accenture and other consultancies, Genpact and operations specialists, Cognizant, Infosys, TCS, Wipro and Capgemini, niche AI companies and the client's own staff. EXL's claimed difference is not that it possesses a model no one else can obtain. It is the combination of industry concentration, an installed base of live operations, global delivery and enough engineering depth to put new technology into old processes without losing the controls.
The human layer gets larger
The August 2026 acquisition of iMerit sharpens that proposition. Valued at up to $310 million when announced, the deal adds model training, evaluation, reinforcement learning and a network of specialists that includes physicians, scientists, engineers and linguists. iMerit's Ango Hub coordinates expert work on complex multimodal data. Founder Radha Ramaswami Basu joined EXL's executive committee.
This is a telling acquisition in an industry fond of imagining people out of the picture. Better enterprise AI often requires more specific human judgment, not none. A radiology model needs expert review. A financial model needs context about false positives and compliance. A customer-service agent needs escalation paths. EXL is betting that human intelligence, organized carefully, becomes part of the product.
“The next generation of enterprise AI will be defined not by the models organizations choose, but by how effectively they can deploy them in real-world business environments.”Radha Ramaswami Basu, founder of iMerit and EXL executive vice president
Where EXL fits now
EXL occupies the increasingly busy space between management consulting, IT services, business process outsourcing and enterprise software. The company advises, builds and runs. That breadth can be hard to explain in a sentence, but it suits the actual shape of enterprise AI projects, which rarely respect vendor categories. Data must be cleaned. Systems must connect. Decisions need controls. Employees need training. Someone must own the workflow after launch.
Its culture has changed along with the pitch. EXL still emphasizes its five founding values - collaboration, innovation, excellence, integrity and respect - but the training machinery now points toward agentic AI, data engineering, cloud, solution architecture and industry certification. There are internal academies, university programs, hackathons and an Innovation Sandbox for prototypes. The aim is to make a huge delivery organization behave with some of the experimental rhythm of a product company.
The test is not whether EXL can produce another polished AI demonstration. It is whether the company can convert its old proximity to work into faster, safer deployment while protecting margins and avoiding the bloat that comes with a vast service portfolio. Its second-quarter growth and raised 2026 guidance suggest customers are buying the proposition. The harder proof will arrive one workflow at a time: a cleaner claim, a fraud alert with fewer false positives, a decision that can be explained, a process that costs less next year than it did this year.
That may sound modest beside the usual promises of artificial intelligence. It is also where the money is.