By Michael G. Kollo, CEO Evolved Reasoning
I had a surreal day last week, where in the same day I presented at two very different conferences about Generative AI use cases. Firstly, to a room full of corporates, the topic was novel, the conversation was around risk, governance, people and adoption. It was cautiously optimistic, though it lacked energy or specificity. That same day, I stood in front of a room full of AI startup builders, a room crackling with energy and enthusiasm that sought grounding frameworks to channel the raw energies of innovation and technical prowess.
In the style of ’48 laws of Power’, these are the ‘7 Laws of Gen AI startups’ for you, a builder of Generative AI. These Laws represent the principles and impressions that could help you frame your thinking, research, product roadmap and pitch.
1. Everyone wants AI, but no-one (really) cares about AI (as much as you do).
Look around you in the building that you’re in, and think about the building material. What is it made out of, what type of metal or brick, how long has it been prepared, what are the kinds of mortars. What are the critical temperatures, and the industrial process that brought it about. Try pitching the idea of improving that building material for three minutes to yourself, or to a friend. For fun, throw in imaginary technical building terms. Now try again, but focusing on the decorations and the colour scheme. Equally make it as technical as you want, but now see how much easier that is to relate to.
AI represents a capability or a discipline, but not a utility in itself. Founders that work with AI may derive benefit, but that’s purely personal, and not shared. Your audience will derive benefit the more you talk about them, and their problem, not about you, and your solution.
A frequent mistake that technical founders make is that they want to describe their value in technical terms, often referencing capability not product. How interesting the data is, how great the algorithm is, how powerful the whole stack is. Things that excite them. Mostly, these terms like ‘neural networks’ lose parts of your audience, than they get them involved in the conversation.
When you pitch and communicate, you have to be entirely focused on the person you are speaking to, not yourself. For any conversation, the human brain has the ability to incorporate some amount of new information, but for the most part, wants to hear and observe things that it already knows. Even the most trained VC investor will blank out after a while, unless they hear specific terms that they are familiar with, or they think they should.
You have probably 10% of your words and content that can be novel, so choose well. The rest of the 90% must be things that are familiar language to your audience.
When you practise, play ‘AI word bingo’. Each time you say a technical term, cross one off. The more you cross-off, the worse your performance is.
Be pragmatic about the tool of AI fitting the problem. Take time to describe the problem in a credible way. For example, “Emergency switch-boards receive 2.5 million calls a day in the United States, of which 22% are not emergencies at all. This means that 400,000 calls should be screened out before they reach a human operator. This is a known problem, and so far has been unmet because the technology hasn’t been responsive within a window of five seconds, which is often too late for emergency calls. However, our AI is able to respond to phone-calls within 0.5 seconds, which solves this problem effectively. We are in talks with several emergency service providers, and … “ Out of the 90 words or so, I spent 17, or just about 20% talking about the AI solution.
The counter example: people still want magic. Many organisations have built their initial sales through the promise of very sophisticated systems, using highly technical language to essentially overwhelm their investors. This can get you off the ground quickly, but is unlikely to last, and requires a certain kind of ‘bluster’ from the CEO especially.
2. Everyone wants the idea of Generative AI, without the (imagined and real) risks
The public narrative of AI is terrible. It involves inhuman machines that want to take over, destroy us, take our jobs, and so on. Probabilistic Generative AI tools have inherent risks that can only be managed and not eliminated. Buyers of AI and data technologies will not be used to conversations around probabilistic outcomes, and will often confuse these with the idea that you haven’t done a good job, or you’ve missed something. This will grow their uncertainty, and the bar for risk will be raised.
You need to ensure that you have either addressed these risks, but also that you’re open to discussing them. The scale and power of Generative AI is immense, and its risks are comparable to the human operator. Picking up the example below: “In our testing, our systems picked up and correctly classified calls to emergency departments in 92% of cases. This compares with the human equivalent of 85%, so we see the system making mistakes, but less frequently, and importantly, much quicker. An average human operator will take 32 seconds to incorrectly evaluate a call, whereas we will take less than 12 seconds after a brief conversation. We improve on both accuracy and speed.”
