Leena AI: From HR chatbot to the agentic enterprise

The back office of every enterprise has a mountain of tickets needing human resolution buried in its IT and HR systems. This is the mundane, high-volume work that clogs queues and pulls skilled people away from other projects. The big software vendors saw this problem coming. ServiceNow built Virtual Agent, Workday built the Digital Assistant, and SAP built Copilot. Loads of engineering talent aimed squarely at trying to automate the enterprise back office. But the tools were too locked to a single vendor’s ecosystem to handle the messy, multi-application reality of how companies actually operate.

Into this gap stepped three IIT Delhi graduates with no corporate work experience, a chatbot startup behind them, and a small cluster of paying customers doing something unexpected with their software. CEO Adit Jain and his co-founders at Leena AI didn’t initially set out to build an enterprise agentic AI platform. They discovered the problem the way the best founders often do: by paying close attention to who their loyal customers were, and why.

What followed was a seven-year build through two distinct technological eras, a complete architectural teardown in 2024, and a company that now deflects 70% of enterprise support tickets, having grown 3x in the past year.

We’re sharing Leena AI’s story as part of our case study series, Launching AI products that win, where we sit down with CEOs and AI leaders about how they successfully commercialized AI. We learned from CEO Adit Jain how Leena survived two key pivots to arrive at true agentic AI for the enterprise back office. He shares their early customer traction that drove the team to rebuild for an overlooked use case, the pricing pivot they made, and how they maintained flat headcount while tripling revenue.

The situation	Founded in 2018 out of a pivot from a former company, Chatteron (a horizontal chatbot builder), Leena AI initially targeted HR ticket deflection using BERT/NER models. Pre-LLM, the founders were competing against well-resourced chatbot products from ServiceNow, Workday, and SAP, plus startups like Moveworks and Aisera.
The challenge	Early BERT/NER technology limited automation rates. Customers kept wanting to pull Leena beyond HR into IT, finance, and beyond. The 2024 transition to an agentic architecture required rebuilding the entire product from the ground up and migrating every customer. This was a painful but necessary reset that also involved replacing ~60% of the team.
The solution	Leena rebuilt around a modular agentic architecture, including an orchestrator, “AI colleagues” (purpose-built Level 3 agents), Agent Operating Protocol (AOP) for grounding agents in enterprise-specific SOPs, 200+ pre-built integrations, and a context graph that improves over time. The platform went from copilot to full autopilot, with human-in-the-loop gates available by process.
The result	The company has reached over $20M ARR, 3x growth in the last 12 months, with strong customer retention. Customers are now using the platform to automate recruitment, accounts payable, and soon payroll processing—all use cases Leena wasn’t originally designed for.

Key takeaways from Leena AI’s pivot from chatbot to enterprise automation

• Follow the customers paying you, not the vision you started with. Adit and his co-founders noticed a small cluster of Chatteron customers using them for internal HR/IT chatbots before they made their first pivot. That 15-person cohort became the early customers of Leena AI.
• Rebuilding is a competitive moat in itself. While Moveworks and Aisera got acquired, Leena stripped down its entire architecture in 2024 and rebuilt it from the ground up, including migrating every customer. Their competitors either couldn’t or wouldn’t do the same.
• The real edge is the integration layer. While it’s possible to assemble Leena’s functionality through Claude Code, ElevenLabs, and API connectors, most CIOs don’t have the number of engineers and years it would take to replicate. Leena’s unified architecture and speed to live production with positive ROI is the product.
• Pricing should follow value delivery, even if it creates revenue uncertainty. Leena moved from PEPY (per employee per year) to a platform fee and consumption model, and has experimented with outcome-based deals tied to self-service ratios. Variable revenue is still under 10% of ARR, but Adit is honest that the right model is still evolving.

How Leena AI stumbled onto an enterprise-level problem

Large enterprises run on an invisible infrastructure held up by an army of IT and HR specialists fielding the daily flood of employee requests. From password resets and benefits questions to onboarding tickets and PTO approvals, the average white-collar employee raises around 36 of these tickets a year across IT and HR alone. At $20 in hard costs per ticket—and up to $70 when you factor in the opportunity cost of keeping an employee blocked and unproductive—it adds up fast. 

The first wave of tools meant to solve this didn’t meet expectations. Every major platform vendor had essentially built chatbots that don’t fully work and, as Adit put it, are too “narrow, siloed, and locked into a single vendor’s ecosystem.” Employees touch several different applications, navigating approvals across departments and finding information scattered across systems, so a ticket-deflection tool that only knows one system can’t really automate anything meaningful. Competitors like Moveworks and Aisera were addressing the same problem, but the underlying technology was a constraint for everyone. In the pre-LLM era, automation ran on BERT and NER models that topped out at 35-45% ticket deflection on a good day. It was better than nothing, but not transformative enough to restructure how a back-office team operated.

