I built TWL to watch the people and companies a market already pays attention to. The uncomfortable bit was discovering that collecting the engagement is not the hard part. Deciding what an engagement actually means is.

A competitor's LinkedIn post can attract customers, future buyers, job candidates, peers, friends, bots, vendors and people who simply liked the photograph. Put all of them into a sales sequence and you have not created signal-based prospecting. You have created a colder list with a more elaborate origin story.

The useful approach is to treat competitor engagement as evidence of attention, then qualify it with source context, ICP fit, recency and repeated behaviour. That is the principle behind TWL Signals. This guide explains the complete workflow, from a manual test to an automated system.

The short answer#

To track competitor LinkedIn engagement well:

  1. Choose a small set of competitor, creator and category sources your buyers genuinely follow.
  2. Record engagement only from content you are permitted to access, preserving the post, action and timestamp.
  3. Enrich each person with role, company and firmographic context.
  4. Score fit separately from signal strength.
  5. Look for repeated or stacked signals rather than treating one reaction as intent.
  6. Route only qualified people into review or outreach.
  7. Use the signal to prioritise and research the conversation—not to write a creepy opening line.

The distinction matters: a person can be a perfect fit with no current intent, or show strong category interest while being unable to buy. Your system needs to represent both.

Why competitor engagement is useful—and easy to misuse#

LinkedIn itself treats activity and engagement as part of its buyer-intent picture. Its Sales Navigator guidance recommends combining alerts, profile activity and account-level intent to identify moments when outreach may be more relevant (LinkedIn Sales Navigator).

Competitor engagement can be useful because it gives you three pieces of context a static database does not:

  • Topic: the subject that earned the person's attention.
  • Source: the company, creator or peer they chose to engage with.
  • Timing: when that attention happened.

None of those proves the person wants your product. But they can give you a better reason to investigate now rather than three months from now.

The failure mode is calling every visible action a “buying signal”. A like is often lightweight attention. A detailed comment describing an operational problem is materially different. Three relevant interactions across two weeks are different again.

The signal does not replace qualification. It tells you where to spend qualification effort.

Use a signal ladder, not a yes/no label#

A flat engaged = true field throws away most of the useful information. I prefer a ladder that keeps the original action visible and increases confidence only when more evidence appears.

Level Example What it reasonably suggests Default action
1. Exposure Viewed or followed a relevant source Possible awareness Store only if the source permits it
2. Lightweight engagement Reacted to one relevant post Topic attention Enrich only when ICP fit is likely
3. Expressed engagement Added a substantive comment or asked a question Active interest in the discussion Review the comment and person
4. Repeated pattern Engaged with several relevant posts or sources Sustained category attention Prioritise for research
5. Stacked signal Relevant engagement plus job change, hiring, website activity or a stated pain A plausible timing event Route for human review or appropriate outreach

This is not a universal scoring model. The relative value depends on what you sell. A recruiter may care about a job-change signal. A data platform may care about a RevOps leader discussing broken enrichment. A local service provider will have an entirely different signal map.

The useful question is: what observable event has historically appeared near the moment someone needed what we sell? Start there.

Step 1: Choose sources with a reason#

Do not begin with every competitor you can name. Begin with the attention graph around your buyer.

Create three source groups:

Direct competitors#

These are products a buyer might compare with yours. Their product launches, customer stories, pricing changes and integration announcements can reveal people paying attention to the category.

Category creators#

These are practitioners and educators whose audience contains your buyers. For TWL, that may include people writing about outbound, RevOps, social selling or GTM systems. They can produce cleaner problem-level signals than a vendor post.

Problem sources#

These are companies, communities or individuals that regularly discuss the operational problem you solve, even if they sell something else. A strong problem source may be more useful than a famous competitor with a broad audience.

For each source, record the buyer hypothesis: “Heads of growth at small B2B companies follow this person for practical outbound systems.” If you cannot write a credible hypothesis, the source is probably noise.

Start with five to ten sources. A smaller monitored set teaches you what a useful signal looks like before automation multiplies your mistakes.

Step 2: Preserve the event, not just the person#

The original event is the reason this lead exists. Keep it.

