Dear Advisor: What should my AI/LLM Go-to-Market Strategy be for my (Pre)Seed Start-up?

(or) AI vs. Product-Led Growth

April 2024

Because I help funds with AI due diligence and my expertise in the field, I (understandably) get this question from start-ups a lot! I’m flattered.

AI FOMO is real because of the promises of AI and LLMs in demos, pitches, and the media. But here’s the thing: it’s typically not the lowest-hanging fruit for your early-stage start-up right now. I’ll share why and what you can try instead to get better results and extend your runway.

Here are 4 common misconceptions about AI/LLM GTM strategy for an MVP – with advice on what to try instead.

Please note: This is one opinion; please feel free to take/leave/adapt as needed.

Please note: I will follow the colloquial use of "AI” to also mean ML, LLM, and ChatGPT – and use those interchangeably here, even though they are actually not equivalent. I explain how they’re different here: AI AMA.

Misconception 1: In Absence of Customers, AI Will Make Recommendations

As I cover in this AI AMA, one way to think about algorithms/AI is that they find patterns in the data and then flag them the next time. Hallucinations, then, find patterns when none exist.

At a (pre)seed start-up, we don’t have many/any customers yet. As a result, we don’t have (much) data to learn from to personalize the customer experience.

Why focus your time and budget (more on that in #2 below) on developing algorithms as part of your MVP when you won’t be able to validate them without customers?

Instead of focusing on AI/LLMs, I’d encourage you to identify one customer persona with which to go to market (GTM) and learn from – without AI, to start. Here’s how to find that persona and determine a strategy.

Step 1: Identify 1 Customer Persona

What is a very specific pain point you’re solving with your product?
How does that manifest for 1 very specific customer persona excited to buy your product today?
What claims are you making to solve this customer’s challenge end-to-end?
How do the claims translate into an ideal workflow, end-to-end, from a customer’s pain point to a solution that provides them value?

Step 2: Uncover GTM by Making Their Life (a Little) Easier

Each interaction with the workflow is customer feedback! If you track this engagement/data for each point in the workflow, evaluate the following:
- Where are the drop-offs happening for a specific customer persona? Why?
- Who’s upgrading and not? Why?
- Who is sticking around, e.g., retained? Why? What’s resonating?
- What is 1 way to reduce friction and/or improve their efficiency?
  - Bonus: Consider pivoting, exploring, or looking into this. This core customer group might help you identify your “secret sauce” and create a better niche/focus for the products to help you differentiate yourself from the competition. Consider scaling that!

Bonus: You just started on your product-led growth journey!

I dive deeper into these topics, strategies, and solutions in the following blog posts:

Talk: Power of Data for Product Market Fit to help you identify inefficiencies in your business
Talk: Reverse Pitch: Interactive Pitch Advice to Tell Your Product, Tech, and AI. Story, to help you uncover your customer persona – and put you in the investor seat to tell a better pitch deck story
Dear Advisor: How can Founders Build Customer Trust around AI?
Dear Advisor: What should (not) be your AI roadmap? (or) Why You Don't Need AI in Your SaaS MVP
Dear Advisor: How do I avoid the biggest data/AI/ML mistakes others make?
(Step 3) Dear Advisor: How do I do Product-Led Growth (PLG) as my Go-to-Market (GTM) strategy for Product market fit (PMF)?

Misconception 2: Your Moat is Your Start-up’s AI Capabilities

I’d argue that your moat is how you solve your very specific customers' problems because this is the sole reason your customers will pay for your product; they won’t care how you solve their problem, just that you solve it for them!

Personalization and efficiency (via "AI” algorithms, as discussed in #1 above) will come from learnings from customer engagement (and the resulting data) to inform product development and solutions to customers’ pain points (as you get more customers). It’s just a tool to help you understand and retain customers!

How does “AI” add value to your customers? Bonus: This will help you pass the "AI Litmus Test.”

