I had an interesting exchange with Benedict Evans over on LinkedIn, prompted by this post:

The obvious reason for Amazon to block Muse is a trade-off that goes back 25 years, and 150 years - do you want to give up customer ownership in exchange for distribution? Generally, if you’re small you pick distribution and if you’re big you pick ownership.

But the more interesting issue is that Amazon sold $75bn of ads in the last 12 months. Ads are big business for a lot of retailers now and they’re most of the profits at places like Instacart and DoorDash. Never mind losing your upsell - the agent won’t see the ads. And on the other side, no-one really knows what agents will do to returns and fraud, but it won’t be pretty, and that was already part of why people block bots.

Prompted by some recent thoughts of my own about “customer ownership”, I responded:

Customers might well object to the notion that it is Amazon’s data to own. And we are seeing the same conflict play out in the enterprise space, e.g. with Salesforce blocking Glean. It’s far from settled what the landscape will look like. More here: [https://findthethread.blog/Personal-Data-Silos/(/Personal-Data-Silos/)

To which Benedict curtly replied:

Dominic Wellington I didn’t mention data ownership.

Which is a fair point; in trying to keep my point pithy, I had elided that important difference. I clarified:

Benedict Evans Neither you did, but I suspect if (big if) Muse takes off the way Meta hopes it will, data ownership will become a competitive issue, leading to potential changes in what had previously seemed pretty solid customer ownership.

Here on my own blog, I don’t even try to be pithy, except sometimes in the titles, so I can expand a bit more on that thought — because it is quite critical to AI adoption, on both consumer and enterprise fronts.

In both cases, the issue at hand is that one entity (Amazon, Salesforce) is sitting on a whole trove of data that is generated by their customers, either directly or through analysis of their usage of the platform. Because they “own” that customer, they can leverage that data to provide various services, from simple “people who bought this also bought that” product recommendations, to the full Agentforce AI stack. And because of the value of the data and the services, and the difficulty of moving the one and replicating the other elsewhere, customers tend to stick around. This is known as “data gravity”: a sufficiently large amount of data will generate its own attraction and make it difficult to escape.

Now a second entity (Meta, Glean) would like to access that data to deliver services which are presumed to be valuable to those same customers.1 Because of the value of the data that they “own”, Amazon and Salesforce block the newcomers.

SQUEEEEEEZE

The point that I was trying to make to Benedict is that the two aspects of ownership of the customer and ownership of the data are intimately related. The data gravity is stronger as users receive more value from the services which rely on that data. But if the value provided by the third-party entrants (Meta and Glean) is high enough, the situation becomes unstable.

On the one hand, Amazon and Salesforce are reluctant to weaken their grip on the customer by granting access to the data. Those ancillary services do not just have value to customers; they also have value to the vendors, as in the example of the inline apps within Amazon that Benedict himself cites. If a customer is accessing Amazon via Meta’s Muse, or Salesforce via Glean’s enterprise search, they are not encountering those services, and Amazon and Salesforce lose an opportunity to monetise that interaction.

But on the other hand (and here is the point I was trying to make on LinkedIn), if Amazon and Salesforce lock access to data down too hard, customers may start asking inconvenient questions, such as “hang on, whose data is it anyway?”. This is the sort of thing that doesn’t make much of an impression at an individual level (“where are you going to go instead? haha, lol”), but over time and at scale can reduce the pull of the data. After all, if customers can’t use it in the ways they want to, it becomes less useful to them. Weaker data gravity maps directly to weaker “ownership” of the customer (and wow, we should really unpack that term some other time).

This is the premise of the SaaSpocalypse in the enterprise world, and of “agentic shopping” on the consumer side. Both imply disintermediation of existing customer relationships. After all, if you are chatting back and forth to your agent, do you really care what back-end services are powering its responses, as long as you get what you need? It’s the same with self-driving cars: assuming for a moment that they ever happen, you won’t care about the badge on the front — which is great for you, but sucks for the car makers, who have invested enormous sums in building a customer relationship that they had hoped would be lasting.

And of course you don’t see any ads if you are letting your agent do the browsing — although of course in that scenario the agent may be seeing ads of its own, designed to influence its behaviour, and through it, your own.

🖼️  Photo by Raquel Baires on Unsplash

  1. For the purpose of this discussion, we will leave out my immense scepticism of Meta as a company, Zuckerberg personally, and all of their works. ↩