At minimum, store:

  • source profile or company;
  • post URL and publication date;
  • post topic or a short classification;
  • engagement type;
  • engagement timestamp where available;
  • public profile URL;
  • capture time;
  • access or provenance notes;
  • the raw evidence required to audit the classification.

Do not reduce the record to a name and email. Six weeks later, nobody will remember why the person entered the pipeline, and your outreach will collapse back into generic personalisation.

There is also an important access boundary. LinkedIn says post visibility depends on the author's choice: posts can be visible to anyone, connections only or a group (LinkedIn Help). Respect that boundary and do not treat restricted content as public data.

Step 3: Enrich after capture—but before routing#

The engagement tells you what happened. Enrichment tells you whether the person belongs in your market.

Useful enrichment fields commonly include:

  • current role and seniority;
  • company and domain;
  • industry, employee range and geography;
  • functional ownership;
  • existing CRM owner or account status;
  • verified contact route, where you have a valid reason and permission to use it.

Keep enrichment separate from the source evidence. Providers change, job titles go stale and company matching can be wrong. You should be able to refresh an enriched profile without rewriting the historical signal.

This separation also controls cost. A raw post may contain hundreds of engagers. Run cheap filters first—location, title fragments, known exclusions—then pay for deeper enrichment only on plausible matches.

Step 4: Score fit and signal independently#

One composite score looks convenient but hides why a person was prioritised. Use at least two dimensions.

ICP fit#

Fit asks whether the person and company resemble someone you can help. Inputs might include role, authority, industry, size, geography, stack and known exclusions.

Signal strength#

Signal asks whether the observed activity is relevant, recent and meaningful. Inputs might include engagement type, source quality, topic relevance, repetition, recency and additional events.

A simple routing matrix works:

Weak signal Strong signal
Low fit Ignore or retain for aggregate research Review only if the signal reveals a new segment
High fit Add to a watchlist or normal account plan Prioritise for research and timely action

Add a third dimension—confidence—when enrichment or classification is probabilistic. A high score built on an uncertain company match should not become an automatic message.

Step 5: Look for patterns and stacked signals#

Single events are easy to collect and easy to overvalue. Patterns are harder and more useful.

Examples of stronger combinations include:

  • a relevant comment followed by a profile visit;
  • engagement with several category sources in a short period;
  • a new role plus repeated interest in a problem your product solves;
  • a company hiring for the function that owns your category;
  • several people from one account engaging with related content;
  • competitor engagement combined with an existing first-party website or email signal.

Do not assume every combination indicates intent. Use historical outcomes to learn which patterns predict a useful conversation for your business.

Recency belongs in the model too. A signal should decay. The right decay curve depends on the event: an operational question may matter for days, while a leadership change may remain relevant for months.

Step 6: Route to the right next action#

The strongest automation is often a routing decision, not an automatic message.

Useful destinations include:

  • ignore: irrelevant or excluded;
  • watch: good fit, insufficient timing evidence;
  • research: promising but needs context;
  • account alert: several people at one company show relevant activity;
  • human review: strong fit and signal, but judgement is required;
  • approved outreach: the reason, contact route and message have passed your rules.

Every routed record should show why it moved: high ICP fit + substantive comment + 2 related interactions in 14 days is auditable. AI score: 87 is not.

Step 7: Use the signal without sounding like surveillance#

“I saw you liked our competitor's post” is technically personalised and socially awkward.

The signal should usually help you decide who to research and when to contact, rather than become the first sentence. Read the underlying discussion. Understand the person's role. Find the business problem behind the content. Then decide whether you have a legitimate, useful reason to start a conversation.

Good outreach is contextual without narrating your data collection. It might refer to the operational topic, a public comment the person deliberately made, or a relevant change at their company. It should not exaggerate what the signal proves.

If you would feel uncomfortable explaining how the person entered your workflow, the workflow needs another review.

A manual workflow you can test this week#

Do this manually before buying or building an automated stack:

  1. Pick three high-confidence sources.
  2. Review their relevant public posts from the last seven days.
  3. Capture no more than 30 plausible people with the original post context.
  4. Enrich role and company manually.
  5. Score fit and signal on separate three-point scales.
  6. Review the top five and write down the next appropriate action.
  7. Repeat for two weeks and compare which sources produced genuinely useful prospects.

