Floworks (YC W23) - your AI SDR 📈

Plus: Cofounder Sudipta on the challenges of building AI employees...

October 22, 2024

CV Deep Dive

Today, we’re talking with Sudipta Biswas, Co-Founder of Floworks.

Floworks is building AI-powered sales assistants to automate and optimize the sales process for companies. Starting with their flagship product, an AI SDR (Sales Development Representative) named Alicia, Floworks is transforming the way businesses handle lead generation, outreach, and meeting bookings. By leveraging advanced AI models, Alicia autonomously engages with prospects, personalizes outreach, handles objections, and even schedules meetings—without any human involvement. 

Floworks focuses on helping B2B companies streamline their sales processes, allowing teams to scale their outreach without needing to hire more staff. With its highly effective approach, Floworks has seen impressive results, including significantly higher reply and meeting booking rates. 

In this conversation, Sudipta shares the journey behind founding Floworks, the challenges they’ve faced in building AI employees, and what’s next on their roadmap as they expand their AI offerings for the sales stack.

Let’s dive in ⚡️

Read time: 8 mins

Our Chat with Sudipta 💬

Sudipta - welcome to Cerebral Valley! First off, give us a bit about your background and what led you to found Floworks? 

Hey there - my name is Sudipta, and I'm one of the co-founders at Flowworks. We started this company about two years ago. The idea behind it was to build an AI tool specifically to help people in sales. We chose sales because it’s the major revenue driver for most companies, but we hadn’t really seen a significant boost in productivity for sales teams. On a personal note, my co-founder and I also experienced the challenges of founder-led sales, and we thought an AI assistant, kind of like a sidekick, could really alleviate some of those challenges and make sales more efficient.

This is my second startup; I started my first company back in 2018 and did decently there. But throughout my career, I’ve mostly focused on building AI products. I’ve also published papers and hold patents in this field.

Give us a top level overview of Floworks - how would you describe the startup to those who are maybe less familiar with you? 

At Floworks, we're building a suite of AI employees designed to help businesses automate a significant portion of their sales processes from start to finish. We're currently focusing on our first product, AI SDR, Alicia. Alicia handles prospecting and the sales development role just like a top-performing human SDR would. She manages everything from lead generation to personalized outreach, engaging with prospects and eventually booking sales calls—all autonomously without any human involvement.

Talk to us about your users today - who’s finding the most value in what you’re building with Floworks? 

Our main users today are B2B companies, particularly those selling software products and services and who already have a solid inside sales engine in place—complete with inside salespeople, account executives, and SDRs handling outreach on behalf of the company. These companies tend to benefit the most from our product. Our sweet spot is companies that are Series A to Series D funded, but we're also seeing adoption in the enterprise segment. 

One of the reasons for this is that we use our own model, ensuring data privacy with no leakage to third-party providers like OpenAI. However, startups that haven't yet established their sales processes may not find as much value in our product.

How do you measure the impact that Floworks is having on your key customers? Any use-cases that you’d like to share?  

Let's say your company just raised a solid series round and now you're looking to scale your sales process. One way to do that is by increasing your outreach. We have companies in that exact phase who are using our product, like Unscript, to achieve this. So, how does it work with Flows? You go to our console and provide some basic inputs to our AI employee, Alicia. The inputs could be a description of your product, your ideal customer profile (ICP), and maybe who the decision-makers or champions might be. Our AI can also assist by suggesting these based on your product and company description.

Once you input that information, Alicia takes over. It searches our database for relevant, high-intent leads and then crafts personalized messages for them. The personalization comes from multiple sources like LinkedIn or the web. Alicia looks for key triggers, such as recent funding rounds or interesting posts by your target. It then integrates this data into a value-driven message that feels natural and relevant to the lead. It’s not just generic outreach—it’s tailored based on what’s happening with the person or company you’re reaching out to.

Absolutely. So, Alicia doesn’t just mention that someone posted on LinkedIn or that a company was in the news—it actually understands what that post or article was about and seamlessly weaves that information into the email. The value proposition of the product is tailored in a way that feels relevant to the recipient, making the email both engaging and meaningful.

Once Alicia sends the email—through our platform, not external email providers—we’ve seen very high open rates, around 25-30%, and reply rates between 15-20%. This happens because the content is highly personalized and not just clickbait. Even better, our meeting booking rate from this outreach is currently between 6-10%.

So, what happens after the initial email is sent? If the prospect responds with questions or objections, Alicia can autonomously reply using information from your sales documentation, website, or other sources. If Alicia can’t find the answer, it will alert the human SDR to jump in. In most cases, these objections are common and can be handled by the AI. The real kicker is that Alicia replies almost instantly, within seconds or minutes. This rapid response helps keep the conversation going when the prospect’s interest is at its peak, which is crucial for maintaining engagement.

