Why a Better Market Ranking Does Not Mean a Better Decision
There is a question coming up more often in buying conversations, and it is a fair one. If a competent research summary, a country comparison and a preliminary recommendation can be produced in an afternoon, what is a company actually paying for when it engages an adviser?
The question deserves better than a defensive answer. The evidence that analytical output has become cheap is not coming from technology vendors. It is coming from the consulting industry itself, which has spent the past two years restructuring how it prices work that used to be billed by the hour.
It is worth being precise about how much has changed. In a preregistered experiment conducted with Boston Consulting Group and published in Organization Science, 758 consultants completed realistic consulting tasks with or without artificial intelligence (AI) assistance. On tasks inside the model's capability range, those using it completed 12.2% more work, roughly 25% faster, with quality scores more than 30% higher.
Those are not marginal gains, and nothing here disputes them. Analysis has become abundant, and it has become good.
A Better Market Ranking Does Not Necessarily Mean a Better Decision
Market attractiveness scoring is the most automatable part of expansion work, and the output is usually sound. Market size, growth rate, digital maturity, ease of doing business, competitive density, purchasing power. Weight them, rank them, and a shortlist appears. The mechanics are sensible and the arithmetic is not the problem.
The problem is what the ranking is measuring. A market attractiveness score answers how good a market is in general. Almost no company needs that answer. What a company needs to know is how good a market is for the specific thing it sells, to the specific buyer it sells to, given the resources it currently has.
Those are separate questions, and the second one is not a refinement of the first.
Two Variables, Opposite Signs
RebusLabs, a Swiss IoT platform that tracks and monitors large-scale assets including temperature-sensitive inventory, was already trading successfully at home and trying to decide which European market to enter next. Germany was selected, but not on the grounds a ranking would have supplied.
Two factors carried the decision. German university hospitals hold a large installed base of high-capacity cold-chain refrigerators, which is exactly the asset the platform was built to monitor. And the digital maturity of the German health sector left considerable room for improvement.
The second factor is the interesting one. Low digital maturity is a negative on every digital readiness index in circulation. For a company selling digital monitoring into an undigitised sector, it is the opportunity. The same variable, read against a different product, changes sign.
A ranking cannot make that inversion. It applies a fixed weight to digital maturity because it has to, and a fixed weight assumes every user of the ranking wants the same thing from it.
Where the Ranking Would Have Landed
Germany scores well on market size in any European comparison, so a company following the ranking would probably have arrived at the same country. This matters more than it sounds. It would have been the right destination reached through the wrong reasoning, which means the entry would have been built around the wrong opportunity: general market scale rather than a specific installed base inside a specific institutional buyer.
The failure mode is not a wrong answer. It is a correct answer whose reasoning does not survive contact with execution.
Segment Blindness
Country scores are also blind to segment, and B2B purchasing behaviour is not consistent across company sizes. A mid-market e-commerce merchant and a multinational brand in the same market run procurement differently, on different timelines, with different approval structures and different tolerance for an unproven supplier. Enhencer, an AI platform predicting purchase intent for e-commerce brands, faced exactly this range when prioritising European and North American markets.
One country score cannot be right for both segments at once, because the thing that makes a market accessible to one is often what makes it slow for the other. Any single number applied across segments is an average of at least two answers, and the average is a market nobody is actually selling into.
The Pattern in Both Cases
Neither company lacked analysis. Neither lacked capital. RebusLabs was already succeeding in Switzerland. Enhencer had closed a funding round and was already winning clients in new markets.
Both were stuck on the same thing: which market to concentrate on. That is a selection problem, and selection is not what better analysis produces. Ranking countries is the input. Deciding which ranking criteria matter for this company, this product and this buyer is the output, and it is the part that has not become cheaper.
Where Market Entry Strategy Stops Being a Data Problem
Entry mode is where the gap between analysis and decision becomes visible, because it is the point at which a single dataset stops narrowing the options.
A market assessment that supports entering a country supports several ways of entering it. Direct entry with a local team. A distributor. A channel partner or reseller. Licensing. Acquiring a company already trading there. Each of these can be defended from the same market data, and the data rarely distinguishes between them, because what distinguishes them is not in the market at all.
