6.6.24: MIT STEX Demo Day - Maven AGU

SAMI SHALABI: Hi, everyone. Thanks MIT Exchange, and wonderful to meet you all. I'm Sami Shalabi, MIT alum and founder and CTO of Maven AGI. Maven AGI is building AI agents to reimagine customer experience, starting with support.

Prior to Maven AGI, I built three companies, exited two, and used to run all of product and engineering for Google News.

We believe that today's support experiences are broken. For customers, support is slow, painful, and getting worse, driven by budget pressures and increased product complexity. For many enterprises, support is human intensive, relies on fragmented systems, making it nearly impossible to deliver a consistent, great, personalized experience at scale.

Imagine if you could offer your customers a personal support assistant that knew everything about them, is a product expert, and is proactively working with them to get results they need, all at the fraction of the cost you are paying today.

At Maven, we've built a set of products that solve key support use cases, whether they're ticket deflection, which help your customers self-serve their issues; agent assistance, which helps your agents serve your customers better, faster; and AI insights that help your leadership to understand where the issues are and where the hotspots.

Here is an example of a travel site helping one of their hotel owners update their hotel web page. Using NLP and our actions engine, Maven is able to guide the user and update the URL, all from within a chat.

This is an example of the Maven co-pilot integrated into the travel company's support desk. Here, a Spanish question came in, and Maven was able to help the human agent quickly answer the question in English while responding in Spanish. The Maven co-pilot is natively integrated with every major support system out there, reducing training time and improving speed and quality of responses.

And finally, we've built AI-driven analytics that help support leaders understand the state of their support, automatically categorizing tickets, Maven's answer quality, and understanding overall customer sentiment, creating kind of a holistic view of the major issues that could exist in the support.

The way this works is Maven ingests enterprise knowledge-- APIs, CRMs, content in any form-- websites, PDFs, support tickets, data lakes, whatever, structured or unstructured, allowing it to answer questions, suggest follow-ups, and take actions.

Knowledge does not have to be clean. We know how to deal with dirty and incomplete data and have all sorts of flows that help identify gaps and help generate the missing knowledge. Our solution uses AI, natural-language processing, information retrieval, and large language models to help and has been proven to generate results in just hours. Our results are grounded in truth and data driven, making it possible to create a world-class personalized customer experience.

And we're at scale, and the results speak for themselves. We've processed over a million tickets in over 50 languages, and we've answered 93% of questions autonomously. The cost of serving tickets for our customers went down by 80%, decreasing the average ticket cost from $40 to $8. And the average resolution times have decreased by 60%.

And this is not just us. We were featured on OpenAI's website as a case study showing how AI agents can transform customer support at scale. In the case study, we talk about how we help companies like Tripadvisor autonomously handle over 90% of their queries, exceeding their initial expectations. So do check it out on OpenAI.

We're at scale now. We're working with companies across industries and have support for over 50 languages. We're excited to partner with any of you so you can help build a great customer support experience.

SPEAKER: Thanks so much, Sami. So let us know, how long does it take to train Maven to answer questions?

SAMI SHALABI: So we've built the technology in a way that you can actually get results within hours. But one of the things that we've been most proud of as part of the system we've built is we built an evaluation framework very early on, so it is a very, very data-driven decision. So initially, because we're able to ingest data in whatever form it's in, you can get results quickly. But then because of the evaluation framework, most of our customers are able to get to the 80%, 90% range within weeks, and in some cases, we've gone from first conversation to deployment in about six weeks.

SPEAKER: OK, and talk to us a little bit about security.

SAMI SHALABI: So when you work with large enterprise, one of the biggest challenges is, how do you protect the data? And when we kind of look at that, we started off by building an enterprise-grade solution. So we've done all the SOC 2s, the GDPRs, all the ISOs to make sure that the data is protected. Architecturally, we do the appropriate level of partitioning, redactions, et cetera to ensure that. And also the approach that we have taken is that we guarantee everyone that we will not use their data to train other agents. It is specific to them, and the solutions are siloed and only available to our customers.