How Engagement Data Should Drive Sending Infrastructure Decisions

• February 2021
• Engineering Memo · External Release

Engagement data is collected primarily for marketing purposes — optimising subject lines, send times, and content based on what recipients interact with. Its value for infrastructure decisions is largely underutilised. The same engagement signals that tell a marketer which content resonates tell an email infrastructure operator which list segments are safe to send through high-reputation IP pools, which warrant isolation in lower-risk infrastructure, and which have reached the retirement threshold where continued sending creates more reputation risk than revenue opportunity.

This note documents how engagement data — specifically open rate by IP pool and ISP, click-through rate by list segment, complaint rate by acquisition source, and engagement cohort age distribution — should inform specific infrastructure decisions rather than being treated as a marketing analytics concern disconnected from the infrastructure team.

The Infrastructure Decisions That Engagement Data Should Drive

Four infrastructure decisions benefit directly from engagement data input: IP pool assignment, list segmentation for infrastructure routing, sending frequency calibration, and engagement-based list retirement thresholds. Each of these is currently made in many organisations without engagement data input — the infrastructure team sets pool assignments based on traffic type alone, and the marketing team sets frequency based on campaign schedule rather than list health signals.

IP pool assignment — which list segments route through which IP pools — is typically determined by traffic type (transactional vs promotional) and static rules. Engagement-informed pool assignment adds a quality dimension: highly engaged segments (30-day openers) route through the highest-reputation pool where ISP rate limits are most generous and inbox placement is highest. Moderate-engagement segments route through the standard promotional pool. Low-engagement segments (90+ days no open or click) route through a lower-reputation, isolated pool that absorbs the higher complaint risk from sending to partially disengaged contacts.

Sending frequency calibration — how often each segment receives campaigns — should be driven by engagement rate per send as the primary signal, with complaint rate as the constraint. A segment that consistently opens 35% of campaigns can sustain higher frequency without complaint accumulation than a segment with 8% open rate. The infrastructure implication: higher complaint rates from lower-engagement segments at higher frequencies drive up the deferral rate at complaint-sensitive ISPs, adding retry pressure on the IPs used for those sends. Reducing frequency to lower-engagement segments protects IP reputation as well as list health.

Open Rate by IP Pool: The Metric That Connects Engagement and Infrastructure

Per-ISP open rate, when segmented by sending IP pool, reveals how engagement differs across the programme's infrastructure tiers. A promotional pool with 22% open rate at Gmail and a re-engagement pool with 8% open rate at Gmail are sending to fundamentally different quality audiences. The ISP reputation implications follow from this: the 8% open rate pool is generating significantly fewer positive engagement signals per delivered message than the 22% pool, making it more dependent on clean list hygiene (low bounces, low complaints) to maintain its reputation than the high-engagement pool.

If the re-engagement pool's open rate is below 5% and complaint rate is above 0.06%, the pool is accumulating net-negative reputation signals. Each campaign from this pool generates more negative signals (low engagement, some complaints) than positive ones (clicks, opens). Over time, this produces IP reputation degradation that limits the pool's effective throughput at major ISPs — not through a single triggering event, but through the slow accumulation of below-expectation engagement signals.

The infrastructure decision this drives: when a pool's open rate drops below a threshold (typically 5–8% for consumer email) and complaint rate rises above 0.05%, the list segments in that pool need intervention before the pool itself needs remediation. Intervention options: more aggressive engagement threshold for inclusion in the pool (raise from 90-day to 60-day no-open criterion), a re-validation pass on the lowest-engagement contacts, or retirement of the sub-threshold segment entirely. These interventions improve the pool's engagement profile, which improves its reputation trajectory, which improves its effective throughput at ISPs.

Engagement-Based List Retirement Thresholds

Every active email list has a retirement threshold — the point at which continuing to send to a contact costs more in reputation damage than it generates in revenue. The retirement threshold is not a universal number; it depends on the programme's average revenue per email, the ISP's reputation sensitivity to low-engagement signals, and the complaint propensity of the specific contact population. But it can be calculated from first principles using programme-specific data.

