Is your data worthless?

In the hazy afterglow of CES - when inspired digital business leaders were dreaming up killer AI apps - we hosted a highly credentialed expert to dampen the buzz.

This expert has decades of academic and corporate experience in the cutting edge of AI and shared with our members a cold hard fact about this technology: killer apps require “trillions of observations” to get anything meaningful out of AI. If your business can’t access trillions of observations, does that mean your data is worthless?

This was not exactly good news. Some of our members run successful businesses that have amassed a ton of proprietary data that initially gave them the confidence to dream in AI. CES was the catalyst, offering hope and hype. Our virtual event was the cold water that woke them up. But there is more to this story.

This week, we spoke with a SaaS co-founder who pushed back pretty hard on our interpretation of our own event. As you will see, this pushback starts by addressing the question at hand but then goes deeper. It reminds us of a theme we keep hearing in our discourse: AI is not magic to be revered; it is a tool that needs to be managed.

Our ON_Discourse Overhead thread:

First off, your data is valuable, period.

All data has value. At least until the next paradigm shifting technology that could be so profound that it's impossible to conceive whether this will always be true. AI is not that for at least the next five or ten years.

Secondly, you don’t need trillions of data points. You just need to own it.

But owning the data is just the start. That data has to hold value for somebody who will want to buy it. That only happens if you are capable of dressing it up and turning it into a data product. Speaking of data product…

This reminds me of an interesting trend in data.

Data monetization is the new search monetization.

Companies with “small data” can monetize their data with AI platforms.

Smaller companies (basically 99.99999% of all companies) can augment LLMs through grounding.

Do you know about grounding?

Grounding is a concept where you focus on making sure that the information or data a company uses is closely connected to the real world and makes sense. Yext is an example of this; they power landing pages that are maintained and updated by distributed people within companies. Wendy's uses Yext to update their store hours. If you ask an LLM “is the Wendy's near me open?” That question references the standard corpus of search material and therefore might give you a wrong answer. So instead, Yext sells that real time data on behalf of Wendy's to the LLM. Now, when that prompt is entered, the dataset that Yext is providing gives a hard query return which ensures that the data is accurate.

Grounding can help your business.

When data is grounded, companies can make the most out of their small data sets, using them efficiently to drive decision-making without the need for massive amounts of information. The money flows from the query to LLM which is all monetized. … Note: at this moment the answer to the direct question was settled, but our guest was not satisfied. There was more subtext to address about the underlying question…

Is your question about data or risk? In other words, is AI going to drink your milkshake?

We are living through a moment of tremendous uncertainty. And in moments of tremendous uncertainty, good businesses are built in two ways, 1) taking huge risks and being able to withstand the downside, or 2) building businesses that are in support of risk taking.

Don’t forget that AI is ultimately the biggest risk to take.

Remember one thing: AI is tremendously expensive to run. The economics are a problem for anyone who thinks they can play in this space.

The scales tipped in favor of the hyper-scalers.

Let me put it this way: Sam Altman had like a billion dollars of venture capital. You cannot compete with this already and still it was not enough to develop anything remotely interesting. And so now Sam is going after Trillions of dollars. Unless you're in one of those places that has that level of resource, you're not going to be competitive.

So you’re wondering if hyper-scaler AI platforms will drink your milkshake? One thing we know for sure is that LLMs alone are not a moat.

There's a huge last-mover advantage in AI, because, you know, none of the models are defensible. In other words, you see what everybody else has done and then you just slide in there and do a slightly better version of it.

In AI, the only real moat is resources, ie: capital + compute.

OpenAI is forever dependent on Microsoft because of Microsoft’s resource moat. This is inarguable.

There are few reliable conventional moats for digital businesses in the AI era.

Very few. Hardware is no longer a moat, software is not a moat, and even if they could be, at best you have one year’s first mover advantage. That’s it.

But do not despair; there are accessible moats for successful businesses.

Good moats make it almost impossible for a customer to want to leave. AWS is a good moat because their security tooling and scale means I can never leave even if I wanted to. Similarly, ecosystem moats are good and highly defensible. Shopify offers an ecosystem that customers will not want to leave. Finally, strong brands with deep customer relationships are a moat that can endure strong technological challenges from competitors.