Account warmup — manual engagement and the behavioral baseline.

The cold-start asymmetry

On day zero, the detection model has exactly one prior on the account: it is new. New accounts are not treated as neutral — they are treated as untrusted, because the empirical base rate of fresh accounts that engage in outreach-shaped behavior within the first 72 hours is dominated by automation, ban-evasion, and scraping. Every signal the model evaluates is interpreted against that prior.

Sessions during the first weeks of an account's life carry disproportionate weight in the construction of the behavioral baseline. Inter-action timing distributions, click-target heuristics, scroll patterns, navigation graphs, login times, session duration — the values observed in week one establish the model's expectation, and the values observed in subsequent weeks are scored relative to that expectation. Compressing the first session into a thirty-minute burst of profile views and outbound connects is not merely high-volume behavior; it permanently anchors the account's behavioral baseline as automation-shaped, with no documented mechanism for re-baselining.

The asymmetry compounds: early sessions are both when the baseline is set and when the operator has the lowest tolerance for restriction, because there is no accumulated value in the account to insure against the loss. The combination produces the empirically observed 60% permanent-restriction rate among accounts that receive a first-tier automation flag in their first thirty days. Operators who skip warmup are not gambling on whether they get caught; they are gambling on the recovery path, and the recovery path on LinkedIn is the worst in any major outbound channel.

What warmup is constructing

Warmup is not a delay imposed for its own sake. It is the deliberate accumulation of four distinct signal categories, each of which the detection model evaluates independently and each of which a cold account scores zero on:

Profile depth

Recommendations from existing professional connections, posted content, group memberships, work-history detail, education detail, skill endorsements. A profile at 90% completeness with no recommendations, no posts, no groups, and no inbound engagement is — from the model's perspective — a profile that exists on paper without existing in the social fabric. The All-Star completeness threshold (the published LinkedIn UI signal) is necessary but not sufficient. The threshold the detection model evaluates is denser than the public completeness widget suggests.

Connection-graph density

1st-degree connections, with particular weight on inbound connection acceptances from a coherent professional cluster (former colleagues, alumni, industry peers). A connection graph constructed primarily through outbound requests to 2nd- and 3rd-degree network is structurally distinguishable from a graph constructed primarily through inbound acceptances and reciprocal connections. The 500-connection All-Star threshold is the floor, not the goal.

Engagement signal

Likes, comments, reactions, post views, dwell time on connection content, profile views of relevant people, search activity consistent with the profile's claimed industry. Engagement is the signal the model uses to distinguish an active human from a dormant account being rented for automation, and the engagement-to-action ratio is one of the more robust per-account features in the published behavioral-classifier literature.

Consistent session patterns

Login times that cluster within a coherent timezone, session durations that follow a human distribution, action types that mix browsing, messaging, content engagement, and search. A session pattern that is consistent week-over-week is itself a signal of legitimate use, and the construction of that consistency is the secondary purpose of the warmup runway.

The ramp curve

Production warmup is a 2-4 week runway, with action volume escalating across roughly four phases. The volumes below are observed operational values for accounts being prepared for sustained outbound, not strict regulatory limits — they describe the curve the detection model has been observed to treat as normal account onboarding behavior:

An "action" in this table is any user-initiated interaction the model logs: a profile view, a connection request, a message, a like, a comment, a search query, a content post. Compressing the curve — week 1 at 15 actions/day, week 2 at 40, production volume by day ten — is the modal operator failure and the modal cause of first-tier flags within the first 30 days of an account's life.

Profile completion as the first task

The first operational task during week one is not engagement; it is profile construction. A profile photo (professional headshot, consistent with the claimed identity), a headline that aligns with the work history, a complete work history with descriptions on each role, education with detail, an "About" section with substantive text, listed skills, and group memberships in relevant professional groups.

The All-Star profile-completeness indicator surfaced in the LinkedIn UI is a public floor. The denser threshold — recommendations from prior colleagues, endorsements clustered around the claimed competencies, group memberships consistent with the industry claim — is what the detection model evaluates. An operator standing up an account should secure 3–5 recommendations from existing 1st-degree connections during week one, both because they signal profile authenticity and because soliciting them produces the kind of inbound engagement (notifications, replies, profile views from the requested connections) that the model interprets as embeddedness in a real professional graph.