You also don’t have the runway (years or budget!) to develop novel IP – and get a PhD (or 5!) – for LLMs as part of the product roadmap!

Even if you did, as soon as you file a (provisional) patent, your invention may become public (and searchable); others may decide to improve upon it. Please consult a patent attorney to understand the nuances of your case.

Instead of focusing on AI/LLMs as part of your MVP, I’d encourage you to:

Buy anything that’s not core to the product – and build anything that is if there are no existing solutions. This way, you don’t have to maintain it!

Please consider improving your customers' efficiency (and see #1 above), especially if you work in Health or Medical Technology, as I advise in this blog post.

This way, the scope of the MVP becomes the shortest path to getting customer feedback through product engagement (as discussed in #1 above), helping you iterate on the product to find your niche faster!

Should you decide that AI/LLMs are still the way to go – and you need to develop novel algorithms – consider partnering with universities! (I can introduce you to the UCLA MASDS program, where I’m on the advisory board, to have current MS, MASDS, and/or PhD students in Statistics support you in learning about your customers.)

Bonus: This may help you qualify for an STTR rather than an SBIR grant!

This way, you prioritize extending the runway and uncovering the “secret sauce” (as discussed in #1 above) to help you differentiate from the competition! This strategy will help you scale – and outline the market opportunity, not just size, for investors.

Misconception 3: AI Will Do It For Me, Team Optional

Founders often ask me: “Do I even need to hire anyone? Can’t AI just do it for me?”

As with everything, the answer depends… It depends on many factors, including what you’re trying to do, the risk and cost of ChatGPT getting the answer wrong (especially if you’re in a regulated industry such as Fintech and MedTech), and how well ChatGPT solves this problem for you now (more on this in #4 below).

If your solution is a wrapper for ChatGPT, will you be able to pass due diligence? How does the tool add value to your customers – and tie into business ROI?

I was impressed and horrified to meet a non-technical founder (and a company of size 1) last year who had ChatGPT code up their MVP – to review real estate legal documents! All it takes is one software issue – or lawsuit – to bring down the house of cards. Will ChatGPT maintain the technology and adhere to Cybersecurity and other best practices?

If the technology itself is core to the value you provide the customers (more on that in #2 above), and you outsource it (e.g., buy it), you can’t afford not to have SLAs (service-level agreements) in place with the vendors! Understandably, ChatGPT won’t be on-call to fix issues for your customers when an outage or another software issue happens, though how I wish that were the case!

That’s not to say that ChatGPT won’t help you – or that your hiring needs won’t change with and without this technology; they might! It depends on your team’s business, strategic, and technical expertise – and how that comes together to help your customers live better lives.

If ChatGPT is core to the business, you’ll still need a team; more on that in #4 below.

Having said that, here’s advice on How … Founders [can] Build Customer Trust around AI.

Misconception 4: AI Will Just Work

If all the context your LLM model needs to support your diverse customers and their needs is available on Reddit, Twitter, and Project Guttenberg (e.g., what ChatGPT knows about) – that’s great, right?

If that’s the case, can you pass due diligence? (Please see #3 above for why that’s likely not the case.)
Are you sure that the algorithm is not biased against (a subset of) your customers?

If the model needs more context, you’ll need a team to customize, maintain, support, and iterate. That’s expensive! There are no guarantees that AI will work with (especially without much) data, and it depends highly on what you’re trying to do!

To avoid the biggest data/AI/ML mistakes others make, I recommend starting with if/then statements to create a baseline for evaluating Building Customer Trust around AI. It’s not sexy but it’s practical because it’s easier to implement, evaluate, and maintain.

Putting It All Together

Start with product-led growth, and check for ways to make your customers’ lives more efficient with the simplest algorithms.
- Please see #1 and 4 above for more details.
“Buy” over “build,” as needed.
- For more details, please see #2 above, evaluating predictions and recommendations on key business outcomes.