How to build a commenting and DM factory for manual-quality outreach

A useful LinkedIn outreach workflow gives you two daily queues: posts worth commenting on and conversations worth continuing. Each item needs a reason to act and a draft grounded in the relationship.

A sequence is predictable by design. The timing, message, and next step are decided when a prospect enters the sequence. Personalized fields alone do not make a fixed sequence responsive to changes in the prospect’s situation or yours.

Manual outreach keeps the judgment and context. It can feel natural because a person decides what makes sense now. In practice, it is usually more art than system. It depends on a founder or sales rep's motivation, and it breaks on the work around the work: find the prospect in HubSpot, locate them in Sales Navigator, decide what to say, send the interaction, and log the activity. As soon as other work takes priority, long-running nurture is usually the first thing to stop.

The useful middle ground has to do three things at once:

1. Keep the quality close to manual outreach.
2. Keep the interaction window open for as long as the relationship needs.
3. Keep the effort per prospect low enough to sustain the work.

Extrovert founder Oleg Sobolev calls this an outreach factory: one factory for comments and one for DMs. His framework combines preparation, prioritization and adaptable follow-up. Manual-quality describes the relevance and judgment you want in each interaction; you can review the work yourself or delegate selected actions.

What an outreach factory actually does

A “factory” organizes and surfaces the work for the rep. Sending can remain personally reviewed, or selected work can be delegated under explicit rules.

The raw material is everything happening around a prospect: their posts, the people they engage with, your campaign goal, previous comments, DM history, new signals, and follow-ups that are due. The factory turns that mess into a ranked workload for today.

LinkedIn gives you three basic ways to stay in touch:

• Posts make your profile and outbound more credible. Even if a post brings no inbound leads, it still gives a prospect context when they check who contacted them.
• Comments create light touches and visibility before or between direct conversations.
• DMs move the relationship into a private conversation and keep it moving over time.

This factory focuses on comments and DMs. Posts remain the credibility layer behind both.

The output is two daily feeds:

• Where should we comment today?
• Which DM conversations need attention today?

Each feed must answer the same two questions:

Factory	Two decisions to make
Comments	Which person and post deserve attention now? What useful angle can we add?
DMs	Which conversation needs attention now? What is worth saying given its history and current context?

Both have to work. If prioritization is weak, the rep still spends the day searching feeds and checking trackers. If the drafts are weak, every suggestion becomes an editing project. Either failure pushes the effort per prospect back toward fully manual work, and the factory stops being a factory.

The process is sustainable only if choosing the next interaction and preparing its draft take less work than starting from scratch.

The commenting factory: who and why now?

Prospects post unevenly. One person may publish every day, another once a month, and another not at all. A chronological feed gives the most active people most of your attention, even when they are not the most important people to engage.

A useful commenting factory needs four controls:

1. Prospects: who belongs in this engagement strategy?
2. Filters: which posts should never enter the queue?
3. Priority: what makes one opportunity more important than another?
4. One main feed: how should several engagement strategies compete for limited daily capacity?

The framework connects the prospect list, filters and campaign priority to one ranked feed. Set capacity around work you can review or supervise; it is an operating budget, not a LinkedIn-approved activity allowance.

Warm-up and slow cooking need different priorities

A warm-up campaign has a next stage. You may want three useful comment touches before sending a connection request or DM. In that case, it makes sense to prioritize prospects closest to completing the warm-up.

In Oleg's example, John needs one more comment and Mary needs three. John appears first because completing his warm-up moves him to the next stage. A comment still needs to be useful; a stage target is not a reason to post filler.

Long-term nurture works differently. A prospect who said “not now” does not need to be pushed through a short stage. Oleg calls this slow cooking. The goal is to stay familiar without stalking them, so touches should be spread across the list at a lower pace.

Oleg's example moves an account from frequent warm-up to a comment every few weeks or once a month. Those are nurture examples, not a required schedule. The campaign goal should determine the pace and sorting strategy.

See how campaign goals sort the feed

Extrovert’s prioritization illustration shows three ways to order posts: newest first for visibility, fewest comments left before outreach for warm-up, and longest since your last comment for nurture. Choose the strategy that matches the campaign; evenly distributed attention is based on time since your last comment, not a measured relationship-strength score.

A campaign is an engagement strategy

Different campaigns deserve different shares of your attention. Narrow campaigns for different ICPs, products, and value propositions also preserve more specific context than one catch-all campaign.

An active outbound campaign may be high priority. Warm leads who replied to an email or visited the website may deserve another strategy. A slow-cooking nurture campaign can run at lower intensity. A topic campaign may focus on visibility.

The main feed combines them. Campaign priority, pace, filters, and relationship stage decide what reaches the top, so a person can spend limited review time on the best opportunities.

When prospects rarely post

Some markets are full of lurkers. Buyers in healthcare, construction, and other less-active categories may read LinkedIn without publishing often.