Be open about questions about ethics, bias, data sourcing and so on. In many cases, it wont be relevant, but when dealing with personal data, or any kind of profiling, take a page to reassure your listener that you’re a responsible citizen, rather than a crazed AI builder.
The counter example: Your business may be about addressing these risks. In this case, highlight them carefully, and ensure that you dont dissuade the company from using AI, but that your solution of managing/mitigating/monitoring these risks sits along-side the solution. Risk is your friend, the language of risk aversion and governance is what you should seek, and ensure you embed yourself with any conversations around regulation.
3. Companies will pay for products that fix problems, not capability
We live in an unusual time, where a raw capability like ‘AI’ is being advertised and considered as a ‘hot topic’. In its raw form, Generative AI is still a capability, and far away from a product that actually fixes or restructures a problem. Remember that as founders with domain expertise, if you are going to build a business on a better form of AI, without a specific use case, you will be selling to other (bigger) technology and AI companies. As soon as you turn your capability into a product, you will spend much time talking about that problem and the customer, than you will about AI. On the other hand, if you stay as a capability provider, you will likely have to narrow your sales efforts to actual technology companies that are trying to solve a use case.
Equally, talking too little about your technical capability will pitch you as a wishy-washy consultant type, with broad brushstroke solutions, and little in the way of actual IP. Walk this path carefully, giving just enough to communicate your significant technical prowess but without being off putting.
The Counter example: In some cases, a company will use your capability as marketing such as “We are using AI to..”. These are the rare cases where you have value as a capability provider. You will be invited for corporate presentations and investor meetings as the ‘smart company we are using’. Use these opportunities to embed yourself closer with the client.
4. Finding a way in: the AI enthusiast
All organisations will have mavericks and first movers. These are individuals who have a keen interest in innovation, technology, and often feel constrained somewhat within their corporate roles. Their interest is to connect with start-ups, and technology vendors and to encourage innovation. They will often bring you into the business to engage, and introduce you around.
Many of these folks will not have been where you are, and they wont exactly know what it means to spend your time socialising within a corporate without payment. They have not felt the stresses associated with short runways and juggling of priorities. Don’t be shy to ask them early on about their commercial scope: do they have budget for Proof of Concept trials, do they have deep connections within the business to test for use cases. Can they at least help you refine your product research. Your corporate network is vital, so develop all relationships. But don’t mistake enthusiasm for commercial prospect.
Some of the most mundane conversations have flowered into the most interesting commercial opportunities, whereas the most exciting meetings of minds have ended up with little more than the conversation. It is hard to tell. Stay pragmatic, stay focused. Enjoy a meeting of the minds, but don’t assign too much to it. Equally, seek out those conferences and coffee conversations with very different people from your walk of life, and learn to understand the true drivers of a sale.
The counter example: There are moments in the corporate lifecycle where the investors and board focus on innovation as the key to growth. These are usually around times when there is a change of leadership at the top, or a critical strategy meeting occurs. Your enthusiastic contact will suddenly rise in fame and importance, and may even command a significant amount of budget. Help them use their moment in the sun to embark on ambitious projects that the company so far has put off.
5. Finding a way in: the internal tech resource
Often you will encounter kindred spirits within organisations who are themselves domain experts. They will often be flooded by requests from within the firm to take on projects for which they don’t have time or resources, but are personally interested in. Try helping them do their job better, share with them your product and insights, and help formulate projects that you can take away, and that they will pay for.
For the time being, most companies will throw ‘AI’ into the ‘data and enablement’ pile, so many of the corporate professionals looking after this domain will have themselves technology backgrounds. That is good in many ways, as it will help you connect with them. It is bad because they will themselves inherit the problem of translation and connecting with the business, and the challenge of finding (funded) use cases. Help them achieve this, and they will be grateful.
The counter example: the technical internal resource may or may not be in the right circles in the firm. Their own political prowess will see them in various committees, controlling the flow of information or projects, or will see them isolated from the business. Be wary that technical capability rarely comes with political acumen.