Adit and his two co-founders, Chief Scientist Mayank Goyal and CTO Anand Prajapati—all IIT Delhi 2015 grads with no professional experience between them at the time—had built Chatteron, a horizontal chatbot platform that allowed companies to create bots for their websites, Facebook pages, and Twitter handles. It was technically sound but commercially unremarkable. After a couple of years, they saw they weren’t making much money and asked themselves: “Who is paying for us?”

The answer was a cluster of 15 customers who had subtly started using Chatteron outside its original use case: internal IT and HR chatbots. “We had zero idea what the hell these guys were doing,” Adit recalls. The founders had never worked in an enterprise before, so they didn’t know how much a helpdesk ticket cost or why a company with 10,000 employees had the same small team answering the same questions repeatedly. The team spent time with each of those customers learning that as companies grow, the sprawl of systems, departments, and applications creates a constant friction tax on every employee. This is what sparked their first pivot.

Two business pivots, one “painful” year

Phase I: BERT/NER era (2018–2023)

In late 2017, Adit, Mayank, and Anand ditched Chatteron and became Leena AI, going from horizontal to deliberately vertical, to enable enterprise HR. The team applied to Y Combinator, moved to the U.S. in the summer of 2018, and started over. They began in HR, and soon customers started pulling Leena in deeper and deeper, which helped realize their vision to expand across the enterprise. “Since we were just three people at the time, we started only in HR,” explains Adit. “But soon employees were saying, ‘Whether my problem is an HR problem or IT problem, or a finance problem, I want one place to go and get my work done.’”

This was the catalyst for their second pivot, which came six years later and was even harder. By 2024, it was obvious BERT and NER architectures had a ceiling, and LLMs had changed what was possible for agents, but taking full advantage meant rebuilding everything once again. “We fully stripped down our entire architecture in 2024,” shares Adit. 

Phase II: The architectural reset (2024–)

It was a “painful year,” according to Adit, but what came out the other side was an agentic platform that truly automated end-to-end business processes. When the team redesigned its architecture, they brought all of their customers on, introducing them to “AI Colleagues.” The first three AI colleagues were Harrison, a universal HR assistant, Iris, a universal IT assistant, and Fiona, a universal finance assistant—these became the primary, foundational agents that Leena and its customers built upon. Leena’s 70% ticket deflection rate, which is nearly double what the original technology could achieve, was the proof point.

“Turnaround across HR, IT, and finance has gone from two days to 6 hours, and it’s given us a reliable way to reach 100% of our employees digitally.”  — Makeen Khan, Chief Information Officer at Coca-Cola

Leena’s new architecture is organized around AI Colleagues, which are purpose-built Level 3 agents designed to do what individual contributors do in the back office today: interpret business processes, collaborate with people, and interact with various applications to get work done. The most important of these, and the one Leena always leads with, are the IT and HR operations colleagues built to deflect tickets. But the architecture was designed to optimize more. At its core is an orchestrator that evaluates incoming tasks, builds an execution plan, and routes to the appropriate AI Colleague. 

Leena isn’t a model company. The orchestrator calls on whichever frontier model performs best for a given task, including Claude and GPT, switching freely between models as the landscape evolves. The architecture is built as a set of modular services, meaning any component can be swapped out if a disruption occurs, so users can continue to use the technology.

What makes the AI Colleagues work in enterprise environments is four layers lying underneath:

1. The first is an Agent Operating Protocol (AOP), which grounds each agent in company-specific business processes. A customer uploads existing SOPs or writes the process in plain English, and the AI Colleague follows suit. 
2. The second layer is the Workbench, a forward-looking schedule of planned work and a task list for each agent.
3. The third is the skills library, consisting of over 200 pre-built integrations with enterprise applications like ServiceNow, SAP, Workday, and Salesforce that customers can activate in seconds. A skill is a single, deterministic unit of work inside one of those applications (promoting an employee, resetting a login, matching an invoice). For cases where APIs, MCP, or A2A protocols aren’t in place, Leena’s operating browser allows AI Colleagues access to legacy systems for the information, just like humans would.
4. Fourth is Leena’s context graph and memory layer that improves with each use and eventually even suggests refinements to the AOP itself.

Hallucination and error risk are handled with two LLMs running in parallel at every step: one makes the plan, and the other checks it. Top-tier models hallucinate at roughly 2.5-3% in isolation, but running two simultaneously knocks that rate down to approximately 0.09%. A third model then flags anomalies post-execution and surfaces them to a human manager, who isn’t a Leena employee but rather the customer’s own best practitioner of the business process. Leena even offers a training program for these managers to help them oversee and spot-check the AI Colleague’s work.

The most consequential outcome from Leena’s rebuild is what customers started doing with the platform once it could handle more than tickets. One customer used it to automate their end-to-end recruitment process; another automated accounts payable. And now, a payroll processing AI Colleague is on the horizon.

Leena’s pricing bet on following the outcome

Leena AI launched on PEPY: per employee per year. It’s a familiar SaaS construct that’s predictable and straightforwardly aligned with how enterprise software budgets work. More importantly, it shielded Leena from a risk that hadn’t existed yet: LLM cost exposure. In the BERT and NER era, there were no token costs to absorb, so PEPY was the right model for the technology they had. 