The revenue side: expected revenue per email from a contact in the 90–120 day no-open cohort. This is typically 10–30% of the revenue per email from the 0–30 day cohort, because click-through and conversion rates decline sharply with engagement age. For many programmes, contacts who have not opened in 120 days generate revenue-per-email below €0.01 — essentially zero.

The cost side: the reputation cost of sending to this cohort. If the 120-day+ no-open cohort has a 0.12% complaint rate and the programme's overall complaint rate is 0.03%, sending to this cohort adds 4× the average complaint load for each email delivered. At scale, this incremental complaint rate contributes to reputation degradation at Yahoo and Microsoft that, at 0.12%, risks triggering their automated throttling responses. The monetised cost of this throttling — reduced throughput requiring extended delivery windows, which reduces the revenue per campaign for the programme as a whole — typically exceeds the direct revenue from the 120-day cohort by a significant margin.

The retirement threshold calculation: when the revenue per email from a cohort falls below the estimated reputation cost per email from sending to that cohort, retire the cohort. This calculation should be performed quarterly for each engagement age cohort (0–30 days, 31–60, 61–90, 91–120, 120+) and should drive a rolling retirement policy that automatically segments out contacts who cross the retirement threshold into either a long-term re-engagement track (attempting 1–3 final sends with significantly different content before full retirement) or permanent suppression.

Building the Data Pipeline: Connecting Engagement Metrics to Infrastructure Configuration

The engagement-to-infrastructure connection requires a data pipeline that flows both directions: engagement metrics from the sending platform (MailWizz) into the infrastructure configuration decisions (PowerMTA pool assignments, campaign suppression criteria), and infrastructure performance data (per-IP delivery rates, deferral rates) back into the engagement context that explains why certain segments are performing differently than others.

The MailWizz database contains the engagement history needed for cohort classification — last open date, last click date, total opens and clicks per contact, FBL complaint status. Querying this data to classify each contact by engagement cohort requires periodic (daily or campaign-triggered) SQL queries against the contact/subscriber tables. The cohort classification output feeds directly into the list segmentation rules that determine which contacts are included in each campaign's recipient list for each IP pool.

This loop — engagement data informing pool assignment, pool assignment affecting delivery quality, delivery quality influencing engagement signals — is the foundation of a self-improving email programme. Programmes that implement it consistently see reputation stability improve over time as the most problematic engagement cohorts are progressively isolated, reduced in frequency, and ultimately retired, leaving the active sending programme with a higher average engagement rate that generates more positive ISP signals per send.

Open Rate Reliability for Engagement Cohort Analysis

Open rate data is the primary input for engagement cohort classification, but it is not perfectly reliable. Open rate measurement depends on an image pixel being loaded when the message is viewed — recipients who view email in plain-text mode, who have images disabled by default, or who use email clients that do not load images automatically will not register as opens even if they read the message carefully.

The estimated under-counting of genuine opens due to image blocking has historically been 20–40% in corporate email environments (where IT policies commonly disable images by default) and lower in consumer webmail environments where images load automatically. For engagement cohort classification, this means a contact who appears to have not opened in 90 days may actually be a regular reader whose client suppresses image loading. This is an argument for supplementing open-rate classification with click-rate data — clicks require active user engagement that is not affected by image blocking, and click history provides a more reliable engagement signal than open history for contacts who may be image-blocking readers.

The practical adjustment: maintain two parallel engagement classifications where possible — open-based (primary, for email clients that load images) and click-based (supplementary, immune to image-loading variations). Cohort classification should treat recent clickers as engaged regardless of their open history, since a click unambiguously demonstrates the recipient saw and acted on the message. Contacts with no open or click history over an extended period are more reliably classified as disengaged, because the absence of both signals is stronger evidence of non-engagement than the absence of opens alone.

Acquisition Source Engagement as Infrastructure Signal

Engagement rates are not uniform across acquisition sources. Contacts acquired through different channels — organic website signup, paid acquisition, co-registration, purchased lists, offline acquisition — have systematically different engagement profiles. Understanding which acquisition sources produce contacts with sustainable engagement levels enables infrastructure decisions about how aggressively to grow sending volume from each source.

