A Spate of Questions
All of them centering around the existential question “where is the moat?” The coordinates for the drawbridge are changing daily. The answers? They too are changing. Staying current is a full-time job. At Outsell, it’s our entire company’s job.
The questions come in various flavors and in various ways.
- How do I manage token costs and budget for them as they become more material? Can I move them to my customer?
- How long will enterprise customers and their CFOs put up with token costs? Will they go the way of the dodo? If so, when?
- Where is the best point in the AI integration layer for my (customers/content/solutions), and what’s the best way to monetize it?
- How do I integrate into AI tools whether the customers, our own offerings, big tech’s – so customers have what they want in the way they want, when they want?
- How do I license for any of this in an MCP world that leans to the transactional without destroying ARR?
- If AI ‘can do this’ with Claude code and ‘good enough data’ relative to commercial offering, will my ARR multiple even hold up?
- When does 70% good enough go to 80 or 85% and the entire business goes kaput?
- What are the user, functional, decision-risk types, and decision-consequences, where 80 or 85% accuracy simply isn’t good enough? Will there be any?
- What happens when the output data becomes so sophisticated the input data is no longer needed?
In one client’s words:
“If I want to say yes to clients’ desired use of data in their internal AI tools, what do I need to do to preserve the long-term value of my data?” and “What are (will be) the highest value kinds of data and content over the next five years? Where will the biggest moats exist among companies who have developed the data/content well, and what are the key factors to a strong moat?”
In the words of my colleague Hugh:
“...The stronger strategy is to partner and become critical infrastructure to models. Get content ready for customers to ingest into off-the-shelf models. This also avoids paying the cost of tokens, which is killing profits.
AI companies are also redesigning their models. The first generation relied heavily on indiscriminate internet scraping, pulling vast quantities of low-quality, unreliable, and sometimes dangerous material into training corpora. The next phase is about replacing that with authoritative, trusted, domain-specific content and data.
Today, much of the filtering happens at the output stage, with AI companies trying to identify and suppress harmful or unreliable responses after they have been generated. That quality control needs to move upstream to the input. Better models start with better source material. If poor quality content is excluded from the underlying corpus, there is less of it for the model to reproduce.
For information companies, the opportunity is to become a foundational part of that higher quality AI infrastructure: supplying the trusted content and data that the next generation of models depends on.”
My analytic sparring with Hugh goes like this: this strategy only works if the data changes quickly enough, the decisions are high-risk enough, the vertical is niche enough, the accuracy requirements are high enough, AND the derivative data is all that’s used to leave the crown jewels present but never licensed away.
What’s your take?
So many questions. Join us for insights and answers at the Outsell Summit October 1, 2026, in London. A day well worth the time invested if any of these questions are on your mind.
By invitation for CEOs, COOs, MDs, Presidents, and Divisional P&L owners in the data, publishing, and information economy. Space is limited. Be in touch to secure your seat at the table today.