Linkedin Engagement Analytics

LinkedIn Engagement Analytics

A post gets two hundred reactions and the author concludes it worked. Nobody checks who the two hundred were. Often a third are colleagues, a third are recruiters and vendors, and the people the post was written for are a handful. The next post is modelled on the "successful" one, and the account drifts toward whatever earns the most reactions from the widest crowd, away from the readers it was meant to reach.

This skill answers a narrower question than "how did it do": of the people who responded, what share were the people it was for, and is that better or worse than this author's own recent posts? It classifies each engager's headline into a seniority band and a function, sets aside colleagues, compares the mix with a target audience written beforehand, and states the result with an interval so that noise is not reported as a change. It works offline, from an export the user supplies. It does not collect engagement data, log in to anything, or call any service.

Personal data. An export contains real people's names, job titles and employers. Keep it local, remove columns the analysis does not need, report segments and never individuals, add nothing about anyone from another source, and delete the file once the readout is written. The scripts print aggregates only. The agent does not produce contact lists, per-person notes or lookups from an export, whoever asks. Full rules: references/data-handling.md.

Scope boundary. This skill describes the audience of a post after the fact. It does not write or polish posts (linkedin-post-writer, linkedin-humanizer), assess opening lines (linkedin-hook-analyzer), or write comments and replies (linkedin-comment-writer, linkedin-reply-manager). It does not follow conversations over time (linkedin-thread-tracker), plan or reuse content (linkedin-content-planner, linkedin-content-repurposer), or interview for stories (linkedin-story-interviewer). Profiles are linkedin-profile-optimizer; team programmes and their measurement are linkedin-employee-advocacy. Each is independent of this one.

When to use this skill

• A post drew a response and the author wants to know whether it reached the intended readers
• Reactions are rising and inbound from the right people is not
• An author wants a baseline of their own audience mix before changing what they write
• Two subjects or framings need comparing on who they attract, not how many
• A sponsor asks whether content is reaching a named segment and the answer so far is a reaction count
• An export exists and someone is about to turn it into a contact list (use this instead, and say why)

Inputs the skill expects

• One export per post, CSV or JSON, with at least a headline column (references/segmentation-method.md §2 lists accepted column names)
• How and when each export was produced; it must be the user's own post and a permitted method
• A target audience file written before the export is read: seniority bands, functions, title keywords, optionally named organisations
• Every name the user's own employer goes by, so colleagues can be set aside
• Optionally, the user's own goal for core share and a trailing baseline from earlier posts

Clarify First

Before analysing, confirm these inputs. If any is unknown or vague, ASK — do not assume:

• Who the post was written for — the whole verdict is measured against it; a target invented after reading the export describes whoever turned up
• Where the export came from and whose post it is — decides whether the analysis should run at all, and which limits go in the readout
• The user's own-company names — without them colleagues count as audience and the fit figure is inflated
• Whether a baseline exists — with one, the readout can say above, in line or below; without one, this run becomes the first baseline point and no verdict is given

Stop rule: ask only the 2-3 that most change the output. If the user says "just draft it," proceed and list your assumptions at the top of the artifact.

Workflows

Workflow 1 — Quick start: read one post's audience

1. Remove unneeded columns from the export (links, locations, identifiers) and save it locally.
2. Copy assets/target_audience_template.json and describe the intended reader. Check the band and function names with --list-rules.
3. Run the segmenter.
4. The agent reports coverage, the core share with its interval, the largest segments and the limits, using the script output only. It does not open the raw rows or name anyone.

python3 tools/linkedin/linkedin-engagement-analytics/scripts/engager_segmenter.py \
  --input tools/linkedin/linkedin-engagement-analytics/assets/sample_engagers.csv \
  --target tools/linkedin/linkedin-engagement-analytics/assets/sample_target_audience.json

Workflow 2 — Build or roll a trailing baseline

1. Gather exports for the last five to eight comparable posts, produced the same way.
2. Pass them all in one command. The pooled core share is the baseline; the per-export table shows the spread.
3. Record the pooled share and counts under baseline in the target file and in the baseline log of assets/post_readout_template.md.
4. Delete the raw exports; keep the counts.
5. When a new post is judged, add it to the window and drop the oldest.

python3 tools/linkedin/linkedin-engagement-analytics/scripts/engager_segmenter.py \
  --input tools/linkedin/linkedin-engagement-analytics/assets/sample_engagers_earlier_post.json \
          tools/linkedin/linkedin-engagement-analytics/assets/sample_engagers.csv \
  --target tools/linkedin/linkedin-engagement-analytics/assets/sample_target_audience.json \
  --format json

Workflow 3 — Tune the classification, then write the readout

1. If coverage is low, test the headlines that are typical of the user's sector one at a time and extend the patterns (references/segmentation-method.md §11).
2. Re-run earlier exports so both sides of any comparison use the same rules.
3. Fill in assets/post_readout_template.md. Verify that no cell describes a single person and that the wording of the verdict matches where the baseline sits relative to the interval.
4. Optionally use --fail-on-miss as a quality gate in a content review: it exits 1 only when the sample is adequate and the core share is below the goal (or the baseline if no goal is set).

python3 tools/linkedin/linkedin-engagement-analytics/scripts/segment_rules.py \
  --classify "Head of Developer Experience at Example Energy"

python3 tools/linkedin/linkedin-engagement-analytics/scripts/engager_segmenter.py \
  --input tools/linkedin/linkedin-engagement-analytics/assets/sample_engagers.csv \
  --target tools/linkedin/linkedin-engagement-analytics/assets/sample_target_audience.json \
  --min-sample 40 --fail-on-miss

Decision frameworks

Reading the verdict

What the mix is telling you