Unlocking GenAI in the Enterprise: Our Investment in Ema

Studying the history of enterprise software reveals a fascinating evolution of the relationship between computers and people at work. The earliest computers were essentially calculators, relegated to primarily research and defence use cases. Businesses soon started realising their potential as systems of records, concurring with the rise of enterprise software giants like SAP and Oracle. Personalisation of computing at work (Microsoft) followed. Then came Cloud-SaaS: a megatrend still underway. The compelling economics of SaaS inspired the birth of several new vendors, each trying to enable employees to fulfill one (or more) function(s) critical to enterprises faster and better - sales, HR, engineering, support, and so on. The relationship between people and computers at work became that of operators using tools. Today, these tools are omnipresent: the average enterprise uses 200+ SaaS applications.

Our investment in Ema comes at a time when the nature of human-computer interaction is staring at its next big change: from tools to assistants and eventually from assistants to actual AI employees. With the advent of GenAI, it’s increasingly apparent that users no longer need to hand-hold software. The software can now think and get jobs done, very much like human assistants. It has inspired visions of radically more efficient enterprises - environments where employees and AI agents work seamlessly together, constantly evolving while learning from each other. It is, therefore, no surprise that “GenAI transformation” is consistently coming out as a top-of-the-mind topic in CXO surveys.

Amidst all the tailwinds, however, it’s also becoming increasingly clear that while spinning up an AI demo is easier than ever, building production-grade AI is a tedious, expensive, and messy journey. That is where the Enterprise Machine Assistant (aka Ema) comes in. By offering a conversational operating system to launch specialised AI assistants for different internal functions, it abstracts out the underlying complexities of GenAI implementation and catapults clients directly to the business-ROI territory. 

We've known Surojit for a really long time: one of our partners overlapped with him at business school, and we've seen him in action at Flipkart since 2017. We stayed in touch with him,  and as soon as he discussed the idea for Ema with us in December '22, were ready to go on this exciting journey with him. He has seen enterprise-scale problems up close while leading product orgs at Google, Flipkart, and Coinbase. Surojit’s stellar execution experience, combined with a rare ability to see beyond the here and now, made us feel quite confident about Ema’s founder-market fit in the fast-evolving and competitive space of enterprise AI infrastructure.

At its core, the Ema platform is built with four important beliefs: 

(1) The data needed to solve a task at hand can reside in multiple silos in arbitrary formats, and needs to be synthesized correctly to be fed as context to LLMs.

(2) Security, Compliance, and Authorization (RBAC) need to be natively ingrained in the infrastructure vs be an afterthought.

(3) AI agents should continuously learn and get better with time, just like regular employees.

(4) There will always be multiple relevant models vs one large model to rule them all. The best model to use (on quality, cost, and latency measures) will depend on the task at hand.  

Taking an opinionated approach from day one has allowed them to invest in cutting-edge technology ahead of the market. For instance, Generative Workflow Engine^TM (GWE) is a patent-pending engine that dynamically maps out workflows with a simple conversation. While Ema offers standard personas for common enterprise roles like CX, data analyst, sales assistant, compliance assistant, etc., clients can instantly spin up custom personas for specialised tasks using plain English. Secondly, EmaFusion^TM is a model that intelligently combines many large language models, maximising accuracy at the most optimal cost. So the client has to never worry about keeping pace with the increasing number of publicly or privately available LLMs. 

Within a few months of starting out, Ema has begun working with over a dozen marquee enterprise clients - and is already showing significant, measurable business impact. It has also built an A-grade AI engineering team in the Bay Area and Bangalore, aligning quite well with Accel’s long history of backing exceptional teams in the India-US corridor. With the right ingredients in place, they’re in for an exciting journey in shaping enterprise software’s next big leap.

We are delighted to announce our investment and welcome Ema to the Accel family!