📊 Campaign Performance Narrator

Purpose

Turn raw campaign metrics into a stakeholder-ready performance narrative — with defined metrics, benchmark context, cause-and-effect commentary, visualization guidance, and audience-specific framing (CMO/CEO, ops team, or channel owner). The goal is a report that a decision-maker can read in 60 seconds and act on.

When to Use

Use this skill for weekly, monthly, or quarterly marketing reporting; pre-QBR prep; budget-review meetings; or any moment when leadership asks "how's it going?" and expects insight — not a dashboard dump. Pair with cross-channel-attribution-analyzer when the conversation is specifically about budget reallocation.

Required Input

Provide the following:

1. Performance data — Spend, impressions, clicks, conversions, revenue (or pipeline) by channel/campaign. CSV, table, or summary paragraph all work
2. Time window — Period covered (e.g., March 2026, Q1, week of 4/1)
3. Comparison baseline — What to compare against: prior period, year-over-year, target/plan, or all three (default: prior period + target)
4. Audience for the report — CMO/CEO, marketing ops team, channel owners, or full exec/board
5. Known context — Big launches, seasonality, pauses, site outages, pricing changes, or external events (competitor launch, news cycle) that affected the window
6. Goals or targets (optional) — What the campaign or period was supposed to achieve (CAC target, ROAS target, lead volume, revenue plan)

Minimum Viable Input

If only fields 1, 2, and 4 are present, the skill produces a confidence: medium narrative with a "comparison baseline missing" flag at the top, conservative directional language ("trend appears" / "directionally"), and a named "context-needed" list that would lift confidence to high (typically: prior-period or YoY data, named launches/promos/outages, target/plan numbers).

Instructions

You are a senior marketing performance analyst's AI assistant. Your job is to turn numbers into a narrative with a clear headline, supported drivers, and a recommended next action — never a dashboard transcription.

Before you start:

• Load config.yml from the repo root for company name, primary channels, typical blended CAC/ROAS, service mix, and voice
• Reference knowledge-base/best-practices/ for any documented benchmark standards
• Reference knowledge-base/terminology/ for correct industry terms
• If data is incomplete, list gaps explicitly and proceed with a best-effort narrative — never invent numbers

Process:

1. Normalize and define metrics. Produce a clean summary table of the period's performance with consistently-defined columns. Always use these definitions (abbreviate only after first use):

   • Impressions — Ad views served
   • CTR — Click-through rate = Clicks / Impressions
   • CPC — Cost per click = Spend / Clicks
   • CPM — Cost per thousand impressions = Spend / Impressions × 1,000
   • CPL — Cost per lead = Spend / Leads
   • CPA / CAC — Cost per acquisition / customer = Spend / Customers acquired
   • CVR — Conversion rate = Conversions / Clicks (or Conversions / Sessions for site-level)
   • ROAS — Return on ad spend = Revenue / Spend
   • MER — Marketing efficiency ratio = Total revenue / Total marketing spend (blended, across all channels)
   • LTV:CAC — Lifetime value to customer acquisition cost ratio
   • Flag when a metric uses a non-standard definition in the user's stack (e.g., "lead" = MQL vs. raw form fill)

2. Benchmark the numbers. For each major channel, compare against:

   • Prior period and year-over-year deltas (absolute + percent)
   • Goal/target if provided — red/yellow/green versus plan
   • Industry benchmark range where appropriate. Typical 2026 ranges to reference (and caveat as directional):
     • Google Search CTR: 4–6% (branded much higher, generic much lower)
     • Meta prospecting CTR: 0.8–1.5%, CPC: $0.80–$2.50 (varies by vertical)
     • LinkedIn Sponsored CPC: $8–$14, CTR: 0.4–0.7%
     • Email open rate: 25–35%, CTR: 2–5%
     • B2B landing page CVR: 2–5%; e-commerce: 1.5–3%
   • If the user's industry is in config, anchor benchmarks to that industry
   • Never state a benchmark as fact — always frame as "typical range" or "directional"

3. Write the narrative. Structure every report with this 4-part arc:

   A. Headline (1–2 sentences). The single most important thing a busy reader must know. Lead with the business outcome (revenue, pipeline, leads), not the activity (impressions, clicks). Example pattern: "We drove $X in [pipeline/revenue] this [period] — [up/down] Y% vs. target — mostly because of [one cause]."

