On the face of it, Ema or Enterprise Machine Assistant, which came out of stealth mode earlier this week, looked like just another run-of-the-mill generative AI startup raising funds in the rather hot US market.
What it does seems basic too —using generative AI to enhance customer support, and automate mundane tasks engineers and data scientists perform at tech companies etc.
But one thing made everyone sit up is that in less than a year, the startup had managed to land nearly a dozen enterprise customers without a prototype or demo—it had the actual assistant actively running as an AI employee at these clients, automating tasks. Plus, it has an interesting revenue model.
Another point that drew attention is that the startup has raised $25 million across its seed and Series A rounds from a clutch of top Silicon Valley founders and leaders. Those names include former Meta COO Sheryl Sandberg, Yahoo co-founder Jerry Yang, Facebook and Asana co-founder Dustin Moskovitz, Snowflake CEO Sridhar Ramaswamy, and Ribbit Capital founder Micky Malka.
Among VC firms, these rounds were led by Accel, Section 32 and Prosus Ventures, with participation from Wipro Ventures, Venture Highway, AME Cloud Ventures, Frontier Ventures, Maum Group and Firebolt Ventures.
“Our seed round was closed within three days,” said Ema’s founder and CEO, Surojit Chatterjee, a Silicon Valley veteran with over 25 years of experience, with product leadership roles at Google, Coinbase and Flipkart.
Asked how he managed to bring together such an illustrious captable, at a very early stage, he said, “I knew a few people… They're all people I have worked with for many, many years. So, it's a network that's very strong. I'm blessed that they believed in me, the company, and the team.”
“Also, our vision helped get us the folks we have on our captable. We have been oversubscribed in every funding situation,” he told Moneycontrol over a Zoom call from his California office.
The fresh capital will be used for research and development of technology, on the go-to-market strategy, and to expand the team. Currently, the startup has a team of 30 people across India and the US. Chatterjee is looking to hire in both geographies.
How Ema works
Chatterjee claimed that Ema’s AI model can in fact, “match or even exceed human performance”.
“Our customers actually appointed an expert to evaluate Ema’s AI employee versus their own employee, side by side. In most of these cases, we saw that just within a few weeks, Ema learned very quickly and became as accurate, matching or exceeding human performance,” Chatterjee said, noting his ‘AI employee’ will be different from Microsoft’s Co-pilot, Google’s Gemini, ChatGPT and others.
“The distinguishing factor here is that Ema has been built to be an AI employee. So, it's not just transactional. Ema can take feedback and incorporate that feedback very quickly and retrain,” he added.
Chatterjee explained that Ema can morph into different roles quickly—from being a customer service assistant to a data scientist or a sales assistant. The AI assistant can work on specialised tasks too.
For instance, as a pharmacist assistant it can accelerate prior authorisation for medical procedures, or drug administration. This kind of a solution helps in developed countries, where healthcare is expensive and requires prior authorisation for everything, with scores of pharmacists and physicians having to do everything, leaving patients waiting for days or weeks to get approval, he said.
The startup's customers include immigration service provider Envoy Global, financial services infrastructure startup TrueLayer and fintech app Moneyview.
Ema also addresses the accuracy concerns and computational costs involved in current generative AI applications by using its “fusion of experts” method, combining existing large language models and running them through 2 trillion parameters. These LLMs include all the popular names, including Claude, Gemini, Mistral, Llama2, GPT4, GPT3.5, and Ema's own custom models.
Unique revenue model
Ema does not follow the usual SaaS subscription-based model for revenue, as most of its rival generative AI startups do. Instead, it charges customers based on tasks they deploy Ema for being completed.
While some tasks are simple, such as responding to a customer query, Ema also takes on more complex work, like producing a request for proposal (RFP) or a project plan of 50-100 pages.
“We think the current SaaS model is very outdated. Customers end up not using even half of the services they pay for,” Chatterjee said.
He added, “We are dramatically lower in price compared to some of the other folks out there… Our customers will have 10 to 20x RoI or even more.”
The back story
Chatterjee launched Ema after a year-long hiatus following his last stint as the CPO of crypt exchange Coinbase, which he also guided to its 2021 listing. He quit Coinbase after three years, citing a personal crisis—his father had been diagnosed with Alzheimer’s disease, and his mother had passed away suddenly.
At the time of Coinbase’s listing in 2021, Chatterjee’s stake in the company was worth $180.8 million. He was also set to receive share options within the next five years that were then worth about $465.5 million, Bloomberg had reported.
After quitting, he took a break and travelled through India before the idea of Ema hit. “I'm a builder at heart. I had this itch to go back and build something again, from scratch,” he said.
Chatterjee soon started to sift through research papers and meeting researchers and PhD scholars in this space. With nearly 40 patents in his name, he, too, had co-published research papers on AI with his Google Search colleagues.
As a former Google executive, Chatterjee had set up a team and built Google’s mobile advertising business from scratch back in 2007-2008. That division is today a $100 billion-plus business, he said. During a second stint in 2017-2020, he was vice president for product management at Google Shopping.
“There were no mobile ads before I started the team…I grew it from zero to $50 billion. That's the kind of experience I've had in taking on projects,” he said.
He found the inspiration to build Ema from his previous stint at these technology companies, where he would see that a big part of what some of the smartest employees did was tedious and monotonous.
Chatterjee approached his now technical co-founder Souvik Sen, another former Google executive, to build Ema. During his stint at the Big Tech company, Sen had built some of its largest data and machine learning systems and has 37 patents to his name till date. He had also worked on privacy and safety at Google.
The startup’s strategy and head of operations, Swati Trehan, is also a former colleague of Chatterjee from Google, where she was the chief of staff for Google Shopping. Trehan has experience leading large teams and managing operations strategy for companies such as e-commerce software provider Shopify. She joined Ema last August.
Discover the latest Business News, Sensex, and Nifty updates. Obtain Personal Finance insights, tax queries, and expert opinions on Moneycontrol or download the Moneycontrol App to stay updated!
