— 11 measurement & decisions decisions, not dashboards

we don't send dashboards. we send decisions.

every month: a one-page board doc that says scale, hold, kill, or test — with the math shown, the reasoning written down, and a kill-switch attached. dashboards are wallpaper. decisions are the deliverable.

— deliverable
1 page, board-ready
— cadence
weekly read · monthly board
— attribution
3-layer stack + holdouts
— ownership
your stack, your data
01. why this matters now

most reports are summaries, not recommendations.

you've sat through the monthly review. forty slides of dashboards. "impressions are up. CTR is up. ROAS is up." nobody asks the question that matters: so what should we do differently next month.

agencies write reports that survive plausible deniability. all the numbers go up so nothing has to change. nothing gets killed. nothing gets bet on. the report exists to prove the engagement is alive, not to make the business better.

we write reports the way a fund writes a quarterly letter: here's what we did, here's what worked, here's what didn't, here's what we're doing next, here's the math. one page. readable in five minutes. defensible in a board meeting.

and we put a recommendation against every linescale, hold, kill, or test. if we can't make a recommendation, we say so out loud and write down what we'd need to know. no hedging, no "we'll keep monitoring."

~62%
— of marketing reports summarise rather than recommend
5min
— is what a cmo or board member will give your monthly read — if you're lucky
~28%
— of tracked metrics meaningfully change a decision. the rest is wallpaper
0
— recommendations to kill a channel in a typical agency monthly — ever
02. how we think about reporting

four operating rules every report follows.

a report is a document of decisions, not a record of activity. these are the four rules we hold ourselves to — the guardrails that keep us from writing the kind of monthly that gets opened, scrolled, and closed without a single thing changing.

01
— rule 01

one page. always.

if it's longer than one page, it's not a report — it's a defence brief. our monthly is one page because discipline beats detail: the act of reducing the month to a single sheet forces us to decide what mattered and what was noise. detail goes in the appendix; nobody reads it, and that's fine.

— what we cut
impressions, CPM, frequency, every screenshot, every "creative highlight" carousel. anything that doesn't change a decision goes in the appendix.
02
— rule 02

every line gets a verb. scale, hold, kill, or test.

numbers without a verb are weather reports. we put a recommendation against every channel, every cohort, every creative concept — and we sign our name to it. scale means add budget. hold means stay the course. kill means stop spending here. test means we don't know yet, here's what would settle it.

— what's not allowed
"we'll continue to monitor." "we'll optimise further." "let's see how next month goes." every line gets a verb — or it gets cut.
03
— rule 03

show the math. and the doubt.

every recommendation cites the data that supports it — and the data that pushes back. if the model says scale but the holdout disagrees, that's in the report. if we're 60% confident, we say 60%. reports that pretend to certainty get believed once and dismissed forever after. we'd rather be honest about the doubt.

— how we say "i don't know"
"the model says X, the holdout says Y, the platform says Z. we recommend X, with a 30-day re-read because Y is loud."
04
— rule 04

your stack. your data. your report.

we build inside your warehouse — bigquery, snowflake, postgres, looker studio, mode, whatever's already there. the dashboards stay yours. the dbt models stay yours. when the engagement ends, you keep every query, every model, every report template. we never lock measurement behind our tooling.

— what stays with you
warehouse models, attribution logic, dashboard templates, holdout history, scoring rubrics, the report doc itself. zero handover friction by design.
03. the monthly board doc

one page. four verbs. every month.

this is the format we ship every monthly review — redacted from a real client doc. tldr at the top, the four-verb quadrant in the middle, footer with confidence and risk. nothing that doesn't change a decision survives this page.

monthly read — nov 2025
— period: 01 nov – 30 nov
— author: mafia · growth
— status: final · signed by m. shah
— tldr
blended MER held at 2.7×, ahead of plan. the win is tiktok creator taking 22% of mix at 4.6× marginal — we're scaling. brand-search is double-paying for organic clicks; killing $14k/mo. youtube shorts ready to scale; meta retargeting needs a creative refresh and is on hold. one open question on amazon DSP — testing it next month.
— quadrantscale
  • tiktok creator ads — 22% mix, 4.6× marginal ROAS+$32k
  • youtube shorts — brand-search lift +14% over 60d+$18k
  • spark-ads from oct cohort — 7 winners ready to whitelist+$12k
— quadranthold
  • meta retargeting — ROAS 4.2× but creative fatigue at 7drefresh
  • google non-brand — CAC $58, at thresholdwatch
  • email lifecycle — 2.1% revenue, holding patternstable
— quadrantkill
  • google brand-search — 78% incrementality fail, double-paying−$14k
  • meta lookalike of converters — marginal CAC $112, above ceiling−$8k
  • pinterest test budget — 60d learning, no signal−$5k
— quadranttest
  • amazon DSP — needs 4w holdout to validate$10k
  • ugc hook framework v2 — 12 new units, 14d test$8k
  • sms retargeting — vs. paid social retarget control$4k
— confidence: high on scale + kill · medium on test · risk: holiday-window CPMs +28%
scale hold kill test
— note 01 / tldr
five sentences max. the cfo reads only this. if it can't be said in five sentences, the month wasn't understood.
— note 02 / verbs
every line in the quadrant has a dollar delta and a verb. no "let's see" lines. ever.
— note 03 / signed
a person's name goes on every report. accountability is part of the format. the report is owned, not produced.
— note 04 / risk
the risk line is what could break the read. holiday CPMs, ios update, channel saturation — the thing we're watching that isn't in the numbers yet.
04. how we wire it up

five weeks from kickoff to first signed read.

