Home / Role Packs / Marketing Analyst
Data Intelligence · Marketing

Marketing Analyst

Measure what marketing spend actually produced, and name the choices that change the answer. Never present one attribution model as the answer. Say when a campaign is not separable with the data available.

Add this role to my teamSee its 8 analyses
8named analyses
32governed capabilities
8operational responsibilities
1shared Core resolver
Questions this role answers

Choose the role by the work—not by a title alone.

  • Which channel deserves the credit?
  • What does a customer cost, what are they worth, and when do we break even?
  • Would these conversions have happened anyway?
  • Which channels bring customers who stay?
  • Where does more spend stop working?
  • Where do people from each channel drop out?
  • Are customers moving forward or sitting still?
  • Which creative is working, and is the difference real?
What this role actually does

One responsibility for every named analysis.

The Role Pack contains the method. The canonical profile contains the governed capability boundary. They reference each other without duplicating ServAI Core.

  1. 01

    Run every attribution model over the same touch set and report the spread as the real uncertainty.

  2. 02

    Compute CAC from fully-loaded spend and LTV from observed retention rather than an assumed rate.

  3. 03

    Read incrementality only against a holdout assigned before the campaign ran.

  4. 04

    Compare retention by acquisition source on cohorts truncated to equal maturity.

  5. 05

    Fit spend response within the observed range and refuse to recommend outside it.

  6. 06

    Report stage conversion by channel with stage definitions pinned to a version.

  7. 07

    Report lifecycle movement rather than stage counts, so stalling is visible separately from churn.

  8. 08

    Report creative performance with intervals and say when two creatives are not distinguishable.

Named analytical methods

8 analyses with evidence gates and refusal conditions.

Every method maps to a registered capability the profile can actually prepare or recommend. Repeating a precondition sentence is not evidence.

01marketing.attribution.model_comparison

Attribution · Model Comparison

Run last-click, first-touch, linear, time-decay and position-based over the same touch set and present all five. The spread between them is the answer's real uncertainty, and it is reported as a range rather than resolved.

Evidence gatethe conversion window is declared before the analysis runs
Refusesto report one model as the answer — attribution is a choice and the choice must be named
contribution.analysis5 declared outputsconfidential
02marketing.efficiency.cac_ltv_payback

Efficiency · Cac Ltv Payback

CAC = fully-loaded acquisition spend / new customers in the same cohort period. LTV from observed retention curve rather than an assumed churn rate. Payback = CAC / monthly contribution.

Evidence gatespend and customers are matched by cohort period, not by calendar overlap
Refusesto compute LTV from an assumed churn rate when observed retention exists
cohort.analysis6 declared outputsconfidential
03marketing.incrementality.holdout_read

Incrementality · Holdout Read

Compare an exposed group against a held-out control on the same period and definition. Reports lift with an interval; where no holdout exists, says so rather than inferring incrementality from correlation.

Evidence gatea holdout was assigned before the campaign ran
Refusesto claim incrementality with no holdout
analysis.comparison5 declared outputsconfidential
04marketing.cohort.retention_by_source

Cohort · Retention By Source

Retention curve per acquisition cohort split by source, aligned on periods since acquisition rather than calendar date.

Evidence gatecohorts aligned on time since acquisition
Refusesto compare cohorts of unequal maturity without truncating to the shortest
cohort.analysis4 declared outputsconfidential
05marketing.spend.saturation_curve

Spend · Saturation Curve

Fit response against spend by channel over observed range only. Reports the observed range explicitly and refuses to extrapolate past it.

Evidence gateenough spend variation within channel to see a curve
Refusesto recommend a spend level outside the observed range
driver.analysis4 declared outputsconfidential
06marketing.funnel.channel_to_revenue

Funnel · Channel To Revenue

Stage-by-stage conversion from first touch to revenue, per channel, with the stage definitions pinned to a version.

Evidence gatestage definitions versioned
Refusesto read a funnel across an instrumentation change without flagging the break
funnel.analysis3 declared outputsconfidential
07marketing.lifecycle.stage_movement

Lifecycle · Stage Movement

Movement between lifecycle stages per period, with the time spent in each, so stalling is visible separately from churn.

Evidence gatestage definitions versioned and owned
Refusesto report stage counts without movement — a stable count can hide equal inflow and outflow
funnel.analysis4 declared outputsconfidential
08marketing.creative.performance_read

Creative · Performance Read

Performance per creative with an interval, and a note where impressions are too few for the difference to be distinguishable.

Evidence gatecreatives ran against comparable audiences
Refusesto rank creatives whose intervals overlap — it says they are not distinguishable
segmentation.analysis4 declared outputsconfidential
Effective capability boundary

Resolved centrally. Scoped to the approved seat.

The Role Pack cannot grant itself authority. The shared resolver binds only what the profile, organization, tools, data classification and human policy permit.

Can read · 32
  • analysis · assumption
  • analysis · comparison
  • analysis · confidence
  • analysis · export
  • analysis · lineage
  • cohort · analysis
  • contribution · analysis
  • correlation · analysis
+ 24 more governed capabilities
Can prepare · 21
  • analysis · assumption
  • analysis · comparison
  • analysis · confidence
  • analysis · export
  • cohort · analysis
  • contribution · analysis
  • correlation · analysis
  • dashboard · definition
+ 13 more governed capabilities
Can recommend · 15
  • analysis · comparison
  • analysis · confidence
  • cohort · analysis
  • contribution · analysis
  • correlation · analysis
  • dataquality · assessment
  • distribution · analysis
  • driver · analysis
+ 7 more governed capabilities
Can request approval for · 3
  • analysis · export
  • dashboard · definition
  • report · definition
Fixed boundaries

What this role will not do.

  • Approve own request.
  • Bypass approval.
  • Impersonate a person.
  • Run raw sql supplied in a prompt.
  • Alter a source system.
  • Verify own analysis.
  • Hide query lineage.
  • Present single attribution model as truth.
  • Change campaign spend.
  • Compare across tracking break without warning.
  • Report roas without window and model.
  • Claim incrementality without holdout.
  • Recommend spend outside observed range.
Connectors

Give this role only the sources its work requires.

Choose from 1,000 named connector routes. Every connection remains tenant-specific, least-privilege, admitted and health-gated; write authority is never implied by catalogue presence.

Explore all connectorsAll eight roles
Marketing Analyst · Data Intelligence Role Pack

Add specialist depth to an approved seat.

ServAI proposes the exact role manifest. Your people review the profile, data scope, tools and approval path before anything becomes active.

Build My AI TeamDiscuss a deployment
Mission composition

Use this analyst depth across the Industry Packs you actually own.

This is a separately entitled cross-industry Role Pack. It can join an Industry Pack mission only when the organization owns both entitlements.

See Missions →Pricing structure