In an era where artificial intelligence is starting to disrupt many industries and markets, not every company saying they are using machine learning and AI are built the same. Companies capable of genuinely understanding and using these new tools in ways that deeply support their products or services will begin to differentiate themselves from those that lack appropriate understanding and are simply wrapping the term ‘AI’ around more traditional tools. A big part of this emerging differentiator is just defining and understanding the problem space in order to apply these advanced tools in meaningful ways.
These are the kinds of issues that Zayd Ali, the 21 year old founder of Valley, spends a lot of time thinking about. Valley is a rapidly growing startup in the sales development industry aimed at automating the $62 billion-a-year appointment setting market in a way simply not possible even a year ago. The company is developing and using proprietary generative AI models and algorithms to reinvent business-to-business sales interactions.
This isn’t Ali’s first rodeo; despite his young age, he has previously founded, grown, and exited two other companies in the sales industry. Most recently, he exited a company he started called Advisor Appointments via a private equity sale. Ali is using what he learned there as a springboard for his current venture.
How Valley Is Leveraging Generative AI
Zayd Ali’s path to innovation in sales technology began as a non-technical founder, a title that often comes with its own set of trials, particularly in the tech-dominated landscape of Silicon Valley. However, Ali’s previous successes armed him with an arsenal of industry-specific knowledge. This expertise became the cornerstone of Valley’s strategy in disrupting the sales automation market. He was able to identify and carefully structure an industry-specific open problem that machine learning and AI on their own cannott address.
At its core, Valley leverages state-of-the-art generative AI and large language models to automate and streamline the tedious processes involved in identifying, developing, and setting business-to-business appointments. The idea was to build a system that could simulate sales development representatives across various industries, a technically highly challenging goal that required pushing the limits of existing AI technology.
Generative AI refers to algorithms that can learn from a dataset and generate new, similar data points. Depending on how it is trained and used, it can be an effective tool for machine inference — connecting the dots between data points in a meaningful way. Large language models, like ChatGPT, are a subset of generative AI, focused on understanding and generating human language and dialogue interactions based on enormous datasets.
In the context of Valley, this meant creating responses and interactions that a human would perform when setting appointments. A highly non-trivial task but one that, given appropriate industry and domain expertise to provide context, large language models might be particularly well suited to tackle.
The ‘Quality Over Quantity’ Approach for Building An Industry-Specific Generative AI Tool
Valley’s approach was to develop a proprietary model — the Valley Reinforced Learning from Sales Feedback model. Unlike the typical approach of “bigger is better” in data modeling, Valley bet on specificity and accuracy. The model was trained not on all possible sales interactions but on the most effective ones, honing its ability to mimic successful sales representatives in a way that continues to learn and improve through iterative feedback loops.
In fact, the Valley team is leveraging a very active area of research when it comes to large language models and generative AI. One of the biggest limitations of extremely large models such as those that underlie ChatGPT is the sheer size of the networks and billions of parameters that need to be trained in the model. It costs millions of dollars just to train the network, let alone the cost of maintaining it and running queries on it. Furthermore, the physical resources needed to train and run these huge models, things like electricity and cooling, challenge the continued scaling that such models will be able to achieve.
As such, much research both in academia and industry is focused on developing smaller models and networks that can achieve similar or better performance compared to larger more general models as the quality and specificity of the data increases. This is one case where ‘good enough’ really is good enough.
Valley’s approach towards achieving market and technological differentiation has been to focus on a strategy of compounding data — the idea that with each interaction, the system gets smarter, more efficient, and thus more valuable to the individual user. This creates a high switching cost for users, as the more they use Valley’s system, the better it gets at understanding and serving their specific needs, dissuading them from turning to competitors.
Taking Advantage of Industry Data and Experience
Developing such a system required overcoming significant hurdles, from algorithm design to data acquisition and processing.
The team, composed of individuals from Samsung AI, Columbia University, Salesforce, Yext, and more, started by building on top of the application layer of existing generative AI frameworks. This layer is responsible for customizing the broad capabilities of the model to specific tasks — in this case, appointment setting. The workflow creation involved setting up sequences of interactions that would most likely result in a successful appointment, much like a human sales representative would achieve through experience.
But contrary to other startups attempting to leverage generative AI, Valley didn’t rush to compile the largest dataset possible. Instead, they focused on curating a high-quality dataset with each data point correctly labeled and relevant to the appointment-setting task. That data specificity meant that their own model could engage in high-fidelity interactions with promising leads, tailored to the nuances and needs of different industries.
They achieved this in part through the use of reinforcement learning, a type of machine learning where an algorithm learns to make decisions by assessing the outcomes as it learns. Valley’s model is constantly updated with feedback from sales outcomes — a reinforcement learning strategy that ensures the AI’s responses are continually refined to be more effective in real-world sales scenarios.
Growing Success By Addressing the Specific Needs of Industry Challenges
The sales technology industry presents a range of challenges, from high customer acquisition costs to the need for scalable, predictable revenue. Valley is attempting to directly addresses these by significantly reducing the cost of acquiring new appointments and, by extension, the cost per conversion. This optimization offers businesses a newfound efficiency and scalability that wasn’t possible with traditional methods.
As Ali explained: “There are only three options for companies that wanted to set appointments with prospective buyers – and they were all sub-optimal. One is founder-led prospecting where if you have a really early stage company you can spend two to three hours a day trying to get in touch with the prospective buyers – a huge waste of time. The second option is hiring an appointment sending agency for $3000 to $4000 a month – very expensive and many of them rarely do their jobs well. And the final option is if you have the capital you could build a full sales development team which when scaled quickly becomes a 7 figure per year investment.
So the experiment with Valley was could we turn that $85,000 a year sales development representative salary or that $4,000 a month agency budget into a $400 a month software expense. Build a product that could take a cold stranger and turn them into a book sales meeting with zero human involvement, and by doing so dramatically change the customer acquisition calculus that our customers were performing. And dramatically alter the allocation of time towards prospecting versus other critical areas of an early stage company.”
Ali declined to share revenue numbers, but he stated that since their pilot program began in March of this year they have been growing 30% blended month-over-month and have reached seven figures in signed expansion letters of intent from existing customers.
Even as they continue to grow, some of their existing customers include Darwinian Ventures, Front.com, and Masterworks. Valley recently announced raising a $2 million pre-seed round from investors including Antler, Jason Calacanis, Rough Draft Ventures, O’Shaughnessy Ventures, ID8 Investments, Transform VC, and John Pleasants, the former CEO of Ticketmaster, Match, and Evite.