The first choice is still to prioritize prospects who post. Their activity gives you direct context. When that is not possible, a factory can look at the people and accounts they regularly engage with, then surface posts the prospect is more likely to see.

A comment on a colleague's or familiar creator's post can put you in the prospect's orbit. Exposure is plausible, never guaranteed. This is a fallback for markets where filtering for active posters would remove most of the ICP.

Filters matter here too. Politics, sensitive topics, irrelevant posts, or content where the sender has nothing credible to add should never consume review time.

The commenting factory: what should we say?

Correct timing does not rescue a generic comment. Useful drafts combine your context and style with the actual post, then need a check for relevance, usefulness and repetition.

The draft needs four kinds of context:

• Who you are: background, expertise, company, and campaign goal.
• How you sound: tone, approved examples, writing preferences, and edits the system can learn from.
• What you believe: talking points or content pillars you can credibly bring into a conversation.
• What the post says: text, image or video context, topic, and intent.

Then the draft needs a check: Is it relevant? Is it useful? Is it repetitive? Does it sound like something this person would write?

Earlier Extrovert drafts sometimes read like AI. Better models alone did not solve it. Oleg described using five to seven model calls for some comments: one model drafts, another reviews, and another corrects violations of the user's style and Extrovert's anti-slop rules. Oleg's shorthand is “fight slop with more slop.” The extra compute is there to reduce the human editing cost.

The result is a draft to review, edit, skip, or regenerate, with the blank-page work already done.

Extrovert can build voice context through its own assistant. Oleg recommends connecting an external assistant through MCP when it already knows more about the user's business, strategy, and writing. The same shared context can support posts, comments, and DMs.

The DM factory: treat outreach and follow-up as one process

A fixed sequence usually treats a reply as the end of automation.

A prospect can say they will book a call and disappear. Someone who ignored three messages may become relevant after a job change. A quiet opportunity can reopen months later because you shipped something useful for them.

The DM factory treats first outreach, replies, follow-up, and reactivation as one continuous process. Separate starting, following up, slowing down and restarting on fresh context.

Its daily feed should answer a practical question: Which conversations need attention today?

The rep should not have to choose between checking new conversations, following up with old ones, and searching for people who promised to reply. The factory puts due conversations in one recommended view. The operating goal is inbox zero: review what needs attention, act or skip, and move on.

Review the conversations that are due

The Recommended DMs illustration shows conversations with a reason to act, a draft, and Send or Snooze controls. Use the reason and conversation history to decide whether to send, edit or set it aside. A recommendation is work to assess, not an automatic sending instruction.

Cadence should adapt to the market and the conversation

Oleg's rule depends on total addressable market and account value. The human effort you are willing to spend on a prospect should track the economics of that relationship.

If the market is large, two or three unanswered follow-ups may be enough before moving on. For a small list of valuable accounts, space follow-ups farther apart and keep the relationship open longer.

A reply resets the unanswered-follow-up count. Read an unread reply before continuing; it stays out of Recommended until you do. If the conversation then goes quiet, a new follow-up run can become due. If they never reply, the configured limit stops recommendations.

The DM factory: restart when something changes

Oleg recommends tapering pitch-led follow-up after roughly three attempts. That is his rule of thumb, not a universal reply-rate threshold.

A useful restarter comes from one of two places:

1. Something changed for the prospect.
2. Something changed for you.

On their side, that may be a launch, new role, shipped product, company development, or relevant post. On your side, it may be a new resource, webinar, event, feature, or piece of information genuinely useful to them.

A simple question such as “I saw your launch. How did it go?” can reopen a real conversation.

Timing should respond to that context. In Extrovert, a relevant post can bring an existing follow-up forward when that setting is enabled. It does not initiate a first DM, exceed the follow-up limit, override a skipped conversation or bypass an unread reply. If nothing changes, the cadence can remain slow.

In anonymized Extrovert usage data, Oleg reported three figures that explain why this matters. In the social-touch comparison he described, connection acceptance was shown as 27% → 41%, based on what he recalled was roughly 500,000 connection requests. He also reported that 52% of conversations reaching a first reply needed at least two touches. He said about one in five stalled conversations could be revived with consistent follow-up and Conversation Restarter signals. These are speaker-reported observational figures, not controlled estimates of what comments or follow-up will cause for another campaign.

The DM factory: use the whole relationship to draft

A first name and job title are not enough context for a useful DM. Combine conversation history, recent context, your selling points and your voice.

The draft should use:

• the conversation history and what has already been said;
• the prospect’s headline, About section and recent posts;
• relevant selling points, resources, objection handling, and links;
• the sender's voice and campaign rules.

Templates can tell the system when to mention a resource, how direct to be, or which question fits a situation. The message then adapts to the prospect and the conversation. Check claims against those sources: Extrovert's DM drafting does not read the prospect's full work history or company page, or conduct an open-web search.