6. Getting the timing right or staying front of mind.
Most of the time when you approach a prospect with an idea or a product, they will not be ready to hear it. They will listen, and engage, and sometimes they will be interested, if for no other reason that they think they should be. But often they wont be ready, as many things need to align across a business from product strategy, to budget decisions, to other vendors and projects being a greater priority. Be patient, and flexible with bringing them AI. Sign them up for regular updates, stay around them. When they are ready, usually prompted by a new financial year, a strategy session, or a board meeting, they will reach out to you. Staying within their range of vision, without being intrusive or demanding is critical. Equally, don’t spend time developing leads that are not ready. If you have a short window, move them into a holding pattern.
Equally, do not anticipate that you can build a community of corporates as easily as you can with a tech/data community. Tech and data folk in the startup space are inherently curious, they want to connect and experiment. Corporate folk want to execute, they are constantly being sold to (badly) by a range of people, and they don’t inherently have the time to work out what is good and what is bad. An interesting cognitive bias of humans is repetitive engagement. What we see more often seems more right to us. Use this to continue to engage in small pieces, and stay patient.
Technology is cheap, and clients are hard. Many fintech companies learnt that the hard way. Building technology is under your control, it may even be something that you enjoy doing. Finding clients that are ready and able is out of your control, and sales cycles are heavily influenced by macro-economic environments, competitors, and just chance. Start finding clients from day 0, long before you have a product or a solution.
The Counter Example: Some startups will focus solely on capturing the public narrative. They will pivot their product and business around the idea of timing, so to ensure that they are in the top 3 topics being considered by companies. Typically more built around consultancy capabilities, these businesses are also really good at using media to get their first customer. The lesson: timing can be created, not just something that happens to you.
7. Be a ‘Swoosh’, not a ‘Tick’.
Broadly, there are two kinds of solutions to problems: a tick and a swoosh. A tick is a kind of solution that is just required, but once its there, it doesn’t really matter how good it is. Many policy documents for example sit in corporate drives, and they don’t need to be super well drafted or excellent, they just need to exist. Think of it as an average quality, almost transactional solution, that ticks a box. A good sign that something is a tick is that it doesn’t have a KPI attached, and more importantly, no-ones salary or job is on the line for its quality.
For ‘tick’ problems, organisations will typically turn to large suppliers and consultancies to meet these, as the upside is limited, and there are costs to evaluating more innovative AI products (like yours) that are really not worth it for these problems. Be cautious that even if you provide a clear solution that is a ‘tick’, you can and will be challenged by larger players who, using brand and reputation, will seek to compete with you. And often will. Technical IP will matter little here, and you wont be able to compete on brand and risk.
For example, take the common use case of creating a knowledge management product using Gen AI, such as a compliance document store, but unless there is a very direct KPI associated with the cost saving of the number of hours of people searching for document is measured, it will soon become a tick solution.
A ‘swoosh’ is where the quality of the solution is directly related to its outcome. An average solution will provide less, and a great outcome will directly provide more. Trading algorithms are examples where a poorly designed algorithm will on average more likely to lose money, and a better one with better data and technology, will make more. Similarly, AI solutions that restructure and organise information specifically linked to a product or to a cost can fall into this category.
These use cases are much more primarily to the business opportunity and conduct, and so may be somewhat harder to prove initially but they will provide much more visible value, and therefore, much greater likelihood of using a small, innovative AI company to solve. As much as possible, use quantitative measures of your solution, and benchmark these, so you can drive the conversation to a ‘swoosh’ solution, rather than a ‘tick’.
Mixing the two types of solutions is often a problem, and for many corporates, they may not be fully aware of which solutions drive which outcomes. Use counter-factuals to ask question: “if you used a non-AI solution here, would it still be usable? How much time are you expecting to save with this?”. Alternately, for many problems, there is a critical threshold that needs to be achieved with a solution, but after which the upside is limited. For example, if you can create a certain report that is 80% of the way there, that is enough for us. For some problems, that will be a huge jump from what there is today, but for others, it may be incremental. Try to push the conversation into ‘swoosh’ solutions where-ever possible as this is where you will shine, and the more pedestrian alternative ‘big company’ solution will falter.
The counter example: many ‘tick’ solutions fly under the radar. They are too specific and too small to be picked up by the bigger companies, and considered kind of ‘boring’ for other start-ups. Regtech is one such example. Seeking industry specialisation in an otherwise ‘tick’ solution can be a very profitable and well defined business as well, even when using Generative AI components.