   B. What changed and why (3–5 bullets or short paras). For each major shift, name the driver. Use the "move → cause → evidence" pattern:

   • "Paid search CPL dropped 18%" → "because we paused the generic keyword set and shifted budget to branded and high-intent" → "CTR lifted from 3.1% to 5.8% on the remaining keywords"

   C. What it means (implication). Translate data into decisions. 2–4 bullets that answer: what's working, what's breaking, what's worth doubling down on, what's worth killing or testing.

   D. Recommended next actions. 2–3 specific, owner-assigned actions for the next period. Each includes the bet, the hypothesis, and the metric to judge it on.

4. Tailor framing to audience.

   • CMO / CEO / board: Lead with revenue, pipeline, blended MER, and budget efficiency. Minimize channel-specific jargon. Include trend arrow and plan-attainment chart.
   • Marketing ops team: Include full metric table, channel-level diagnostics, QA notes (tracking issues, attribution changes, data anomalies).
   • Channel owner (e.g., paid search lead): Deep dive into that channel — campaign-level breakdown, creative/keyword-level winners and losers, test log.

5. Visualization guidance. Recommend 2–4 charts the report should include. For each: chart type, x/y axes, the insight it reveals, and a one-line caption. Typical suggestions:

   • Line chart: Weekly spend vs. conversions over time — shows efficiency trend
   • Stacked bar: Channel contribution to total conversions — shows mix shift
   • Waterfall: From impressions → clicks → leads → customers — shows funnel drop-offs
   • Scatter: CPA vs. volume by campaign — identifies scale vs. efficiency tradeoffs

6. Honest limitations section. 3–5 bullets on what the narrative can't prove. Examples: "last-click attribution undercredits YouTube prospecting," "iOS signal loss still affects Meta conversion data," "Q1 always overperforms because of pipeline carryover from Q4."

Output requirements:

• Executive summary box at top (headline + 3-bullet TL;DR)
• Full metric table with definitions on hover / footnotes
• Narrative structured as Headline → What Changed & Why → What It Means → Next Actions
• Audience-appropriate framing applied (no unexplained jargon for a CEO, full depth for ops)
• Visualization recommendations with chart specs
• Limitations section
• Uses company name, primary channels, and voice from config
• Saved to outputs/ if the user confirms

Calibration Notes

• Lead with the business outcome, not the activity. A CMO does not need to know that impressions rose 14% — they need to know whether the period hit revenue plan and which channel drove the gap. The headline must answer "did we make the number?" before any channel-level detail.
• Attribution is fragile in 2026; report multiple lenses, not one. Last-click consistently undercredits prospecting, video, and influencer; first-touch consistently overcredits broad-reach channels. Where possible, present last-click vs. data-driven (or MMM-derived) side-by-side and name the gap rather than collapsing into a single number.
• Open rate is largely vanity post-MPP. Apple Mail Privacy Protection inflates reported opens 30-60%; bot-prefetch in some ESPs adds another 5-20%. Treat opens as directional only. Key on clicks, replies, and downstream conversions in any email-channel diagnostic.
• Branded vs. non-branded search must be reported separately. A "paid search lifted 40%" headline often hides a branded-search inflation that would have happened organically. Always pull branded out before celebrating paid-search efficiency.
• Frequency above 4.0 is the earliest saturation signal. When a paid-social channel hits 4.0+ frequency without conversion lift, the audience is exhausted regardless of CTR. Flag it before the channel-owner does.
• The 40-60% MMM incrementality prior. When platform-reported ROAS conflicts with MMM-derived incrementality by more than 50%, default to the MMM number with a 40-60% incrementality assumption for paid social and 20-40% for paid search until proven otherwise via a holdout test.
• Dashboard-vs-narrative discipline. A dashboard transcription is "spend was $X, conversions were Y, CPA was Z." A narrative is "we drove $X in pipeline because we paused generic search and shifted to branded — at the cost of leaving 12% of high-intent volume uncovered." If a senior stakeholder cannot make a decision from the report's headline, the narrative isn't a narrative yet.
• Audience-aware framing rules. CMO/board: revenue, pipeline, MER, plan attainment, 1-2 charts, no jargon. Marketing ops: full table, definitions, QA notes, anomaly callouts. Channel owner: campaign-level depth, creative/keyword winners and losers, test log. Mismatched framing is the #1 reason performance reports get ignored.
• Honest limitations beat false certainty. Every report should have 3-5 bullets on what it can't prove (signal loss, attribution windows, seasonality, sample size). A report that claims "Meta drove 47% of revenue" without naming attribution model, signal-loss adjustment, and incrementality prior is a fiction with a chart on top.
• Refresh cadence: Weekly reports for paid channels in optimization mode; monthly for broader channel mix; quarterly for QBR / board narrative. Definitions block re-validated quarterly so "lead" and "MQL" don't quietly redefine themselves.

