Evaluating one adjacent market segment properly takes an analyst about 3 days. Use cases, buyer priorities, applicable regulation, who already serves it, how saturated the coverage is. Multiply that across every segment worth a look, and the reason most teams assess 2 or 3 a year becomes obvious.
So the shortlist gets drawn before the analysis starts. Someone proposes 3 segments in a planning meeting, those 3 get researched, and one of them wins. The decision looks like the output of a wide search when it is closer to an audit of whatever was suggested in the room.
That is a cost problem rather than a rigor problem, and the cost changed this year.
What changed
The Model Context Protocol reached close to 500 million monthly downloads across its Tier 1 SDKs and moved under Linux Foundation governance, backed by AWS, Google, Microsoft, and OpenAI. Connecting an AI system to an external source went from an engineering project to a configuration step.
That matters once you look at what sits on the other end of the connection.
Searching over documents gives you passages that resemble your question, and someone still has to read them and work out what they mean together. Structured intelligence gives you entities and the relationships between them: which company covers which capability, which capabilities a buyer segment prioritizes, which regulation governs a use case in a given jurisdiction, who moved and when. An agent can traverse that, going from a segment to its use cases, then to the buyers with that need, then to the vendors already serving them, without a person stitching the hops together.
The constraint moves as a result. You stop being limited by what you can afford to research and start being limited by what you think to ask. The State of Context Engineering 2026 survey found 81% of AI infrastructure leaders expect the next phase of AI maturity to be defined by systems that gain value as knowledge accumulates, and 94% consider that compounding quality essential for production-grade work.
Here are 5 questions that become answerable, in the order a where-to-play decision needs them.
1. Which segments should we be evaluating at all?
Segment evaluation is expensive because it is 5 research tasks stacked together. What use cases define this segment, what capabilities do buyers there prioritize, which regulations apply, who already serves it, and how saturated is the coverage already.
Asked as a single traversal, that stack collapses into one query, which means you can run the same evaluation across 15 segments instead of 3 and rank the results. The segments worth entering are frequently the ones nobody thought to put on the whiteboard, and a shortlist drawn in advance is designed to exclude exactly those.
2. Where could we be displaced, and where could we displace?
Most competitive analysis quietly gets this wrong, and it is the section of a where-to-play decision that punishes the error hardest.
“Is B an alternative to us?” and “are we an alternative to B?” have different answers. A platform covering 40 use cases genuinely substitutes for a specialist covering 3 of them, while the specialist does not substitute for the platform. The relationship runs one way. Almost every similarity-based approach treats it as running both ways, including a general model reasoning over vendor marketing material, and the resulting competitive set looks reasonable while pointing strategy at the wrong ground.
Structured capability and use case coverage makes that asymmetry computable. You can ask which companies could displace you in a specific segment, then ask separately which companies you could displace, and get two different lists. The second list is where entry opportunities live.
A segment where you substitute for the incumbents and none of them substitute for you is defensible. A segment where the reverse holds looks attractive on a market sizing chart and rarely survives contact with a sales cycle.
3. What do buyers in that segment actually prioritize?
The usual input here is what competitors shipped, because it is visible and concrete. It also lags the market by however long it took them to build it, and it encodes someone else’s guess about what buyers wanted 18 months ago.
Buyer capability priorities set against current coverage saturation are a better input. A capability that 90% of the market already covers is table stakes no matter how recently a competitor announced it. A capability buyers weight heavily with thin coverage is where an entry earns a return. Asked against structured demand and supply data, that stops being a survey commission and becomes a query, which changes the entry question from whether you can build what they have to what these buyers want that nobody has built.
4. Who is already moving here?
Competitive tracking usually produces a document that is accurate the day it ships and decays from there. By the time it reaches a planning discussion, the movement it describes is a quarter old.
For a where-to-play decision the movement is most of the signal. Funding into a segment, acquisitions that reshape who covers what, capability launches, entrants arriving from adjacent markets. A segment that looked open 6 months ago may have been consolidated by an acquisition that never made the trade press in your vertical.
Querying those signals at the moment of the decision, rather than reading them off a deck someone last updated in March, is a different exercise.
5. Do we enter alone or with a partner?
Once a segment is chosen, entry mode is the next call, and partner identification has had almost no tooling behind it. Most teams invert a competitor list by hand and look for the gaps.
That approach struggles because complementarity weights differently than substitution. Overlapping geography and buyer segment help a partnership and hurt a competitive comparison, while functional overlap works the other way around. Inverting one list does not produce the other.
Asked directly against structured coverage data, the question stands on its own: which companies reach the buyers we want, in the markets we want, with capabilities adjacent to ours rather than overlapping. Answering it properly turns entry mode into an analysis instead of a function of who the CEO happens to know.
The part that compounds
The most useful property here is not any single query. It is that an agent can reason across your own material and verified market structure in the same call.
Your positioning documents, win and loss notes, roadmap, and prior segment research are context nobody outside the company has. Market structure is context you cannot build yourself. A strategy memo grounded only in the internal view comes out confident and parochial, while market data without your context comes out accurate and generic. Neither one alone gets you to a decision.
Queried together, the question sharpens into something more specific: given what we already believe about our differentiation, which segments does the market structure actually support. That question used to require a consulting engagement and now fits into a Tuesday afternoon.
It also accumulates, since the internal context deepens every quarter and every query built on it returns a little more than the last one.
Where to start
Start with the list you already have. Most strategy, product, and revenue teams keep an informal set of questions parked as too expensive to answer. Write them down, then separate the structural questions about market relationships from the genuine judgment calls. The first group changed price this year. The second never will, which is fine, because judgment gets easier once the structural work underneath it is done.
The teams that get value from this over the next 12 months will be the ones who noticed their research agenda had been shaped by cost, and rebuilt it once the cost moved.
Key takeaways
- Where-to-play decisions get distorted by research cost, which turns a wide search into an audit of a shortlist drawn before the analysis started.
- Structured market intelligence lets an agent traverse the relationships between segments, use cases, buyers, capabilities, and regulation in one pass.
- Substitution runs one way. A segment where you substitute for the incumbents and none substitute for you is defensible, and similarity-based analysis misses the distinction.
- Buyer capability priorities set against coverage saturation beat competitor feature launches as an entry input, because launches lag the market.
- Partner identification weights differently than competitive analysis, so inverting a competitor list does not produce a partner list.
- The compounding advantage comes from reasoning across your own context and verified market structure in the same query.
See how this works in practice: the Liminal MCP connects any MCP-compatible AI to structured, source-attributed intelligence across identity, fraud, financial crime, and security.
