Global News & Trends Intelligence
How We Build and Run a Monitoring Assignment
Monitoring is a craft before it is a tool. What follows is the method we apply to every subject: how a scope is written, how sources are chosen, how items are qualified, and how the result reaches you.
Two constraints shape everything. The first is that publication volume on any live subject exceeds what any team can read, so the value lies in what is discarded as much as in what is kept. The second is that automatic detection is fast but literal: it counts occurrences, it does not understand them.
Our method therefore splits the work. Machines handle scale, repetition and continuity: collection, deduplication, first-pass classification. A person handles judgement: what is ambiguous, what is genuinely new, what changes the picture, and how to say it in a form that supports a decision rather than describing the internet.
A seven-step method
Every assignment follows the same sequence, whether it runs for a week or for a year. The steps are the same; the depth given to each depends on the subject.
-
Understanding the need
Before any keyword, we look at the decisions the monitoring is meant to support and the people who will act on what they read.
- Which decisions, meetings or deadlines depend on this information.
- What already reaches you today, through which internal channels, and what it misses.
- What a useful edition would contain, and what would make it be ignored.
-
Defining the scope
The scope is written down: subjects, entities, territories, languages, period, and the exclusions that keep the result readable.
- Named subjects and entities, with their variants, former names and common misspellings.
- Territories and languages retained, and those consciously left out at this stage.
- Explicit exclusions: adjacent subjects, homonyms and recurring noise not to be reported.
-
Selecting the sources
Sources are chosen and ranked for this subject, not taken from a standing list. Coverage matters less than the ability to tell who publishes first.
- Wire services and news agencies, as the origin of most republished copy.
- Trade, specialist and institutional publications, often earlier and more precise than general media.
- Known gaps recorded openly: paywalled titles, non-indexed archives, sources with restricted access.
-
Collecting and qualifying
Collection runs continuously; qualification is what turns it into something readable. Every retained item is dated, sourced and checked for relevance.
- Syndicated and mirrored copy grouped, with the earliest publication identified.
- Ambiguous mentions verified by reading before being counted, particularly on common names.
- Items discarded for a stated reason, so the same noise is not reprocessed each cycle.
-
Human analysis supported by digital tools
Tools measure; people interpret. A rise in volume and a change in narrative are different events, and only one of them is visible on a curve.
- What is new is distinguished from what is merely being repeated more widely.
- Shifts in framing, vocabulary and the identity of those speaking are read explicitly.
- Automatic sentiment scoring is treated as an indication, never as a conclusion.
-
Producing alerts, syntheses or reports
The format follows the use: immediate when a decision may be needed within hours, periodic when the question is the state of a subject.
- Every statement attached to a published source, with title, outlet, date and link.
- What is uncertain presented as uncertain, with the reason for the doubt.
- A stable structure between editions, so recipients read faster over time.
-
Adjusting the set-up
A monitoring set-up left untouched drifts. Subjects change vocabulary, new sources appear, thresholds that were right in March are wrong in September.
- Periodic review of keywords, entities, sources and alert thresholds.
- Feedback from recipients used to correct what is over-reported or under-reported.
- Scope changes recorded in writing, so past editions remain interpretable.
Delivery formats
Ten formats cover the practical needs we encounter. An assignment usually combines two or three of them rather than relying on a single one.
Immediate alert
Triggered by a predefined event: a named publication, a regulatory decision, an unusual acceleration. Short, sourced, and explicit about why it was sent.
Daily summary
One edition per working day at a fixed hour, grouping new items by theme, merging duplicates and opening with what changed since yesterday.
Weekly bulletin
A calmer reading: dominant subjects of the week, notable arrivals, signals that faded, and the questions likely to surface next.
Monthly report
A trajectory rather than a feed. How the subject was treated over the month, which sources set the terms, and what has structurally moved.
One-off study
A single question answered thoroughly and once: what has been published, by whom, when, and what remains genuinely undocumented.
Benchmark
A comparison of how several organisations, markets or subjects are treated in the media over the same period and the same source base.
Mapping
A structured view of who speaks on a subject: outlets, journalists, institutions, commentators, and the relationships visible between them in publication.
Tracking table
A structured file with one row per item: date, source, language, territory, category. Made to be filtered, cited and reused internally.
Custom research
A precise question requiring targeted searching rather than continuous watching: a chronology, an attribution, the origin of a claim.
Strategic support
Regular working sessions in which the monitoring output is discussed with your team, and the set-up adjusted to the decisions ahead.
What we define together
Five parameters determine the shape of the assignment. They are fixed in writing at the start and revisited whenever the subject moves.
- Frequency
- Continuous, daily, weekly, monthly or one-off. Frequency should follow your decision cycle: an edition arriving after the meeting it was meant to inform has no value.
- Languages
- Working and delivery languages are English and French. Sources in other languages are monitored on request, with the treatment for that language agreed beforehand.
- Sources
- Which families are included, which are prioritised, and which are deliberately excluded. Known limitations, such as paywalled or non-indexed titles, are stated rather than hidden.
- Geographical areas
- Territories are chosen for relevance, not for coverage. Two countries watched properly usually serve a decision better than a nominally worldwide scope.
- Delivery channel
- How the output reaches you and who receives it: named recipients, format, hour of delivery, and what happens when something significant appears outside office hours.
Quality principles
- Nothing is reported without a source that can be opened, dated and named.
- What we do not know is written as unknown, rather than filled with plausible wording.
- Automatic scores, including sentiment and influence metrics, are presented as indications and never as findings.
- Only published and publicly accessible content is used; no private account, closed space or restricted database is accessed.
- Scope changes are documented, so a series of editions remains readable months after it was produced.
Questions about our method
Can you monitor a subject in a language your team does not read?
Sources in other languages can be included, with the working method agreed in advance: which titles, what depth of reading, and how items are rendered in your delivery language. We state where machine translation is used, because it affects nuance and therefore interpretation.
What happens with paywalled or non-indexed sources?
They exist on almost every subject and they are often the most substantial. We record openly which relevant titles are behind a paywall or absent from indexes, and agree with you what to do about them rather than pretending the scope is complete without them.
Do you use sentiment analysis?
As an indication only. Automatic scoring struggles with irony, quotation, negation and specialist vocabulary, and it frequently misreads factual reporting as negative. We may use it to sort large volumes, but any statement about tone in a delivery comes from reading the material.
How often should a set-up be reviewed?
For a stable subject, a review every quarter is usually enough. For a subject under active change, monthly is more realistic. The signals that a review is overdue are consistent: rising noise, alerts nobody opens, or an event that should have been caught and was not.
Tell us what you need to monitor
Describe your subject, your markets and your decision timeline. We study every request individually before proposing a monitoring set-up.
Every request is reviewed confidentially. No commitment is required to discuss a scope of work.