Use

Intelligence

A watch follows a topic on RSS feeds and on LinkedIn. On each run, the platform reads your sources, spots the useful trends and writes a synthesis with its sources. You can then draft a post from a trend, or import the people who reacted to a LinkedIn post.

This chapter explains how to create a watch, read its trends, schedule a daily run and go from a watch to a post or to a prospect.

1. Create a watch

Available

To create a watch from the screen:

  1. Open "Intelligence".
  2. Click "New watch".
  3. Enter the "Topic" and the "Keywords" (one per line or separated by commas).
  4. Add at least one source: an RSS feed or a LinkedIn source (see section 2).
  5. Choose a "Target" if you want the relevance of trends judged against an audience. This choice is optional.
  6. Read the cost of the first run, then click "Create watch".

A first run starts when the watch is created. The daily run is not turned on: it is offered to you afterwards, with its maximum cost per day.

You can also click "With the assistant" to describe the watch in plain language (see section 8).

2. Sources

Available

RSS feeds. Enter one RSS or Atom feed address per line. Each address is read once when you save, to check that it works. RSS feeds do not use LinkedIn reads.

LinkedIn sources. They need a connected LinkedIn account. Each source reads the posts from the past week, 20 at most per run:

SourceWhat it reads
KeywordsLinkedIn posts that contain your keywords, most relevant first.
NetworkThe same posts, limited to your connections and the people you follow.
AuthorsThe latest posts of the people or companies you follow. 10 authors at most per watch.
Home feedYour account's home feed. This source is not turned on in production today (Planned).

To add an author:

  1. In "Followed authors", choose the "Author type": "Person" or "Company".
  2. Enter their name.
  3. Click "Search". A search costs 5 credits. LinkedIn returns 5 results at most.
  4. Click "Add" next to the right result.

LinkedIn job ads. Next release. This source searches LinkedIn job ads that carry the keywords of your offers (My account → Offers). A company hiring on a topic you sell is a buying signal.

  1. Tick "LinkedIn job ads".
  2. In "Tracked offers", tick at least one offer. Without an offer, the watch is not saved.
  3. On each run, the watch makes one search per keyword, 12 keywords at most. The first run covers the last 30 days, the next runs the last 7 days.
  4. The number of job ads collected shows in the watch sources. The "Hiring signals" tab groups them by company (see section 10).
  5. After the collection, each ad gets a score from 0 to 10 against your offers. Known job boards, IT service firms and recruitment agencies are set aside without a model call. For an ad scoring 4 or more, the watch reads its full description, then scores it again. An ad scoring 5 or more, from an end client, with a named offer, becomes a signal.
  6. The last run line shows "N ads read · M signals".

Job ad searches have their own daily cap per LinkedIn account, separate from the post-read cap. A description read also counts against this cap; when it is reached, the remaining descriptions are read after midnight (UTC). Their credit price will be set after one month of measurement; today neither searches nor description reads are charged.

3. Runs, relaunching and the daily run

Available

A run reads all the sources of the watch, scores each item found and produces a report.

Cost of a run:

  • the LinkedIn read: 5 credits per run, if the watch has at least one LinkedIn source. It is charged only if LinkedIn answers;
  • the report: 20 credits. It is charged only if the run finds signals.

First run. It starts when the watch is created.

Relaunch. The "Relaunch" button reads all the sources again and produces a new report. It shows the cost, for example "Relaunch · LinkedIn 5 cr + report 20 cr". Only one run at a time: wait for the current run to finish.

Daily run. It relaunches the watch automatically every night, once a day at most. It commits you to a repeating cost. To turn it on:

  1. After creation, click "Turn on daily pass".
  2. Read the maximum cost per day: 25 credits with a LinkedIn source, 20 credits without.

The watch panel shows the last run: its state, the number of LinkedIn posts and RSS articles read, and the state of each source ("read", "no content this time", "daily LinkedIn quota reached", "LinkedIn account disconnected", etc.).

Available

The "Intelligence" screen shows:

  • the "Synthesis": a dated report that cites its sources;
  • the "Detected trends": each trend has a relevance (high, medium, low), a source and, for a LinkedIn post, its reactions, comments and reposts at the time it was read.

To sort the trends, choose "Relevance" (default) or "Engagement". The "Relevance ≥ …" filter hides less relevant trends; "Show all" brings them back. The "Cited" badge marks a trend used in the synthesis.

Click a trend to see its detail: "What the source says", "Why it is relevant", the "Entities" mentioned, the source and the detection date.

The text of the sources is extracted and summarized before it is shown or used by the AI. See Limits and security.

5. From trend to post

Available

To draft a post from a trend:

  1. Open the trend.
  2. Click "Suggest a post".
  3. "Publish" opens with an instruction already filled in. Complete it with your expertise.
  4. Click "Generate" (15 credits).

Nothing is drafted before you click. The trend detail then lists the "Posts written from this trend". See Publish.

6. People who reacted to a post

Available

When a trend comes from a LinkedIn post, you can see who reacted or commented, then import these people as prospects.

To do this:

  1. Open the trend.
  2. In "People who reacted or commented", click "See who reacted or commented". The cost is 5 credits. The read covers 100 reactions and 100 comments at most.
  3. The list shows each person, their connection degree and their activity ("commented", "reacted"). People already in your database are marked.
  4. Choose a "Target". Without a Target, the imported people will not be qualified.
  5. Tick the people, then click "Import (N)".

The result shows the people imported, the duplicates merged and those above your plan's import limit. The text of the comments is not kept. Companies and authors without a profile are left out.

A reaction shows interest in a piece of content. It does not prove a business need. Qualify these people against your Target before you contact them. See Targets and qualification.