Additionally, when it’s time to schedule a meeting, Alicia doesn’t just send a calendar link—it’s integrated with your calendar and coordinates directly with the prospect to find a mutually available time, just like a human would. All of this happens while you're taking a good night's sleep, from lead generation to outreach, and finally, to booking a meeting—all without human supervision.

To give you an example, one of our customers sent out 86 outreaches in a campaign and ended up booking 12 meetings—pretty amazing results!

There’s been an explosion of interest in agentic workflows. How has that shaped the way you’re thinking about building Floworks? 

Absolutely. When we started building this product over a year and a half ago, we initially thought it might be as simple as creating a wrapper around a good LLM. But after just a month into it, we realized it wasn't going to work that way. What we learned is that if you're building an AI employee, it should at least meet the same standards as your best human employee. And ideally, since it's an AI, it should be able to scale those best practices to 10x or 15x to really add value.

When we broke it down, we found that there are three critical components to building an effective AI employee. First, there's the planning phase. Just like a human SDR, the AI needs to have a plan to reach its objectives—whether that's booking 30 meetings in a month or reaching out to a certain number of leads. It needs to understand how to organize its actions around those goals, such as how many emails to send, how many leads to engage with, etc.

Second is the action phase, which involves executing the plan. This is where the AI needs to both analyze data and generate actions. For example, it may need to craft personalized emails or assign tasks. It's not just about understanding information but acting on it—sending emails, interacting with APIs, or updating a CRM.

The final component is evolution. A good human employee learns from their experiences and improves over time. Our AI SDR needs to do the same. If a campaign doesn’t perform well, it should learn from those results and adjust its strategy for the next campaign.

These three components—planning, execution, and evolution—are essential to building a powerful AI employee. We've spent a lot of time developing these modules, particularly the action module. We've even published white papers on how our action module is more reliable and accurate than similar features in current LLMs, like tool calling or function calling. This is especially important when the AI needs to interact with tools like Gmail, CRMs, or task management systems.

One of the key challenges we've addressed is making the AI's actions deterministic. Unlike ChatGPT, which might give different responses to the same prompt, our AI employee needs to perform specific, repeatable tasks. For example, if it's updating a CRM, it can't make errors like updating the wrong field or entering the wrong value. We've built a model, Thor V2, that ensures our AI’s actions are far more reliable and accurate than other models like GPT-4, particularly when it comes to interacting with external tools through API calls.

How do you plan on Floworks progressing over the next 6-12 months? Anything specific on your roadmap that new or existing customers should be excited for? 

Right now, we’re focused on building AI employees for the entire sales or GTM stack—specifically within the sales function, not marketing. We're starting with the AI SDR, but eventually, we’ll expand into roles like AI Account Executives and AI Revenue Operations Managers or AI DevOps. This progression feels natural because where the AI SDR leaves off is where the Account Executive typically steps in, and similarly, DevOps comes into play either in parallel or further down the line. We envision a seamless data flow between all these AI tools working together within a company.

Interestingly, as we were building the AI SDR, we discovered that the foundational blocks we developed could be used beyond sales. These building blocks are so universal that they can be applied to create any type of AI employee, not just sales-focused ones. Looking ahead, we plan to democratize this platform so that others—whether they’re companies building vertical AI agents or customers developing specific AI workflows—can use our platform to create their own AI employees.

This could also include internal teams building AI employees for different use cases. We’re aiming to have this democratized AI employee-building platform ready within the next six to eight months, allowing people to easily build and deploy AI employees for a variety of functions.

Lastly, tell us a little bit about the team and culture at Floworks. How big is the company now, and what do you look for in prospective team members that are joining?

We just hit 20 people last week. The culture at Floworks revolves around working at the cutting edge of LLMs and AI agents, and we encourage collaboration between AI researchers and software engineers. We maintain a strong bias for action. While we conduct fundamental research, we're not a research company—we're a product-focused company at heart. 

Our ultimate goal is to build products that solve customer problems. So, while we continue to publish our model contributions, our focus is on shipping features that delight our customers. Everyone on the team, from research to engineering and design, is encouraged to push forward and deliver valuable products and features on a continuous basis. It’s still early days for us, but this approach has worked well so far.

Conclusion

To stay up to date on the latest with Floworks, learn more about them here.

Read our past few Deep Dives below:

If you would like us to ‘Deep Dive’ a founder, team or product launch, please reply to this email ([email protected]) or DM us on Twitter or LinkedIn.