Navigating Generative AI: A Strategic Framework for Building AI Powered Products

Twelve months on from the launch of ChatGPT, Generative AI continues to be a hot topic among private and public companies. Recent technical advances and product breakthroughs have put wind in the sails of product builders and investors alike. This growing interest has meant that AI is now synonymous with the use of foundational models.

‍

As early-stage investors, we frequently engage with founders seeking innovative solutions using the transformative power of Generative AI. Drawing insights from these interactions, we've developed a decision-making framework for teams considering AI-powered solutions to address real-world problems.

• Identify a pressing pain point for a paying audience and design an AI-driven user journey that seamlessly resolves this issue. This is crucial to eventually realising value.
• Assemble a team with the ability to “hack”, particularly crucial in larger organisational settings.
• Leverage existing AI model APIs and frameworks to swiftly construct a functional prototype. Emphasising speed and adaptability over perfection is key, allowing for early validation and continuous iteration. This approach ensures a quick and responsive testing phase, facilitating efficient refinement and enhancement of the concept as needed.
• Establish clear benchmarks for a successful user experience and engage end-users in rigorous testing to identify corner cases and gaps to refine the UX. Proactively address potential challenges, such as hallucinations, latency, output inconsistency, and cost. Hallucinations in AI are misinterpretations that occur due to various factors, including overfitting, training data bias/inaccuracy and high model complexity.
• Understand the perceived value of the AI-powered solution for target customers and iterate on pricing to identify breakeven points and margin generation potential.
• Once the product's value proposition is validated, focus on optimising cost and performance metrics.

This approach is akin to Eric Reis’ Lean Startup methodology: reach a stage where you can confidently allocate resources to scale something that is starting to work.

‍

Choose the Right Problem Statement & Build a Compelling Product

Businesses must be designed to profitably solve meaningful problems for customers with willingness to pay. This expectation doesn’t change with AI in the mix. In fact, AI powered features can often be a distraction, leading to scope creep and features that are bolted on to check a box. Problems are not worth chasing in some cases, and the economics do not add up in other instances.

While language models have been getting supersized and highly capable under the radar over the last five years, the interface was one of the biggest unlocks that catapulted these models into the mainstream. The chat based interface was the first such interface, but we’re seeing more such opportunities now. From generative UIs (Perlexity, Defog) to design generation, webpage generation & code completion products (uizard, vercel, codeium), there are a number of these innovations. All contextual, sources of user delight, and able to displace a large share of knowledge work.

Companies have also chosen to work in underserved verticals and niches with copilot software and AI powered agents and have seen varying degrees of success. We’re excited about the opportunity for products to commit to the delivery of an “outcome” instead of one that is a “system of record” or a workflow tool. Ultimately intelligence & reasoning underpin the ability to get work done. Foundational model powered products provide exactly this capability, so the way to think about it is to automate mundane and laborious work that was previously done by hand.

‍

Focus on Time to Market & Shorten the Feedback Cycles

The “LLM wrapper” critique is commonly used to dismiss companies that are building AI powered products. In our view, this option is a huge advantage. Unlike previous waves of machine learning where models needed to be trained from scratch, and data availability & technical capabilities were huge moats, the baseline performance from off-the-shelf models has lowered the bar to ship a viable product.

It is also important to note that the AI model serving business is one of economies of scale. Short of running your own server farms, the best bang for your buck comes from reserving compute over one to three year span assuming that you can get high enough utilisation rates on these fixed costs. OpenAI does this better than anyone else, and while they may be discounting to land grab and make their competitors sweat, they are in pole position to invest heavily in reserved infrastructure, cut deals and even pass these savings on. HuggingFace Inference endpoints, Modal’s cloud functions, Google’s Vertex AI, etc. are all options for varying degrees of complexity & need. Self-hosted models, fine tuning, and inference optimisation will eventually have a role to play, but teams must start off by exploring and arriving upon a viable value proposition instead of prematurely optimising.

‍

Embrace the Absence of a Moat

One of the biggest questions associated with AI companies that are fundraising is “what is your moat?”. This is a topic that deserves a dedicated post, but here’s a short answer. Since these models, whether proprietary or open-source, offer very high baseline performance & flexibility out of the box, it doesn’t make sense to reinvent the wheel to begin with. Instead, it’s the other things that should matter way more at first:

• The chosen problem statement
• The effectiveness of the solution, the UX on offer
• The ability to quantify the value delivered and land a pricing model that creates a large market opportunity
• A willingness to gather & incorporate feedback to iterate quickly
• The ability to market, capture mindshare, and generally excel at GTM

Companies will eventually outgrow many parts of their stack and need to do more custom work, and a technology moat may eventually emerge but speed of execution is a company’s biggest strength in such a dynamic market.

The Generative AI ecosystem is evolving at an unprecedented pace, with groundbreaking announcements becoming the norm. As investors in this dynamic space, we are constantly challenging our assumptions and refining our understanding of how best to leverage AI and the potential it holds for transformative solutions. We look forward to learn from and collaborate with exceptional teams working on interesting problems.

‍

If you’re building with a focus on applied AI, we’d love to hear from you. Write to us - LINK