Organic website signup contacts typically produce the highest engagement rates (30–45% open rate within the first 30 days) and lowest complaint rates (below 0.03%). Paid acquisition contacts vary significantly — search-driven acquisition produces higher engagement than display or social acquisition because the intent signal at search is stronger. Co-registration contacts (where the contact signed up for a third party and was shared) typically produce lower engagement and higher complaint rates than organic contacts. Purchased list contacts often produce complaint rates 3–10× higher than organic contacts and engagement rates that are 50–80% lower.

The infrastructure implication of acquisition source engagement differentials: new contacts from high-engagement-potential sources (organic signup, search-driven paid) can be introduced to the premium pool within 30 days of acquisition, because their engagement profile is likely to contribute positive signals. New contacts from lower-engagement-potential sources (co-registration, purchased lists) should be held in an isolated, lower-reputation pool for a validation period — typically 2–3 campaigns — before any graduation to the standard promotional pool, and should only graduate if their actual engagement rate (not assumed from the source) meets the minimum threshold for the pool.

This acquisition source routing is an infrastructure-level enforcement of list quality standards: it operationalises the list quality judgments that the engagement data reveals, rather than relying on marketing teams to remember to apply special handling to specific acquisition sources on a campaign-by-campaign basis. When the routing is built into the infrastructure (MailWizz list tagging and PowerMTA vMTA routing rules), it applies automatically regardless of which team member sets up the campaign.

The Infrastructure Benefits of Engagement-Driven Decisions

Engagement-driven infrastructure decisions produce compounding benefits over time that are difficult to attribute to any single decision but are visible in the programme's long-term deliverability trajectory. Programmes that implement engagement cohort routing, engagement-based frequency reduction, and acquisition source validation maintain IP pool reputation more stably than those that route all traffic through common infrastructure without engagement differentiation.

The mechanism: by isolating lower-engagement segments in separate pools, the programme's primary promotional pools receive only traffic with positive engagement profiles. The ISPs' reputation models for these pools receive consistently good signals — higher open and click rates, lower complaint rates, stable volume patterns. The pool reputation improves or stays stable, ISP rate limits become more generous as reputation improves, and campaign delivery times shorten as throughput capacity increases with reputation.

This feedback loop — better engagement → better reputation → more generous ISP rate limits → faster delivery → campaigns reaching more recipients during peak engagement windows → higher engagement — is the compounding dynamic that produces the significant inbox placement advantage that high-engagement programmes enjoy over moderate-engagement programmes at equivalent infrastructure investment levels. It is not the infrastructure itself that creates the advantage; it is the engagement-infrastructure feedback loop that the data pipeline enables.

Implementing engagement-driven infrastructure decisions requires the organisational alignment between marketing and infrastructure teams that is often absent. The marketing team controls engagement data and campaign decisions; the infrastructure team controls pool routing and delivery configuration. When these teams operate independently, the engagement data never reaches the infrastructure configuration, and the infrastructure performance data never reaches the list management decisions. Building the data pipeline and the organisational processes that connect them is as important as the technical implementation — the best engagement-routing architecture produces no benefit if it is not informed by current engagement data.

Measuring the Infrastructure Impact of Engagement Changes

Once engagement-driven infrastructure decisions are implemented, measuring their impact requires tracking infrastructure performance metrics alongside engagement metrics — not just individually, but as paired signals that reveal the relationship between them. The metrics to track in parallel: average open rate by pool (weekly), per-ISP deferral rate by pool (weekly), Gmail Postmaster Tools domain reputation (daily), and complaint rate by pool and by list segment (per campaign).

The expected patterns after implementing engagement cohort routing: (1) the premium pool's per-ISP deferral rates should decline as lower-engagement traffic is removed from it, because the positive reputation signals from higher-engagement traffic become a larger proportion of the pool's total reputation signals; (2) the re-engagement pool's complaint rate may increase after routing lower-engagement contacts to it, because the isolated pool now handles all the higher-risk traffic that previously diluted across the programme; (3) the Postmaster Tools domain reputation should stabilise or improve as the programme's overall authenticated mail has a higher proportion of positively-engaging traffic.