Content engagement during warmup

From week one onward, the account engages with existing connections' content — likes, thoughtful comments, occasional shares. The inter-action timing distribution matters: a session that produces 20 likes in 90 seconds is not credible; a session that produces 4 likes and 2 comments over 18 minutes of mixed browsing is. The empirical engagement distribution among active LinkedIn users skews heavily toward consumption (scrolling, reading) over interaction; warmup sessions should mirror that ratio, with roughly 3–5 minutes of feed scrolling per interaction event.

Comments specifically carry higher weight than likes — they produce inbound notifications, they generate replies, they create the kind of conversational threading that the model interprets as legitimate professional engagement. A warmup session that includes two substantive comments on connection posts produces materially more signal than the same session with twenty silent likes.

First-content publication

During the 4-week warmup, the account publishes at minimum 3–5 posts. The content is professional, aligned with the profile's claimed role and industry, and substantive enough to attract engagement from the existing connection graph. The point is not content marketing — the point is the post-engagement signal the model records: views from connections, reactions, comments, and the resulting profile views from interested parties.

A posted piece of content that attracts 30 views, 4 reactions, and 2 comments from existing connections produces a substantially richer signal than any volume of outbound engagement, because it is signal generated by other users acting on the account's profile. Inbound engagement of this shape is the hardest signal for an automated warmup service to fabricate and is consequently among the highest-weighted features in the model's evaluation.

Connection-graph density

Week-two connection requests target the operator's existing professional network — former colleagues, university alumni, prior coworkers, industry peers met at conferences. These requests carry high context (shared workplace, shared school, shared connections) and produce high acceptance rates, which in turn produce high inbound signal: notifications, profile views from accepting parties, reciprocal connection requests from people in the same cluster.

The connection requests sent during weeks two and three should largely not be the cold outbound requests that production cadence eventually serves. They should be requests where the acceptance rate is expected to exceed 70%, with personalized notes that reflect actual prior context. The purpose is to construct a 1st-degree connection graph that is coherent, cluster-dense, and structurally indistinguishable from the graph an early-career user organically builds in their first months on the platform.

The synthetic-warmup detection problem

A historical category of products attempted to automate the warmup process — automated profile views, automated likes, automated comment generation, automated connection acceptance — on the premise that the model would not distinguish synthetic engagement from human engagement. That category has been substantially deprecated by detection improvements over the 2022-2025 period.

The detection signal against synthetic warmup is, by observation, dense: the engagement is too uniform across session times, the comments are statistically detectable as model-generated, the connection-acceptance patterns lack the cluster structure of organic graph growth, and the inter-account behavioral correlation across a pool of synthetically-warmed accounts produces a fingerprint that is itself detectable. Operators who outsource warmup to an automated service in 2026 are not warming up the account; they are constructing a flagged behavioral baseline more efficiently than they could do by hand.

The present-day operational reality is that warmup must be human-driven. There is no shortcut category that the detection model has not already learned to identify, and the categories that briefly worked have, by the time they are described in any public guide, been deprecated by the next classifier iteration.

Pattern-of-life consistency

The behavioral baseline the model constructs in weeks one through four is the baseline it scores production-phase behavior against. A warmup pattern that establishes weekday logins between 8 and 10 a.m. Eastern, sessions of 12-20 minutes, and a mix of engagement and browsing — followed at week five by overnight logins, 90-minute sessions, and pure outbound-connect activity — is itself a strong signal. The change-in-pattern itself flags the account, independent of the absolute volume of the new pattern.

The implication is that warmup behavior cannot be fully abandoned after launch. Login times, session durations, action-type mix, and engagement-to-outbound ratios should remain broadly consistent week-over-week. A sustained outbound cadence at production volume must coexist with continued engagement, continued browsing, and a session shape that resembles the shape established during the warmup runway.

Account-team vs founder warmup

When warmup is conducted by an SDR team on the account owner's behalf — common in agency operations and in distributed sales teams — the consistency requirement becomes operationally harder. Different operators have different typing cadences, different mouse-movement profiles, different timezones, and different click patterns. A warmup runway conducted by three rotating operators on the same account produces a behavioral fingerprint that is structurally distinguishable from a single-operator account, and the difference is, by observation, detectable.

The operational pattern that minimizes this signal: the account owner personally conducts the first two weeks of warmup, establishing the baseline under a single behavioral profile. Handoff to the SDR team occurs only after the baseline is set, and the handoff itself happens on the same IP residency, the same browser fingerprint, and the same device profile — the substrate carries the continuity that the human inputs cannot.

The cost

A 4-week warmup at 30 minutes per day of human time is 14 hours per account. For the 5-account multi-account architecture described in Chapter 01, the total is 70 hours of human labor before the first cold connect request is sent. The labor is not delegate-able to an automated tool, is not parallelizable across accounts (each account requires its own session-by-session human input), and is not recoverable if the account is subsequently restricted.