What Each Route Actually Asks of You
Direct entry requires hiring capability, local management attention and tolerance for a long ramp
Distribution trades margin and customer proximity for speed, and depends entirely on which distributors are genuinely available rather than nominally listed
Partnership requires something the partner wants, which is a question about your product's position, not the market's size
Acquisition requires capital, integration capacity and a realistic view of what you are buying
Licensing requires enforceable protection and a willingness to lose control of execution
None of these are market variables. They are organisational ones, and no market analysis can supply them.
Two Columns, Two Conclusions
Investment data illustrates how far apart two readings of the same source can be. UNCTAD's World Investment Report 2026 shows the value of announced greenfield projects rising slightly in 2025, from $1,379 billion to $1,393 billion, while the number of projects fell 10 per cent. International project finance follows the same shape: values up 3 per cent, deal numbers down 11 per cent, with values still around a quarter below their 2021 peak. Cross-border acquisition values fell 7 per cent.
Read the value column and international investment is stabilising. Read the activity column and fewer deals are closing on every route. Both columns are correct and they support opposite conclusions about how difficult entry currently is.
These figures describe institutional-scale investment rather than a mid-market software company choosing a distributor, so the numbers do not transfer. The pattern does. Aggregate values are being held up by a small number of large transactions, which means the headline can rise while the median experience of entering a market gets harder.
Knowing Which Column to Distrust
This is the practical form of experience. Not knowing more variables, but knowing which of the available ones tends to mislead in a given market, and checking the other column first.
Prior exposure to a market does not add data. It changes the order in which the existing data gets interrogated, and it supplies the question that comes before the analysis: what would have to be true here for this route to work, and is it?
Judgement Is the Ability to Say the Analysis Is Wrong
Judgement is easy to invoke and hard to define, and vague appeals to human intuition are not an argument. In an expansion decision it means five specific things:
Deciding which variables carry the most weight for this company rather than in general
Identifying what is missing from the dataset entirely
Recognising when a comparison is structurally misleading
Being explicit about what is uncertain rather than presenting a range as a finding
Naming what would change the recommendation
The last one is the test. An analysis that cannot say what would overturn it is not a recommendation. It is a conclusion with supporting material attached.
The Reversal in the Same Study
The Boston Consulting Group experiment cited earlier produced a second finding, and it is the one that matters here. Alongside the tasks inside the model's capability range, the researchers included a complex managerial task deliberately positioned outside it, combining quantitative and qualitative evidence in a way that led to a plausible but wrong answer.
On that task, consultants using AI were 19 percentage points less likely to reach the correct conclusion than those working without it. The output was not less coherent. It was more persuasive.
That is the failure mode worth taking seriously. The problem is not obviously bad analysis, which anyone can catch. It is well-structured, internally consistent, confidently expressed analysis that is wrong, and the confidence is doing the damage.
Reviewing Is Not the Same as Approving
The researchers found that the consultants whose performance declined tended to adopt the output without interrogating it. They were not inexperienced. They were skilled professionals at a leading firm, and the more capable the assistance became, the less independent effort they applied.
This has a direct commercial equivalent. A company can now generate a defensible market entry recommendation internally in a day, and then hire an adviser to review it. If the review consists of confirming that the reasoning holds together, it has added nothing, because internal consistency was never the weak point. The recommendation was already coherent. That is precisely why it is dangerous.
A review that only validates is not a review. It is a signature.
What a Real Challenge Looks Like
The useful version is uncomfortable and specific. It identifies the variable the analysis weighted wrongly for this particular business. It names the thing the dataset never contained. It states what would have to be true for the recommendation to hold, and what evidence would kill it.
In the RebusLabs work this took the form of an alternative plan tree, mapping how the go-to-market approach would progress under different circumstances alongside budget projections for each. The value of that document is not the primary path. It is that the primary path was stated as conditional, with the conditions written down, which makes it possible to know later whether the assumption failed or the execution did.
