The Network You Cannot Assemble at Short Notice

The phrase "local expert" describes two things that are not related. One is a person who knows a place. The other is a person whose knowledge of a place has been measured. Only the second is an asset, and the difference is invisible until the moment it matters.

This piece is about the measurement, which is the part of the discipline nobody describes because it is slow, unglamorous and impossible to demonstrate at the point of sale.

Reliability is a property, not a credential

The formulation on this firm's own pages is precise and worth reading closely: networks of local influencers, community leaders, academics, religious authorities, officials and businesspersons, drawn from all key sectors of society, retained as full and part-time consultants, and tested for reliability and accuracy over time.

That final clause carries the weight. It says reliability is established rather than assumed, established across time rather than at engagement, and established by testing rather than by reputation.

Consider what the alternative rests on. A credential — a title, an institutional affiliation, a record of publication, a recommendation from someone trusted — is evidence about position. Position predicts access. It does not predict accuracy, and the correlation between the two is weaker than intuition suggests, because the people with the best access frequently have the strongest reasons to shade what they pass on.

What testing actually consists of

Three mechanics, none of which requires the source to know they are being assessed.

Tasking against known answers. A source is asked, in the ordinary course, about something the analyst already knows independently. The answer is not needed; the calibration is. This establishes whether a source distinguishes between what they know and what they have heard, whether they will say they do not know, and whether the confidence in their delivery tracks the accuracy of their content.

Tracking calibration over time. A single comparison is nearly worthless. What is useful is a record across many taskings — how often this person was right, on what kinds of question, with what stated confidence, and in which direction they err. Sources are rarely uniformly good or bad. They are good on some categories and systematically unreliable on others, and the value of a long record is that it identifies which is which.

Separating what a source knows from what a source believes. Every well-placed person holds both, delivered in the same register and frequently in the same sentence. Distinguishing them is analytic work performed on the reporting, not a question that can be asked directly, because the source does not experience the two as different.

None of that can be compressed. A calibration record requires the taskings to have happened, spread across enough time and enough question types to mean anything, and it cannot be assembled retrospectively because the whole point is that the assessments were made before the outcomes were known.

The fixer economy

Set against that, the alternative available at short notice: a name obtained through an introduction, in-country, willing to help within days.

The person may be excellent. The problem is that nothing about them is known in the relevant sense. Their interests in the local situation are unmapped, and everyone in a place has interests in it. Their reliability has never been measured, so their confidence carries no information. And their incentive is structurally adverse: they have been engaged to produce an answer, they are paid for producing one, and "I do not know" reads as failure to deliver. That incentive does not require anyone to be dishonest to distort output.

This is the reasoning behind the stated position that this firm does not rely on local generalists — not a judgement about the competence of any individual, but a statement about what can be known about an unmeasured source under time pressure.

Why a network cannot be bought

The obvious shortcut is acquisition — buy a firm that already has the network, or hire the person who holds it. Both are done routinely and both under-deliver, for reasons that follow from what the asset actually is.

A calibration record belongs to whoever made the assessments. It consists of judgements about how much weight to give particular people on particular subjects, and most of it has never been written down; it lives with the analyst who accumulated it. Acquiring the firm acquires the contact list, which is the least valuable component, and the relationships transfer only to the extent the individuals do.

Hiring the holder has a subtler problem. The relationships were built by a person acting for an institution, and a source's willingness to speak frequently attaches to both. Some of it moves; some does not, and which is which cannot be established in advance.

The result is that a purchased network arrives with its measurement stripped out. It looks identical on paper to the thing that was valuable — the same names, the same sectors, the same countries — and it will not perform the same way, which the buyer discovers under exactly the conditions where discovery is most expensive.

The honest constraint

A network built this way has gaps, and they are not distributed where a client would choose.

Coverage follows history: it exists where relationships happened to be built, sustained through years of work that had its own reasons, and it is thin or absent where that history does not exist. A network cannot be extended into a new territory quickly for the same reason it is valuable — the property that makes it worth having is the measurement, and the measurement takes the time it takes.

So the honest report includes the map's holes. A firm claiming comprehensive coverage of a region is claiming to have measured reliability everywhere, which is either untrue or means something weaker by "coverage" than the client will assume.

The asymmetry worth ending on is one of timing. Everything described above is built during periods when nobody is asking for it, funded against demand that has not appeared, and its value is realised in compressed windows that arrive without notice. That is a difficult proposition to hold as a business and an easy one to abandon quietly — and the abandonment is undetectable from outside until the window opens, which is the only moment the difference between a tested network and a list of names becomes visible.