Operators who calculate the per-account cost of warmup and conclude that it is "not worth it" are calculating against the wrong denominator. The correct denominator weighs the labor cost against the empirical 60% permanent-restriction rate on automation-flagged unwarmed accounts and the operational cost of replacing a restricted account. The expected cost of skipping warmup exceeds the expected cost of conducting it by an order of magnitude.

Production-phase warmup maintenance

After the formal 4-week runway, the account does not stop performing warmup-shape behavior; it continues at a reduced cadence indefinitely. The operational target during production is 5–10 daily warmup-type actions — likes, comments, organic content engagement, occasional content publication — running alongside the outbound cadence.

The purpose is to prevent the behavioral signature from collapsing into pure-outbound mode. An account whose only logged activity is sequential connect requests and templated messages produces a different signature than an account where outbound activity is embedded in a continuous stream of human-shape engagement, and the difference is, by observation, the difference between an account that survives the year and an account that is restricted at month four.

Common operator failures observed in production

• Skipping warmup entirely on accounts that look professional. The operator stands up an account with a strong headline, a real photo, complete work history, and concludes that warmup is for less credible profiles. The detection model does not evaluate visual credibility; it evaluates behavioral baseline. A professional-looking profile with no behavioral history is a flagged account waiting to happen.
• Treating warmup as content marketing. The operator publishes 4 posts during week two, attracts no engagement (because the connection graph is too small to produce it), and considers the warmup complete. Content publication without the accompanying engagement signal — likes, comments, profile views, reactions from connections — is half the signal the model is looking for, and the operator has constructed only the publishing half.
• Running an automated warmup service. The category exists, the providers market themselves as undetectable, and the detection model has, by observation, already learned each public category's signature. The operator pays for warmup, the warmup produces a flagged baseline, and the account is restricted in week six with no recoverable signal.
• Abandoning warmup behavior after launch. The account completes the 4-week runway, transitions to production cadence, and immediately drops likes, comments, and content publication. The behavioral signature shifts from "active human professional" to "pure outbound machine" within the first week of production, and the account is flagged within the first month — not because the production volume is excessive, but because the change-in-pattern itself is the signal.
• Rotating multiple operators through a single account during warmup. Each operator establishes a partial baseline, the composite baseline is inconsistent, and the account's behavioral fingerprint is structurally distinguishable from a single-user account.
• Conducting warmup on a different IP residency than production. The behavioral baseline is established under residential IP A, the account transitions to residential IP B for production, and the IP-shift signal compounds with the volume-shift signal in week five. Chapter 03 covers the proxy-continuity requirement; warmup is the period when that continuity is most consequential.

Pre-deployment checklist

• Profile photo, headline, and "About" section finalized before week-one activity begins
• Work history complete with substantive descriptions on each role
• 3–5 recommendation requests sent to existing 1st-degree connections in week one
• Group memberships in 5–10 relevant professional groups by end of week two
• Inbound connection requests from the existing professional network solicited via direct outreach off-platform during week one
• At least 3–5 substantive content posts published over the 4-week runway
• Login schedule planned around a single coherent timezone, with session times that match the claimed work schedule
• IP residency and browser fingerprint identical between warmup and production (Chapter 03)
• Production-phase warmup maintenance cadence (5–10 daily engagement actions) explicitly scheduled into the operator's SDR workflow
• For multi-operator accounts: account owner personally conducts weeks one and two; SDR handoff occurs no earlier than week three

Where warmup fits in the broader infrastructure

Warmup is the period during which the substrate established by Chapters 01-03 (account architecture, detection model, IP infrastructure) is translated into a behavioral identity the detection model trusts enough to permit production-volume outbound. Without warmup, the substrate is provisioned but the identity is empty, and the empty identity is treated by the model as the prior the empirical evidence supports: an automation account being constructed at speed for outbound abuse.

The chapters that follow — messaging architecture (Chapter 06), automation category selection (Chapter 07), compliance posture (Chapter 08) — all assume the account has a constructed behavioral baseline to operate against. Skipping warmup does not merely increase the risk of restriction; it invalidates every downstream operational assumption those chapters make. An operator who treats warmup as optional has, in practice, built the entire infrastructure on a substrate that the detection model has already classified as suspect, and the subsequent outbound activity is operating against a baseline that was permanently anchored at "untrusted" the day the account was provisioned.