If these patterns do not emerge within 8–12 weeks of implementation, the engagement cohort classification thresholds may need adjustment — the cohort boundaries (30/60/90/120 days) are starting points based on general population behaviour, but programme-specific engagement patterns may justify different thresholds. Analyse the complaint rate and open rate distribution within each cohort to confirm that the boundaries are placing contacts in the correct risk category. A cohort classified as "moderate engagement" that has a complaint rate of 0.10% needs its boundary moved to capture it in the "low engagement" category where it belongs based on actual behaviour rather than the nominal age classification.

Practical Implementation in MailWizz and PowerMTA

Implementing engagement cohort routing in a MailWizz + PowerMTA environment requires configuration at three layers: contact tagging (or list segmentation), delivery server assignment, and virtual MTA routing.

In MailWizz, contacts can be tagged with their current engagement cohort using custom fields updated by a scheduled process that runs the cohort classification query against the contact database. The custom field value (e.g., "cohort_high", "cohort_moderate", "cohort_low") is then used as a segmentation criterion when creating campaign recipient lists. Campaign templates for each pool tier target only the appropriate cohort — the premium pool campaign targets cohort_high contacts, the standard pool campaign targets cohort_moderate, and the re-engagement pool campaign targets cohort_low.

In PowerMTA, each pool tier is implemented as a distinct virtual MTA with the pool's assigned sending IPs. MailWizz delivery servers are configured to use the virtual MTA corresponding to each pool tier. The routing then follows automatically: cohort_high contacts receive their campaign from the premium pool delivery server, which routes injection to the high-rep virtual MTA; cohort_low contacts receive theirs from the re-engagement pool delivery server.

The cohort classification process should run nightly to ensure contacts are reclassified as their engagement ages — a contact who opens today moves from cohort_low back to cohort_high, and their next campaign send routes through the premium pool rather than the re-engagement pool. This dynamic reclassification keeps the pool composition current without requiring manual list management, and rewards re-engaging contacts with the better delivery experience that the high-reputation pool provides.

Engagement data is the most available and most underutilised infrastructure input in most email sending programmes. Making it operational — connecting it to pool routing, frequency configuration, and retirement thresholds — produces infrastructure that self-optimises as the programme's engagement profile evolves, protecting IP reputation by keeping high-quality traffic in high-reputation pools and isolating lower-quality traffic in appropriately managed infrastructure. The technical implementation is straightforward in MailWizz + PowerMTA environments; the primary requirement is the organisational commitment to maintaining the engagement data pipeline that feeds the configuration.

The compounding nature of engagement-driven infrastructure decisions means their value increases over time rather than plateauing. Each quarter of disciplined engagement cohort management produces a pool reputation that is marginally better than the previous quarter, which produces marginally more generous ISP rate limits, which produces marginally faster delivery, which produces marginally higher per-campaign engagement (more recipients see the message during the peak engagement window after send). This 2–4% quarterly improvement in each metric compounds across years into a substantially different deliverability trajectory from programmes that manage infrastructure and engagement independently. The investment in connecting the two through data pipelines and routing configuration pays compound returns that are measurable in both infrastructure performance and programme revenue over time horizons of one year and longer.

Organisations that treat engagement data and infrastructure as separate concerns — one owned by marketing, the other by infrastructure teams — leave the most powerful connection between the two unimplemented. The data already exists in MailWizz: every open, click, complaint, and unsubscribe is recorded. The infrastructure configuration already supports pool routing based on virtually any criterion. The gap is the pipeline that connects the data to the configuration — and that gap, once closed, produces the engagement-infrastructure feedback loop that consistently separates excellent deliverability from merely acceptable deliverability at equivalent infrastructure investment levels.

The relationship between engagement data and infrastructure decisions is ultimately about closing the loop between what recipients do and how infrastructure responds. Programmes that instrument this loop fully — tracking engagement signals at the campaign and segment level, using those signals to guide sending decisions, and monitoring infrastructure metrics to confirm the signals are producing the expected reputation outcomes — are operating email infrastructure as a closed-loop system rather than an open-loop one. The closed-loop system self-corrects continuously through the feedback the data provides; the open-loop system accumulates problems that are only discovered when they manifest as delivery incidents. Engagement data is the feedback signal that makes the loop work, and infrastructure is what the signal governs.