Stop Waiting for the Digest

One of the clearest public examples comes from the St. Gallen Endowment for Prosperity Through Trade, which published the analytical setup behind its Iran Conflict Scenario Monitor in 2026. The evidence is borrowed: the subject is a geopolitical crisis rather than trade or investment. The operating model still transfers, and the team has documented it in unusual detail.

The monitor does not wait for headlines. In the pipeline the Endowment documented in 2026, a structured collection process runs three times a day before any analysis begins. That same documentation puts a single cycle at roughly 38 searches and 200 to 400 results scanned, a dozen or so pages read in full, and 25 to 30 verified developments admitted to the intelligence file.

Everyone can collect more information. The value is in rejecting more of it, consistently and according to rules. In the Endowment's pipeline, duplicates come out, undated material is downgraded or excluded, and an old development is treated as context rather than news. Claims that matter are checked against the original source rather than a search snippet. The system also watches its own source mix. The Endowment's 2026 documentation sets a cap of 40% of citations for any single outlet and requires at least one non-Western source every cycle, so the monitor does not end up repeating a single media environment's view of the world.

It even logs where searches return nothing. In a watched domain, silence can be useful. A quiet week in a competitor jurisdiction tells you something, and so does the moment that silence breaks.

Part of the search surface is deliberately adaptive. The Endowment's setup documents eight subject domains queried every cycle, with a separate scanner that reviews each day what was collected against what might have been missed, then writes new targeted queries to close the gaps. New lines of inquiry are added and retired as the situation evolves, so the monitor reshapes itself around the problem instead of waiting for a human to reconfigure it.

None of this is magic. It is good analytical hygiene, made systematic. These are the habits we try to teach junior officers: check the source and the date, and read the original. The trouble is that busy teams skip those steps because the day gets away from them. A well-designed system does not get tired or bored, and it never decides that this week is too busy to check the original.

The judgment underneath the dashboard is where the professional expertise sits. It decides what the system watches and what would change a decision.

The St. Gallen team's own emphasis is that the governance layer matters more than the model. The person who designs the system is the one encoding professional judgment into its rules. The setup the Endowment published in 2026 describes 14 separate analytical agents, each grounded in a different theoretical framework, each assessing the same evidence in isolation so they cannot herd toward an easy consensus, with a red team whose standing job is to argue the strongest case against whatever everyone else concluded. A trade or investment monitor almost certainly does not need 14 competing analysts. But the principle is portable: a system is only as good as the judgment encoded into what it watches and what it is forced to challenge.

This is also where a lot of off-the-shelf monitoring falls short. If you work in investment attraction, you know which competitor jurisdictions are relevant and which only look similar from a distance. If you work in trade promotion, you know which regulatory changes could affect exporters and which importers matter.

That knowledge is the system design. The point is to encode enough of your judgment that the machine can look in the right places, with more consistency than a human team can manage on its own. A monitoring service gives you someone else's view of relevance. A system you build reflects yours.

The problem with most intelligence products is that they do not force much thought. A digest can be useful, but it is easy to skim and forget, and it gives the feeling of being informed without changing the work.

A useful dashboard should be more demanding. It should tell you what changed and what deserves a closer look. For an investment team, that might mean tracking rival jurisdictions for incentive changes and major project wins, because the pattern shows where competition is heading. For a trade team, it might mean watching a target market for regulatory notices and procurement signals. For a mission team, it might mean keeping a country brief alive between the planning meeting and the wheels-up date, since too many mission briefs are accurate when drafted and stale by the time they are used.

A dashboard that watches everything gives the team one more thing to ignore. The purpose is useful coverage, not comprehensive coverage.

There is an obvious objection here, especially in government: our IT people will never allow this. Sometimes that is true. Often it depends on what you are trying to build, because there are two different use cases, and they should not be treated as one.

The first is an external dashboard built only on public information, with no confidential company information and no cabinet material. This is where most teams should start. Public information does not mean there are no rules; you still need to use tools your organization permits and avoid putting sensitive material where it does not belong. But the risk profile is much lower than anything involving internal files.

The second use case is internal. That version may be more valuable, but it belongs inside approved infrastructure, with privacy and data-classification rules built in from the start. It is not a weekend experiment.

None of this makes the dashboard an intelligence product on its own. A pile of scraped links is not analysis, and a chatbot summary is not judgment. A polished interface can still be wrong, out of date, or biased. The discipline matters more than the tool.

The St. Gallen team is candid about this. The Endowment calls its own system better than headlines but well short of intelligence, and the monitor's published FAQ describes a machine layer that extends human reach, is honest about its limits, and is meant to sharpen expert judgment rather than replace it. That instinct is the one to copy. The system needs dated, sourced, verifiable inputs and a diverse mix of sources. It has to separate a new development from an old fact being repeated, show disagreement rather than smooth it away, and report what it did not find as well as what it did.

And it still needs a human who understands the file. A machine can gather and compare. It can help a small team see more of the field and make weak signals easier to notice. It cannot tell you whether a minister should take the meeting, or whether an investor is serious rather than testing incentives. It cannot read the relationship history behind a company visit, or the significance of what was not said in a meeting. It can sharpen the question. You still have to answer it.

For years, better intelligence mostly belonged to the organizations with bigger budgets and better subscriptions. That advantage is eroding. A small office can now build a public monitoring system that would have required a dedicated analyst or an external provider not long ago. A regional team can track competitor jurisdictions with more discipline than a generic digest provides. The advantage is shifting from who can buy the feed to who can design the system.

That is good news for small teams willing to learn. It is less comfortable for teams still waiting for someone else to summarize the market.

The first version does not need to be ambitious. Pick one question your team cares about: which jurisdictions are beating you in your target sectors, or which companies are showing early signs of expansion. Then build around that question.

Do not start with the tool; start with the judgment. Decide what should be checked, and what would challenge the team's current assumptions. That is the useful work.

The case rests on a gap that vendors have not closed, and it would fail if they closed it. If an off-the-shelf service can be told which competitor jurisdictions matter to one agency, and then rejects material as strictly as the Endowment's pipeline does, the argument for building your own shrinks to a procurement decision. Watch for that over the next two or three renewal cycles.

The full build is left out here, the exact tools and the prompts that make one of these run, because it is better shown than described. A walkthrough is coming that builds a working external dashboard from scratch, so you can watch one come together and then make your own.