Oleg's own example starts with a low-pressure offer to support a prospect's content. If they post, the next message can refer to a real post. If they do not, the factory follows a different rule. The system should follow the sender's actual outreach strategy and use Oleg's opener only where it fits.

Who runs the factory?

Oleg's three modes remain a useful way to decide who controls the work. Choose separately for comments, connection requests and DMs.

1. You review everything

A founder or rep opens the feed and checks each suggestion. This gives maximum control and still removes the work of finding opportunities and starting every draft from zero.

2. Automate selected parts

Comments start in review mode. To automate a chosen campaign, open its Commenting tab, set Commenting mode to Auto, and confirm the switch. Set its daily cap and posting window. The account-level comment limit still applies across manual and automatic activity.

Oleg's example is to review a campaign's output for a week before delegating it. A week is an observation period, not proof that every future draft will be suitable. Keep sensitive topics, customers and live opportunities personally reviewed.

Automatic comments wait in Pending, where you can Cancel them before posting. Switching Auto back to Manual affects new work; already scheduled comments can still post unless you cancel them in Pending.

Connection requests can be sent automatically after warm-up when enabled. DM recommendations still need approval rather than being sent merely because the cadence says they are due. Check connected-agent permissions separately from those app defaults.

3. Your AI agent reviews the feed

Through MCP, an external agent can review feeds and conversations, edit drafts, and approve or send the actions you allow. It can apply business context Extrovert may not have and bring uncertain items back for review. Do not assume connecting an assistant means every DM will still receive your personal review: define which actions it may take and which must come back to you.

The agent handles routine triage. The founder defines the strategy and stays involved where judgment matters.

The three modes can coexist. Review VIP accounts personally. Use bounded automation where the campaign has earned trust. Let an agent reduce routine triage under explicit permissions. Neither personal review nor daily caps establish LinkedIn authorization or guarantee protection from restrictions.

The autopilot settings guide covers Auto, Manual and Pending. For delegated work, see the Claude outreach workflow.

From LinkedIn factory to GTM operating system

Oleg framed Extrovert as the system his team is building. The same questions can organize email or other social workflows: who deserves attention now, why now, and what is worth saying in the current context?

He sees those principles as the basis for a broader GTM operating system. Extrovert’s product workflow is LinkedIn-specific; applying the model to email is a strategy, not a claim that Extrovert sends email or schedules posts.

One practical way to build the factory

Start with the current sales process:

1. Define the relationship goal for each campaign: warm-up, active outreach, slow cooking, visibility, or another specific job.
2. Qualify who belongs in each campaign and filter out posts you would never join. Current Lead Finder and Intent Signals can help build the list; activity is a reason to investigate, not proof of fit or permission to pitch.
3. Decide what “why now” means for that campaign.
4. Set pace and priority around real daily capacity.
5. Combine campaigns into one comment feed.
6. Add voice, expertise, approved examples, and talking points for comment drafts.
7. Bring new DMs, due follow-ups, and stalled conversations into one DM feed.
8. Set separate rules for pitch-led follow-up and context-led restarters. Respect a request to stop. Skip forever stops DM suggestions for one prospect; turn off Direct messages for a campaign when the whole group should stop receiving recommendations.
9. Give DM drafts the full conversation history and current prospect context.
10. Decide which work stays personally reviewed, which campaigns can use bounded automation, and which actions an agent may take. Check Pending when changing modes, and cancel queued work you no longer want.

A rep should be able to open two feeds, review a finite list, and know why every item is there. Remaining search work or wholesale draft rewrites show what still needs fixing.

Get the original framework slides from the replay page.

FAQ

What is MCP in this workflow?

MCP connects Extrovert to an external AI assistant. The assistant can use its broader context to configure voices and campaigns, inspect queues, edit drafts, and approve or send actions you allow.

How many unanswered DM follow-ups should you send?

Oleg's answer depends on the size of your market and the value of each account. For a large addressable market, he recommends two or three follow-ups and then moving on. For a short list of high-value accounts, he recommends much longer nurture with a slower cadence.

Should comments and DMs use the same long-term cadence?

Oleg recommends more frequent comments and less frequent DMs. His examples are a comment every few weeks or once a month, and a DM every two or three months, adjusted to list size and account value.

Can the first DM reference a prospect's profile or posts?

Yes. Oleg says DM suggestions can use the prospect's profile and recent posts, along with your voice, instructions, selling points, and adaptable templates. You can review and rewrite every suggestion.

How do you keep irrelevant people out of the commenting feed?

For outreach, Oleg recommends a people campaign built from a defined ICP list. Topic campaigns scan broader LinkedIn content and are better suited to visibility. Campaign filters, pace, and priority then control which opportunities reach the main feed.

Can different products or ICPs use separate strategies?

Yes. Oleg recommends narrower campaigns for different ICPs, products, or value propositions so each can use its own context and engagement strategy.