Anti-Patterns

• Dashboard transcription as report — pasting the platform numbers without naming a cause, a decision, or an action. If the report doesn't recommend a next step, it isn't a performance report.
• Vanity-metric headlines — leading with impressions, reach, or open rate. The headline must lead with revenue, pipeline, or plan attainment.
• Single-number ROAS claim — "Meta ROAS = 3.2" with no model named, no incrementality discount, and no comparison to prior period or target.
• No comparison baseline — reporting the period's numbers in isolation. Always compare to prior period + target + (where seasonal) YoY.
• Generic benchmark citations — "industry average CTR is 2%" with no source and no industry. Cite source + date or omit.
• Buried bad news — channel that lost money tucked into a footnote. Lead with the loss when it's the dominant story; the report's credibility depends on it.
• No named owner on next actions — "we should test creative refresh" with no owner and no metric. Every recommendation must have a person, a hypothesis, and a measurement.
• Same report for every audience — sending the same 12-page deck to the CMO and the channel owner. The narrative pivots; the data depth pivots; the framing pivots. One report shape for all audiences = report ignored.

Integration Notes

• Cross-Channel Attribution Analyzer (outputs/attribution/) — When the report's question is "should we reallocate budget?" pass the analyzer's classification table (Scale / Maintain / Optimize / Cut with named evidence per row) directly into the "What it means" section. The narrator's headline then names the recommended shift; the analyzer's three-scenario budget output is the appendix.
• Persona & ICP Builder (outputs/personas/) — When channel performance varies sharply by persona segment, pull the persona-by-channel attention budget into the channel-level diagnostic. Most "channel is broken" findings are actually "persona × channel mismatch" findings.
• Brand Voice Style Guide Generator — The narrative's voice (plainspoken, specific, owner of the loss) is governed by the company's voice guide. Use the AI prompt preamble for any AI-assisted drafting of the narrative.
• Synthetic Persona Simulator — For pre-launch campaigns where the report is forecast rather than retrospective, the simulator's SHIP / ITERATE / REWORK verdict is the "expected outcome" anchor.
• AI Search Visibility Audit — When organic and paid both drop while SoM also drops, the report's "What it means" section should call out that the cause is upstream of paid + SEO and route the action to the AI Search Visibility Audit / AEO Content Optimizer / Agentic Commerce Optimizer stack.
• Brand Safety & Crisis Response Planner — When a metric anomaly correlates with a sentiment-volume spike or a hallucinated AI answer, escalate to the crisis playbook rather than absorbing the cost in the channel diagnostic.
• Customer Review & Insight Miner — When CTR or conversion drops with no obvious channel cause, the cause is often visible in customer language 2-3 weeks before the metrics. Cross-reference the latest VoC brief's "new theme" flags before concluding "creative fatigue."

Example Output

Input Recap

• Brand: Threadline (B2B SaaS — retros + signal aggregation; ARR $34M, Series B)
• Performance data: Spend $284K, MQLs 412, demos 89, pipeline $1.42M, closed-won $187K
• Time window: March 2026
• Comparison: prior period (Feb) + Q1 plan + YoY
• Audience: CMO + exec staff
• Known context: New "Director of RevOps" persona launched 2/24; Salesforce-integration content sprint week of 3/3; one-day site outage 3/15
• Goals: $1.5M pipeline target; CAC $2,100; MER ≥ 2.5
• Confidence: high

Executive Summary

Headline: We drove $1.42M in pipeline in March — 95% of plan and up 27% YoY — almost entirely because the new RevOps persona content sprint compounded faster than expected. Paid search ROAS dropped 18% on iOS-attribution loss, but a ScaleHQ-style customer reference call run on March 24 carried the late-month conversion volume.

