Navigating the AI startup landscape

New start-ups that use AI as the core of their offering are appearing at a rapid pace. These AI-first businesses are new to the business landscape. It’s no wonder, given the novelty of the technology and how its use can completely upend a market and create massive value without many of the same pre-existing constraints.

It’s now possible to make new services or products with almost no capital, just a handful of people, and a laptop. This has brought about the small-unicorn phenomenon (Currier, 2023), which will certainly increase.

Individuals have experienced this firsthand since November 2022 when OpenAI released ChatGPT3-5 to the public. Adoption was swift and entirely new fields were created overnight, and the words ‘prompt engineering’ entered our lexicon. Now it’s almost common to ask your favourite Generative Pre-trained Transformer (GPT) for advice, to summarize a large document, or to generate an image for an article in Think. Despite the huge leap forward in technology, it’s become commonplace and now there are more than a dozen publicly available GPTs with many more on the way.

As powerful as generative AI is, it is just a subset of the AI tools available to entrepreneurs. Because of its immediate applicability to the needs of consumers, it is well known now and ubiquitously used. Coupled with other AI techniques such as Vibecoding whereby a GPT creates a large percentage of the code for your product and a human assembles it (New York Times, 2025), entrepreneurs today have cheap, powerful tools to service existing customer needs in new ways that can undercut the existing cost structure or delivery mechanism of traditional solutions. While this affords great opportunity (and probably why there seem to be so many new AI-first start-ups, Ascendix, 2025), it also ushers in new challenges these entrepreneurs now face.

Because there is often no precedence for novel AI-first solutions in the marketplace, there are no pricing model to emulate, communication patterns to follow, or even business models to copy. Perhaps most difficult of all is that there is no budget for these new products, no business leader to champion the offering, or an obvious place to integrate the solution. Taking lessons here from the field building an AI start-up in 2021 from the ground up to now, we can learn several strategies to address these new challenges.

Let’s look at this real-world example to see how things can be different for an AI-first start-up. I co-founded Verilytix Inc in 2021 to address a multi-billion pound problem in indoor agriculture. That’s the amount of money lost per year due to poor plant yield mostly due to human error or negligence. Crops don’t reach their full potential or are lost altogether because humans didn’t see equipment failures, pest issues, or suboptimal growing conditions. But this is a tough job for humans since they are typically responsible for thousands of plants and growers have difficulty finding these (often minimum wage) employees. Employee churn is high, so building a suitably experienced workforce is challenging. Agriculture in general is an optimisation problem as the goal is to provide as much yield in a given space over a period time for the lowest cost to produce. Any plant loss is revenue loss.

The solution we created, Budscout, was a low-cost, robot chock full of environmental sensors that would move over the growing canopy of indoor plants recording conditions like temperature, humidity, plant height, CO2 concentrations, and light reflectance while taking multiple images. Each plant can be scanned every hour around the clock if needed. All this data is uploaded to the cloud where it is analysed and used as an input stream to AI systems ready to alert growers of poor conditions.

AI can detect plant stress due to a poor growing environment before a human can see the issue. Instead of asking employees to manually scan plants, growers can now send the few employees they do have to the areas within the grow that have issues brewing. And while all eyes are on the moving robot, the scout is simply a glorified data collection device sent out on jobs to collect fresh data to feed the analytics and underlying AI systems. Catching issues early can prevent catastrophic crop loss and improve yield by over 20% (Verilytix internal results).

AI capabilities must integrate into existing processes

While the concept seemed simple and logical enough, we were faced with challenges due to the novelty of the solution and the business environment we worked. First, we knew we must integrate into the existing process of growing plants indoors. There is already a set way this is done including the physical setup, human grower plant maintenance and even the plant lifecycle.

The reality is that we had crafted a moving device that would carry sensors and report those data to a cloud service. We needed to adapt our solution to the existing infrastructure of indoor grows and get the device in the right spot to detect the exact growing environment for each plant. We noticed for indoor grows, growers typically used a metal rack system to hold the plants and the lights above the plants. We attached our scout to the existing racks and took up space just below the lights. Even with robots and AI, growers are not particularly interested in high tech and are primarily driven and held to the yield of their crops as their key metric. We knew we needed to tie our solution to that.

You’ll notice in the above description of what we do, it is tightly bound to the concept of improving yield -not the underlying technology of AI or robotics. This drove our major design decisions going forward. Additionally, while we knew we could alert growers of issues to avoid catastrophe, we had to quantify this value and show growers what the potential benefit (such as avoiding a loss of revenue) could be. Because this was a new capability created by AI, we had to put the value of the AI in terms the customer cared about before they became interested in the AI.

