Every CAC and ROAS figure below is last-click attributed: useful for daily pacing, but directionally biased. Brand search and comparison pages both sit late in a PLG research journey that organic content and word-of-mouth largely originate, last-click over-credits the close and under-credits the assist. The Measurement Lab tab holds the causal layer (geo holdouts, lift tests, MMM backbone). Until incrementality is fully built out, treat channel-level numbers here as relative signals, not absolute ROI.

In a freemium PLG motion, raw signups are a vanity metric on their own. The funnel bottleneck is signup → activation: 60.0% of this month's signups completed a first meaningful workspace action (created a task, doc, or list) within 7 days of signing up. Activated workspaces convert to paid at roughly 6× the rate of non-activated ones. This reframes the marketing mandate: channels should be scored on activated-signup delivery, not raw signup volume, which is why the channel table above will get an "Activated Signups" column once the Layer 3 identity spine (Measurement Lab) is live.

Every asset class below is produced by a dedicated agent workflow, not a person prompting ChatGPT ad hoc, and not one agent doing all of it: a comparison-page agent that pulls competitor feature-parity data and drafts structure, a paid-creative agent that generates hook/body/CTA variants against a live performance library, a lifecycle agent that drafts subject-line and send-time variants per segment, an SEO-content agent that drafts blog and organic articles, and a video-script agent for UGC-style ad scripts. Every one of these agents' output is checked against the same shared Playbook by a separate, single-purpose model call, not by the agent that wrote it, before a human ever sees it, none of them grades its own homework (see Systems & Learning Loop for what's actually in the Playbook and how it changes over time). None of it ships un-reviewed: see the Quality Gate tab. The point isn't zero-touch output; it's that a small review team can hold a much larger volume of output to one shared standard.

Video/UGC ad scripts have the lowest 1st-pass rate (43%) and the lowest final rate (72%) of any asset class. The agent workflow is newest here and hasn't accumulated as many logged rejection patterns yet. This is a deliberate, visible weak point rather than a hidden one: the Systems & Learning Loop tab shows exactly which rule (v1.4, quantified-claim requirement) closed an 11-point gap in this category, and why video is the next priority for Playbook investment.

Each card below is one AI-drafted marketing candidate, produced by one of the specialized agents named in Content & Campaign Systems. The score and flag come from an automatic check against the current Playbook, not a person's note, so reviewers start with a head start, not a blank page. Analysts approve, revise, or reject: that decision is the actual human review, and it's the raw input that improves the Playbook over time, see Systems & Learning Loop for how.

Every Quality Gate decision is logged with a reason tag. Nightly, a Marketing Systems Agent clusters recurring reasons and proposes a specific Playbook update that's never auto-deployed: it routes to Director/CMO approval in Action Center first. Once approved, that version becomes the Playbook every agent generates against, so each fix compounds across every future asset instead of resetting each quarter.

It would be technically straightforward to let the Marketing Systems Agent push Playbook changes automatically once confidence crosses a threshold. This is explicitly not built, for the same reason a poisoned or over-fit context is worse than no context at all: an unreviewed Playbook change propagates to every agent immediately, and a bad one compounds silently across hundreds of assets before anyone notices the pattern. The one-week lag introduced by human sign-off is a small cost against that downside.

Layer 1, Operational Truth (Weeks 1–4): Last-click attribution via GA4 + product analytics. Sets the daily-pacing baseline. Biased but necessary for anomaly detection.

Layer 2, Strategic Truth (Months 2–4): Incrementality testing: geo holdouts on paid social and non-brand search, matched-market tests on comparison-page SEO refreshes, conversion lift studies via platform-native tools (Meta, Google). This is the causal layer that corrects for last-click's over-crediting of brand search and under-crediting of organic content.

Layer 3, Foundational Truth (Months 3–6): Unified warehouse model (BigQuery/Snowflake), keyed by a stable identity spine (workspace ID + hashed account email) spanning free and paid. Unlike a business with only a couple of physical locations, Acme has years of channel-spend history and thousands of geographies to split on, media mix modeling is not just viable here, it's overdue. The data this layer logs today is exactly the MMM input set for next year.

Setup: 14 matched market pairs (DMA-equivalent), split on pre-period signup trend (parallel-trends check required). Treatment markets run normal paid social; holdout markets get spend cut to zero for 6 weeks.

Estimator: Difference-in-differences. Treatment effect = (treatment post − treatment pre) − (control post − control pre). Pre-trend window: minimum 8 weeks.

Why this matters for the Non-Brand Search flag in Acquisition: the 0.5× last-click ROAS on non-brand search is exactly the kind of number that under- or over-states truth depending on how much of that traffic would have converted anyway via organic. No budget reallocation ships off a last-click number alone. See Action Center.

Primary: New ARR added (weekly) | Activated-signup rate | Free → paid CVR (30d rolling)
Secondary: 1st-pass Quality Gate rate | Content pipeline velocity (brief → published) | Referral program ROI
Channel: Organic session share | Comparison-page CVR | Blended paid CAC | Email open + CTR
Efficiency: Revenue per marketing dollar (post-incrementality) | Human review hours per published asset

Deliberate non-investment: No production MMM output in Quarter 1. The identity spine (Layer 3) needs to stabilize first, or the model will be fit on incomplete, mis-keyed data. Building the model before the data foundation is a common failure mode this dashboard is explicitly avoiding.

Not every decision needs the CMO's sign-off, but some categorically must; especially anything that changes what the system generates next.
Analyst / Content Lead (self-approve): Single-asset Quality Gate decisions, tactical spend changes <$5K/wk impact.
Growth / Content Lead: New pipeline launches or pauses, spend changes $5K–$50K/wk, single-asset-class Playbook tweaks.
CMO / Director: Any Playbook version change with company-wide scope, budget reallocation >$50K/wk, brand-voice guideline changes, anything with legal/compliance exposure.