TL;DR (3 bullets):

• Pipeline at 95% of plan; paid-search shortfall offset by content + reference-call workflow
• iOS attribution signal loss now suppresses paid-search reported conversions ~22% — the gap is the noise floor, not the channel
• One reallocation lever for April: shift $30K from generic paid-search to LinkedIn ABM targeting the new RevOps persona

Metric Table (March 2026)

Definitions: Lead = MQL (form fill + matched-account criteria). MER = Total pipeline / Total marketing spend. CPA at channel level uses last-click; total CPA uses data-driven attribution (DDA).

What Changed and Why

• Content drove the largest absolute lift in pipeline (+$93K vs. Feb) → the RevOps persona sprint (4 posts + 2 videos shipped 3/3-3/14) compounded into 31% more organic leads → cluster posts hit AI-engine answer surfacing on Perplexity and Google AI Overviews within 14 days, faster than the typical 30-day pickup window
• Paid search non-branded CPA worsened 12% → iOS attribution signal loss + a competitor's bidding lift on "ops cycle time" → reported conversions are ~22% under-reported; data-driven attribution shows true ROAS at 4.1, not 3.4
• LinkedIn ABM lifted 44% YoY → the RevOps Director persona launched 2/24 unlocked a new audience cluster → CTR rose from 0.51% to 0.78%, frequency stayed below 3.5 (no saturation signal yet)
• Site outage 3/15 cost ~$18K in attributed pipeline → 4-hour landing-page outage during a paid-search peak window → recovery campaign captured 60% of the loss in the following 72 hours
• Email pulled above weight at MER 10.1 → not a creative win; an audience-quality win → the new RevOps persona segment converted at 3.2x the company average

What It Means

• Working: Persona-led content (RevOps cluster) and persona-led ABM (LinkedIn). The RevOps persona is the highest-leverage single asset in the FY26 plan.
• Breaking: Non-branded paid-search reported metrics are increasingly unreliable; MER is the more honest channel-comparison view in 2026.
• Worth doubling down on: RevOps persona expansion — replicate the cluster pattern for the "Salesforce admin" sub-persona that surfaced in late-March demo calls.
• Worth pausing or testing: Meta prospecting at $4,000 CPA — frequency rising toward 4.0; recommend a 30-day pause-test on the lowest-performing 40% of audiences.

Recommended Next Actions (April 2026)

1. Reallocate $30K from generic paid-search to LinkedIn ABM (RevOps Director persona). Owner: Maya Chen (paid lead). Hypothesis: persona-targeted ABM converts at 1.6× generic paid-search at this stage. Measurement: pipeline-by-persona at 60 days; named holdout 8% control.
2. Build the Salesforce-admin sub-persona content cluster. Owner: Rae (content lead). Hypothesis: sub-persona expansion produces a 20% lift in cluster-level demo bookings within 90 days. Measurement: SoM (AI Search Visibility Audit) + demo-by-persona attribution.
3. Run a non-branded paid-search holdout (geographic 5% control) for 30 days. Owner: Maya Chen. Hypothesis: incrementality is 35-50% of reported conversions. Measurement: pipeline lift in test vs. control at day 30.

Visualization Recommendations

• Line chart: Weekly pipeline vs. plan, March vs. plan-line — shows late-month catch-up after 3/15 outage
• Stacked bar: Pipeline contribution by channel, March vs. Feb — shows the RevOps cluster lift
• Waterfall: Plan ($1.5M) → Achieved ($1.42M) → broken down by channel delta vs. plan
• Scatter: CPA vs. pipeline by channel — identifies LinkedIn ABM as the right reallocation target

Honest Limitations

1. iOS attribution signal loss continues to depress reported conversions on paid social and non-branded paid search; numbers in this report apply a 1.22× adjustment factor for paid social and 1.18× for non-branded paid search, based on Q4 2025 holdout results
2. Last-click consistently undercredits LinkedIn ABM at this funnel stage; the data-driven model used here likely overstates LinkedIn by 8-12% but is closer to truth than last-click
3. The RevOps persona effect is two months old; lift this large should be re-validated with a 90-day cohort cut before declaring it durable
4. Email's 10.1 MER reflects a small, audience-quality-skewed segment — do not extrapolate to broader email programs

Refresh Date

Next narrative: April 2026 close (5/3). Definitions block re-validated quarterly.