Deep horizontals

AI adoption is facing the same problems that Geospatial has been struggling with. Services-driven products are one solution, MIT, A16Z, and OpenAI all agree.

Under a baking sun, I’m watching tumbleweed blow down the main street of my inbox.

Summer is a time for forced reflection. When seemingly everyone is taking the time for holidays and long weekends, the space that would be filled with calls instead is filled with consideration.

Photo by Jonas Allert on Unsplash

A focal point of my reflection is always the place of geography within society. With geospatial technology as the digital expression of geography. Whether in consumer applications, such as Google Maps or Strava, or in more industrial uses of geography for logistics or natural resource management. I care about how geography is used. Partly, this is due to the complex mysteries that geography constantly presents, but also because of geography’s ability to connect data and experiences through space and time. I’ve often referred to geospatial as a deep horizontal, because location is often a component of consumer and commercial workflows, but rarely the whole story. While the deep horizontal is an opportunity for geospatial technology to be everywhere, this is also one of geospatial’s greatest problems for broad adoption or even in defining geospatial or GIS as an industry.

Because it’s not, geospatial is a community of practice, precisely because it is everywhere.

Geospatial data and technology should not be siloed; it works much more effectively when built into workflows. However, the practices that have developed around GIS have become siloed, and this separation becomes a needless battle in many enterprises. I’ve talked ad nauseam about these subjects for years. There is nothing new there. My supposition is that in most business situations, decision-making is almost always made better with robust and accurate geographic technology. That geospatial technology must be embedded into workflows so profoundly that it is virtually invisible. This is also the nub of the geospatial product problem. So, while geography is rarely the whole solution to a problem, it’s almost always a component, a consideration. This is what I think of as a deep horizontal: an activity that could be a valuable component in numerous applications, but without workflow or domain context, may not be helpful in and of itself.

Interestingly, I now see the same patterns in the AI community, particularly in GenAI. AI is also a deep horizontal. It is rarely the whole solution; instead, it can help to speed up or optimize parts of a workflow.

This is borne out in one recent publication by MIT and one older article by a16z. MIT’s study suggests that 95% of GenAI pilots in enterprises are failing to produce any meaningful economic benefit. The reasons for this abject failure are multiple. There are common themes: brittle workflows, lack of contextual learning, and misalignment with day-to-day operations. Seemingly, the real benefits are for individuals using the common consumer-grade tools: ChatGPT, etc, to do things faster. These are not being used within the corporate environment but instead on personal accounts outside the control of corporate IT and any security measures. This is referred to as a shadow AI economy. So though the article is damming of GenAI pilots, it’s really more of a battle between corporate and consumer tools.

In the second article, a16z, back in 2020, said that the AI economy would need to be supported by services. Indeed, most robust AI implementations would need a human team of experts to create appropriate workflows. Why? because almost every robust enterprise use case would be a bit custom. For a16z to say services matter is quite a thing. Traditionally, mixing services into the SAAS revenue stream would be anathema, and frankly, frowned upon by most traditional Valley investors. Fixed cost people really screw up the software economics, and any sniff of a software company behaving like a consulting company in disguise would damage their valuation enormously. For reference, software companies can be valued at 10-40x revenue, whereas services companies are lucky to be valued at 1.5x revenue! So for a16z, one of the most “Valley” of the Silicon Valley investment giants to say “services will matter for AI” was big and interesting.

That a16z article was prescient. MIT agrees that strategic partners are the way to go. Companies building AI tools internally were failing to even finish the products, while bringing in teams to help provided instant expertise, those enterprise pilots that did succeed were built through strategic partners.

Finally, OpenAI has just announced a Palantir-style services team, indicating that the biggest consumer-oriented AI company sees the problem too. This transition is happening in real time. Palantir has always valued the concept of the “forward-deployed engineer, " which, to you and me, is an engineer running their software on behalf of a customer. However, the ability of this individual to provide value to the customer and then go back to the product company they serve is undeniable. You could also call this staff augmentation, or a managed service.

Or simply, a services-led product strategy.

As we move towards a future of geospatial foundation models, where the value still needs to be properly quantified*, we will see a similar pattern. Humans will be necessary to sculpt the commercial workflows to ensure operational alignment. Even with the traditional geospatial technologies we have today, greater use of services teams would create a better return for the EO pilot investment. Sigh, yes, we are still doing those.

Clearly, geospatial technology products, whether it’s Earth observation, data platforms, location tech, routing, or GeoAI** could all benefit from a services strategy. To some extent, our friends at Esri have always approached their market with a serious consulting component to their revenue mix. As a private company, they have always been less prone to the foibles of market valuation.

*To be clear, I believe GFMs to be an inevitability. They WILL be the way to go at some point. At this point, they are not ready for deeply commercial use cases. But the present innovative activity is absolutely necessary, and our entire community will benefit from continued and vigorous experimentation. Valuable use cases will emerge. At Sparkgeo, we are certainly investing time and effort in GFMs.

**My own personal (dystopian) Idaho, the name, not the practice!