7. Edit, archive, delete a watch

Available

  • "Edit" changes the topic, the keywords, the sources and the Target. The changes apply from the next run. Saving does not relaunch collection.
  • "Archive" hides the watch from the screen and from the assistant. Its trends and reports remain. A run in progress stops.
  • "Archived (N)" lists the archived watches. "Restore" makes a watch active again, without relaunching collection.
  • "Delete" permanently erases an archived watch, with its runs, trends and reports. Posts already drafted from its trends stay in "Publish".

8. With the assistant

Available

The assistant's "Intelligence" tab creates and edits watches, relaunches a run, lists the trends and explains them. Examples:

Follow LinkedIn posts that mention “generative AI” or “customer service”.
Watch AI trends in B2B.
Show me my trends.
Explain the first trend.
Add posts from the company Example Cloud to my watch “Sovereign cloud”.

To follow an author, the assistant searches for their name on LinkedIn (5 credits) and shows you the results. You choose the right one.

The assistant does not see archived watches.

9. What Intelligence costs

ActionCost
Read RSS sources0 credits
Read the LinkedIn sources of a run5 credits per run (only if LinkedIn answers)
Report for a run20 credits (only if signals are found)
Daily run25 credits per day at most with a LinkedIn source, 20 without
Search for a LinkedIn author5 credits
See who reacted to or commented on a post5 credits
Import these people0 credits (qualification afterwards: 2 credits per person and per Target)
Draft a post from a trend15 credits

One credit is worth one euro cent. See Credits and subscription.

10. Hiring signals

Next release

The "Hiring signals" tab of the Intelligence screen groups the collected job ads by company. It shows the companies hiring on a topic your offers serve, over the last 30 days.

For each company:

  • the score of the best job ad, out of 10;
  • the matching offer (My account → Offers);
  • "N job ads in 30 days" and the number of signals;
  • the title of the best job ad and the reason for its score;
  • "See the job ads": the list of ads, with their score, their contract type and a link that opens in a new tab.

Companies are sorted by best score, then by number of signals. By default, only companies scoring 3/10 or more are shown. Click "Score ≥ 3/10" to see them all. IT services firms, agencies, software vendors and job boards never appear. With several offers, the chips at the top filter by offer.

Find the decision-maker. This button opens the Searches tab of Prospects, pre-filled: a people search and the company. When you click "Search", Maya runs 3, 4 or 5 LinkedIn searches, one after the other, one per set of titles:

  • A: "systèmes d'information" OR DSI OR CTO;
  • B: infrastructure OR cloud OR production;
  • C, depending on the signal's offer: cloud offer → architecte OR "head of cloud" OR "cloud privé"; AI offer → IA OR "intelligence artificielle" OR "data & AI"; RUN offer → exploitation OR SRE OR support. If the offer's theme is not recognised (from its name, then its keywords), there is no search C;
  • D, for every offer: CIO OR "Head of IT" OR "Director of Data";
  • E, for the AI offer only, after D: innovation OR "transformation digitale" OR "chief digital".

An AI offer therefore runs 5 searches (A, B, C, D, E), a cloud or RUN offer 4 (A, B, C, D), and an offer whose theme is not recognised 3 (A, B, D). Each result row shows the letter of the search that found it.

Each set has at most 3 titles: LinkedIn returns no result beyond that in a classic search. Nothing runs before your click. On the first click, a window shows the cost before any call to LinkedIn: the number of searches and their price (0 credits today), the maximum cost of the scoring that follows the import (2 credits per imported prospect, bounded by the import cap), and the total. Click "Run the search" to continue. The screen shows "Search 2/4…" while the searches run.

Results are merged without duplicates, those of search A first. A small letter (A, B, C or D) next to the name shows which search found the person. A banner shows how many people each search added, and how many are already in your base. The import line separates the two reasons a person is not imported: "N already in your base" for people you already have, "M left out (plan cap)" for people above your plan's import cap. If LinkedIn or your search budget refuses a search (daily limit, cap), Maya stops there, keeps what was found and shows the reason. "Load more" is not offered on these merged results. If you change the company or the title in "Advanced filters", "Search" runs a single search, with "Load more".

The import Target comes from the offer. Each offer can have a Target (Workspace → Offers, "Target" field). The "Find the decision-maker" link carries the offer of the company's best job ad. Imported people are scored against that offer's Target, preselected in "My base + Target". If the offer has no Target, or its Target is archived, the import asks you to choose a Target. It never uses your default Target without your choice. What follows is the same as any search: results, import, qualification. See Prospects.

Project summary and account context. When Maya has read the description of a job ad kept as a signal, it writes a summary of the project: the technologies mentioned, the stakes (technical debt, business continuity, sovereignty…) and one sentence describing the project, without copying the ad. The company card shows this summary. If a job ad has been read but has no summary yet, the "Prepare the summary" button creates it (no LinkedIn call, 0 credit during the measurement period). People imported with "Find the decision-maker" are linked to the job ad. For them, the campaign writer receives an "Account context" block: the summary, the name of your offer, the company name and the fact that the company is hiring on this subject. The message mentions the project indirectly ("I see your team is strengthening…"); it never says it saw a job posting. The person's record shows this context in the "Account context" section.

Not relevant. This button removes the company from the list. Pick a reason (IT services firm or agency, no real project, outside our target, already a customer or already in talks, other) and add details if needed. The reasons are used to measure detection quality. "Show dismissed" brings them back.

When the list is empty, the screen says why:

  • no offer: describe your offers in Workspace → Offers;
  • no job ad watch: create a watch with the "LinkedIn job ads" source, then arm the daily run;
  • no signal yet: the screen gives the date of the next daily run.

The assistant's "Intelligence" tab can also list these companies:

Which companies are hiring on my offers this month?

Viewing the signals and dismissing a company cost nothing.