Being open to unexpected benefits

Our plant scanning system takes images of every plant daily and sends these pictures up to the cloud. There, an AI algorithm scans the images and identifies and places bounding boxes around the important part of the plant that generates revenue for the growers such as the fruit or flower. This way we knew we could count and size up the entire crop to give the grower a precise sense of how well the crop is progressing day by day.

Upon showing this to the growers, they asked if we could display the increasing numbers and size changes over time. It turns out the growers wanted to be able to project out their yield at harvest in financial terms so they could use this information for business reasons such as crop insurance and taking out financial loans against future revenues.

Growers also told us they could mark the effectiveness of their staff that trim plants as part of the growing process. By analysing the images each day of all the plants, they can see when trimmers worked the plants and quantify their effectiveness. This allows growers to better deploy and scale labour to address identified plant problems instead of trying to find them and address them. We would have never guessed these use cases to enhance the benefits but collaborating closely with our initial customers and beta partners were able to align the AI better to business needs. Beyond the obvious capability of improving yield and operations for the growers, additional stakeholders took notice and found previously unseen benefits from our solution. For example, because we are scanning plants every hour of every day, inventory managers saw the potential for us to scan RFID tags on the plants so they could track thousands of plants as they move from seedling rooms to grow rooms onto harvest and final processing.

Operational leaders approached us to see if we could detect when a plant was no longer able to take energy from its light source or when it had enough water to alert management to turn off the lights and reduce water. These benefits were initially unforeseen by us until the customer pointed them out. Further, operators also liked telling their staff that every plant had a camera watching it to deter theft. Now we are working towards these goals. It’s because of the novelty of these AI capabilities that it took the customer to show us the added value it could provide.

Opportunities for industry disruption

Technology solutions for growers have generally involved significant initial capital expenditures requiring millions of dollars of upfront investment. These would be environmental and irrigation systems, buildings, lighting systems, and rack infrastructure to hold all this equipment. Because we created a low-cost robotic device, the value we were delivering to the grower wasn’t the robot per se but insights about their grow.

This was a completely new solution on the market, it’s no surprise that growers didn’t have an existing budget for our robotic scout system. We knew we couldn’t approach customers with a large upfront capital expenditure, so we decided to take a completely different approach to pricing our solution. Instead of charging growers up front for the robots and installation, we’d provide the robots and infrastructure and set them up for no cost. We simply charge customers a monthly fee based on the size of the crop under our watchful ‘eye’. We made these changes for several reasons.

1) Frictionless: Firstly, it meant a new customer wouldn’t have a cost barrier to entry to start using the solution. Clearly for novel solutions, one wants to remove all potential reasons a customer would reject your solution and stick with the status quo, and there was no budget to start with in this case.

2) Value: Secondly, we wanted to tie the costs of the solution as closely as possible to the benefits to the customer. In our case, the benefits are the insights. The robot is there simply to collect data. It doesn’t carry out any action based on the insights; that is still up to the humans to carry out. We decided to charge based on the value we bring which is all about providing insights over more of their crop.

3) Scalable: Lastly, we needed a scalable pricing model based on the benefits were delivering. And again, this was all about how many plants we were monitoring each day. The size of the canopy we are monitoring is directly related to our costs which include the number of robots we need as well as the infrastructure in the cloud to store and run AI models against the data those robots collect. Our pricing model needed to align both with the benefits we brought our customers but also our internal costs to deliver our solution. It was the novelty of this AI solution which allowed us to introduce an attractive, yet disruptive, pricing model into this established industry.

With careful planning and a thoughtful approach, your AI solution can completely disrupt an industry. Of course, starting with an industry that embraces change and technological advancements will ease your journey.

But first, it must align with existing processes and behaviours your customers already embrace. It needs to provide benefits to the areas and in the same metrics your buyers care about.

On the upside, you may have the opportunity to invent new pricing models, distribution methods, or introduce simplicity in the purchase process that will compel new customers to sign up. And while you may start with an intended benefit, find ways to engage with customers to learn about novel ways to add value that you may never have imagined. AI technologies can be so radically different from existing solutions, so remain flexible and open as you build and create value for your customers.

Toby Velte, PhD is an AI for Business contributor and the co-founder of Verilytix, Inc. He is also a lecturer and Sloan Fellow (2024) at the London Business School.