we don't take over a measurement stack on day one. we audit, model, instrument, then ship the first board doc — and only after it's been signed off do we move to the always-on monthly cadence.

01
— week 01data audit

where the numbers live, what they mean.

we map your actual data flow: ad platforms → warehouse → dashboard, plus shopify/stripe/ga4/crm. document where each metric is defined, where it's broken, where it's double-counted. usually 30% of "the numbers" don't agree with each other — that gets fixed first.

data lineage metric definitions discrepancy log
02
— week 02attribution stack

three layers, one source of truth.

we set up the three-layer attribution: platform-reported (with skepticism), MER vs. forecast (the truth-line), and quarterly incrementality holdouts (the validator). models live in your warehouse, not ours. dashboards expose the math, not just the answer.

3-layer model holdout calendar MER baseline
03
— week 03report scaffold

the four-verb format, on your data.

we build the one-page report scaffold — tldr, four-verb quadrant, footer — and populate it with last quarter's data as a backtest. you read the historical version first: would those calls have been right? if the format doesn't survive the backtest, we tune it before going live.

report template backtest format sign-off
04
— week 04weekly read goes live

monday, every monday, no exceptions.

weekly read is a one-screen friday-to-friday rollup — pacing, marginal CAC, MER, three calls. lands monday 9am, signed. by week six the rhythm is the rhythm: monday read, thursday creative ship, end-of-month board doc.

monday read three calls friday rollup
05
— week 05first board doc

one page. four verbs. signed.

end of month one: first signed monthly read. tldr, quadrant, risk, signature. presented in 20 minutes, defended in 30. after that, it ships on the same day every month — and the format never bloats. discipline is the deliverable.

monthly board doc live presentation cadence locked

documents boards read. decisions that ship.

— 01 / format
1
page per month, every account, every time. detail in the appendix — if you ask
— 02 / discipline
~38%
of tracked metrics typically retired on engagement — wallpaper, not signal
— 03 / verbs
100%
of report lines carry a scale / hold / kill / test recommendation. no exceptions
— 04 / ownership
0
measurement assets locked behind our tooling. your warehouse, your queries, always
"

first agency that ever wrote "kill this" in a monthly. and they were right.

M
marcus l.
— cmo · b2b saas
05. faq

questions, answered straight.

can't find it here? email hi@socialmafia.agency — replies in under 24 hours.

do we still need our existing analytics team?
yes — we work with them, not over them. our reporting layer sits on top of the warehouse your team owns. we don't replace data engineers or analysts; we replace the layer that turns numbers into decisions, which is usually the layer they're under-resourced to do.
what if we don't have a data warehouse?
we'll set one up — bigquery or snowflake, your call — or use a leaner stack (postgres + dbt-core + metabase) if the volume doesn't justify it. either way, the warehouse and the models are yours. we don't run measurement out of our own tenant; that's a lock-in pattern we refuse on principle.
how do you handle attribution post-iOS / cookie deprecation?
three-layer stack: platform-reported (skeptically), MER vs. forecast (the financial truth), and quarterly incrementality holdouts (real, not modelled). we don't believe platform ROAS in isolation — that number is a marketing aid, not a financial one. our reports always cite which layer is talking.
can you report on more than just paid?
yes — the four-verb format works across paid, organic, lifecycle, partnerships, and product. most of our reports cover the full marketing surface, with a separate appendix per channel. the one-page summary is unified.
how long until the first useful read?
the weekly read is live by week four. the first monthly board doc ships at the end of month one. by month three the cadence is fully dialled in — you stop noticing the report exists, which is the point. good measurement disappears into the rhythm.
what if your call is wrong?
the report names the call, names the confidence level, and names the conditions under which we'd reverse it. we track our own hit-rate across recommendations — scale, hold, kill, test — and publish it internally each quarter. if our calls aren't beating chance, that's our problem to fix, not yours.
do we have to switch off our existing dashboards?
no — keep them. we usually add the four-verb monthly on top of whatever exists. over time, the daily dashboards naturally get pruned because nobody opens them, but that's an emergent thing, not a forced migration. tools are not the problem; the format is.

stop summarising. start recommending.

— one page · four verbs · signed · your stack, your data