That distinction is impossible to recover after the fact if nobody wrote down what the plan depended on.
Why This Is Getting Harder, Not Easier
The incentive runs the wrong way. Challenging a coherent recommendation is slower, less comfortable and harder to justify than endorsing it, and as analytical output improves, the endorsement becomes more defensible each time. The reviewer who signs off is rarely wrong in any given instance. They are simply not contributing, and the pattern only becomes visible when a market that scored well underperforms and nobody can say which assumption broke.
From Producing Answers to Owning Recommendations
The commercial question, then, is not whether analysis retains value. It does, and it is now available to almost everyone at almost no cost. The question is what remains scarce once it is.
What remains is the willingness to be answerable for a judgement. Producing a recommendation and owning one are different acts, and only the first has become cheap. A recommendation that nobody is accountable for is a document. A recommendation somebody has staked their credibility on is a decision, and the difference shows up eighteen months later, when the market has either worked or it has not, and someone has to explain which assumption failed.
That is not a defence of expertise as a category. It is a narrower claim. The parts of advisory work that consisted of gathering, comparing and synthesising were always the parts most exposed to automation, and their exposure has now been priced. What is left is interpretation, challenge, and responsibility for the call, which is harder to sell, harder to demonstrate in a proposal, and harder to automate.
Companies buying expansion advice should probably be asking a different question than they were three years ago. Not what analysis they will receive, but who will tell them the analysis is wrong, and what that person is prepared to be held to.
Frequently Asked Questions
Can AI do market entry research?
Yes, and it does the research layer well. Market sizing, competitor comparison, regulatory summaries and preliminary country shortlists are all within reach of current tools at low cost. What it does not do is decide which of those variables matter for a specific company, or identify what the dataset never contained. Research is an input to a market entry decision, not the decision itself.
What is the difference between market attractiveness and company-market fit?
Market attractiveness measures how good a market is in general, using variables such as size, growth, purchasing power and digital maturity. Company-market fit measures how good that market is for one company's specific product, buyer and current resources. The two often disagree. A market can score highly on attractiveness and still be wrong for a particular business, and a variable that counts against a market on an index can be the actual opportunity.
How do you choose the right market entry mode?
Entry mode is decided by organisational capability rather than market data. The same market assessment will usually support direct entry, distribution, partnership, licensing and acquisition. What separates them is hiring capacity, local management attention, margin tolerance, capital available, integration ability and appetite for risk. Start from what the business can realistically execute and sustain, then test which routes the market actually permits.
Is AI-generated market analysis reliable?
It is reliable inside its capability range and unreliable in a specific way outside it. In a preregistered experiment with 758 consultants at Boston Consulting Group, published in Organization Science, AI assistance improved output on suitable tasks but reduced correctness by 19 percentage points on a complex task designed to fall outside the model's range. The output on that task was more persuasive, not less. The risk is confident, coherent analysis that is wrong.
What should companies expect from a market expansion consultant now that analysis is cheap?
Interpretation, challenge and accountability rather than research volume. A useful adviser identifies which variables were weighted incorrectly for that business, names what the analysis omitted, states what would overturn the recommendation, and stands behind the judgement. If the engagement consists of confirming that an existing recommendation is internally consistent, it has added little, because coherence was rarely the weak point.
How do you know whether a market entry recommendation is wrong?
Ask what would have to be true for it to hold, and what evidence would disprove it. A recommendation that cannot answer either question is a conclusion with supporting material attached. Recording the conditions the plan depends on, alongside alternative paths and their budgets, makes it possible to tell later whether an assumption failed or the execution did. That distinction cannot be reconstructed after the fact.
How Metheus Can Help
We work with technology, SaaS and fintech businesses on the part of expansion that analysis alone does not settle: which market to concentrate on, which entry mode the organisation can realistically sustain, and what would have to be true for the plan to hold. Our approach is data-oriented throughout, but the data is where we start rather than where we finish. We bring direct operating experience in the markets our clients are entering, and we tell them when a recommendation does not survive scrutiny, including our own. The objective is a decision somebody is prepared to stand behind, not a document that agrees with itself.