A spreadsheet is enough for this test. Suggested columns are:

source, post_url, topic, engagement_type, profile_url, role, company, fit_score, signal_score, confidence, next_action, reason, captured_at.

The goal is not to generate outreach volume. It is to learn whether this attention graph contains your market and which behaviours deserve operational weight.

What an automated system should actually do#

Once the manual process produces a repeatable decision, continuous workflows should remove collection and coordination work while leaving important judgement visible.

A robust workflow looks like this:

  1. Monitor approved public sources on a schedule.
  2. Capture new posts and engagement with stable deduplication keys.
  3. Classify post topic and engagement type.
  4. Filter obvious exclusions before paid enrichment.
  5. Enrich person and company records.
  6. Score fit, signal, confidence and recency independently.
  7. Resolve repeated activity to one person and one account timeline.
  8. Route qualified records according to explicit thresholds.
  9. Request review before consequential outreach where required.
  10. Record outcomes so the scoring model can be challenged and improved.

Retries, duplicate events, missing enrichment and provider failure are normal states. Design for them. “The workflow ran” is not the same as “a qualified prospect reached the correct owner exactly once.”

Account safety, permissions and privacy are product requirements#

Public visibility is not blanket permission for any collection or use. LinkedIn's User Agreement prohibits unauthorised scraping and automated methods, including browser plugins or scripts used to copy the service (LinkedIn User Agreement). Applicable privacy, direct-marketing and anti-spam rules also depend on your location, data source, purpose and outreach channel.

Build the access decision into the system:

  • use permitted data sources and documented provider rights;
  • do not bypass access controls or automate a personal LinkedIn session;
  • retain provenance and the purpose for each record;
  • minimise collected fields;
  • provide suppression and deletion paths;
  • review channel-specific outreach rules before sending;
  • keep a human approval boundary where the context is sensitive or uncertain.

This is an operating principle, not legal advice. Get appropriate advice for the markets and channels you use.

Common mistakes#

Tracking too many sources#

More sources create more records, not necessarily more signal. Expand only after a source proves it contains your ICP.

Enriching everyone#

Apply cheap relevance filters first. Otherwise the provider bill grows faster than the qualified pipeline.

Treating comments and reactions equally#

Preserve engagement type and content. A substantive question usually contains more evidence than a lightweight reaction.

Hiding decisions inside one AI score#

Store the contributing facts and thresholds. A reviewer should be able to understand and override the route.

Automating outreach before learning the motion#

First prove that the sources, signal ladder and message are useful manually. Automation makes a weak process consistently weak.

Measuring replies without measuring quality#

Track qualified conversations, opportunities and source-level yield. A curious reply is not the same as commercial relevance.

The metrics that help you improve the system#

Measure the funnel by source and signal type:

  • captured people per source;
  • percentage that match minimum ICP criteria;
  • enrichment success and cost per qualified profile;
  • strong-signal rate;
  • time from signal to review;
  • qualified conversations per reviewed prospect;
  • opportunity rate by source and signal pattern;
  • false-positive and suppression rates;
  • manual review time;
  • duplicate and workflow-failure rates.

The first useful question is often source yield: which monitored sources produce people who both fit and have a credible reason to care? Remove weak sources before tuning a sophisticated scoring model.

Build a filter, not another lead list#

Competitor LinkedIn engagement can be a valuable prospecting input. It is not intent by default.

The system becomes useful when it keeps the original event, separates fit from behaviour, recognises patterns, respects access boundaries and routes only the people who deserve attention. That is a quieter promise than “turn every like into pipeline”. It is also much closer to how a reliable sales system should work.

Start manually. Learn which sources and patterns matter. Then automate the parts that repeat without making the judgement disappear.

Keep exploring

TWL Signals

Turn market attention into a qualified prospect feed.

Monitor the people and companies your buyers already follow, keep the original context, enrich the profiles and route only the strongest matches into outreach.

Start building your signal workflow