But then the architecture reset changed their pricing course. An agentic platform running multiple LLMs in parallel, such as orchestrators, AI Colleagues, and dual-model error checks, has real and variable compute costs underneath, and PEPY no longer fit. It also no longer reflected what Leena’s customers were actually buying. A company automating its entire recruitment process is consuming outcomes at scale, and the pricing should follow.

Today, Leena runs on a platform fee plus consumption. The platform fee covers access and includes enough tokens to onboard the first one or two AI Colleagues, which is enough for a customer to go live and start generating ROI. Beyond that, usage drives billing. Customers deploy more AI Colleagues as they automate more processes, so consumption scales with them. Leena also ran a handful of outcome-based deals that go further: if the self-service ratio (i.e., share of tickets resolved without human intervention) falls below a defined threshold, Leena pays back. If customers exceed it, Leena earns more. It’s a structure that bets directly on product performance, and Adit describes it as a natural expression of how the company thinks about delivering value.

The honest complication here is predictability, but variable revenue is only less than 10% of overall ARR, which provides enough of a baseline for the team to forecast reasonably well. Adit shares that the industry is mid-transition: “Pricing in our market is evolving. People like outcome-based pricing. People are wary of consumption-based pricing.” With OpenAI and Anthropic having done much of the customer education work by normalizing consumption models at scale, the team at Leena is now watching the landscape closely and treating the current model as a starting point. Leena is for pricing on the outcome, not the seat, even when it introduces uncertainty.

How Leena’s customers fueled a land-and-expand motion

Adit and his co-founders’ early instinct after the pivot from Chatteron was to stay narrow. They had already learned the cost of being horizontal. Starting Leena AI as HR only was a deliberate constraint for a three-person team with no playbook or prior enterprise experience, and it worked. Customers started using the product, the technology improved, and the ROI case of ticket deflection, cost per resolution, time saved, etc., was concrete enough to sell. 

The eventual expansion into IT and finance wasn’t a product roadmap decision. Leena’s own customers made the decision. That’s because enterprise employees don’t think about their problems in terms of department ownership. Once Leena proved it could handle one category, customers wanted it to handle all of them. This customer-led dynamic is now core to Leena’s land-and-expand motion. Customers essentially “design” new features based on what they learn after deploying their first AI Colleagues. 

Adit’s CIO pitch is simple and direct: 

“You could build everything Leena does yourself. Our architecture is open. You could assemble the orchestration layer, the LLM integrations, the voice interface, the compliance stack, the 200+ application connectors—piece by piece and vendor by vendor—as long as you have 20 engineers and a few years to spare.” 

Leena’s offer is 90 days to production. The question Adit poses to CIOs is whether that time and money spent DIY-ing is worth it when boards are seeking positive ROI, now. Vendor neutrality is the structural advantage that makes this argument land. ServiceNow, Salesforce, and Workday each have their own agentic layers. CIOs who have watched these platforms fail to interoperate with each other are most likely not going to build their automation strategy on top of any one of them. Leena sits above the stack, integrating with all of them, which is exactly the position enterprise buyers want a vendor to be in when they can’t afford to bet on one single ecosystem winning. 

"We chose Leena AI for its strong integrations with SAP and our broader tech stack. We wanted to avoid creating tickets that should never have existed in the first place, and Leena demonstrated that." — Hira Hashmi, Sr Director, People Operations, MongoDB

The 2024 reset included the go-to-market team as well. All three founders relocated to New York, and a full U.S. GTM function was hired. The team skewed heavily toward customer-facing roles doing sales, professional services, and customer success, while their engineering headcount held flat. Leena tripled its growth in 12 months, and the team is aiming for that same trajectory without adding headcount.

The next era for the AI enterprise: back office automation

Leena AI’s immediate roadmap is a straight line to where its customers are already pulling it. Payroll processing is next, which will capture exactly the kind of high-volume, multi-source, back-office work the platform was rebuilt to handle. The broader opportunity Adit sees is the enterprise back office as a category. Most of the automation conversation in AI has focused on customer-facing workflows like customer support, sales, and marketing. The internal functions that keep large organizations running have been slower to transform because they’re harder to reach, and the integration complexity is real. In Leena’s case, this complexity is the moat. 

For teams building enterprise automations, Leena reinforces that the model that matters isn’t necessarily the AI model. The integration layer, the enterprise trust stack, and the ability for businesses to build and orchestrate on top of that foundation are what determine tomorrow’s new systems of action. 

About Leena AI Leena AI is an enterprise agentic AI platform that automates back-office operations across IT, HR, finance, recruitment, and beyond. Founded in 2018 by CEO Adit Jain, Chief Scientist Mayank Goyal, and CTO Anand Prajapati, Leena AI is trusted by large global enterprises, including Coca-Cola, MongoDB, and McKinley. Leena’s purpose-built AI Colleagues are autonomous agents that deflect up to 70% of support tickets, automate end-to-end business processes, and integrate with 200+ enterprise applications out of the box. Learn more at Leena.ai