A six-step workflow that turns a campaign export into a decision queue and a weekly brief.
Most campaign analysis produces a dashboard nobody acts on. This produces four lists: scale, investigate, hold, test. Each with the finding, the numbers behind it, and the next action.
Inside: the input templates, the reconciliation checks that catch bad data before you analyse it, the cohort comparison that stops you reading noise as a trend, all six prompts written out, the action board, and a complete worked example.
Step one is the one people skip and the reason most analysis is wrong.
AI Driven, Personalized Outreach on LinkedIn: Prosp 💜
Setup: The Three Inputs
Three inputs. Get these right and the six steps run on their own.
Get them wrong and you will produce confident analysis of bad data, which is worse than no analysis.
The three inputs
campaigns/
├── export.csv the raw campaign data
├── targets.md what good looks like, with numbers
└── events.md the events that actually matter
Nothing runs without all three. The workflow fails closed on a missing file rather than inferring what the targets probably are.
export.csv
One row per campaign per day. Not per week, because weekly rollups hide the day a change happened.
REQUIRED COLUMNS
date
campaign_name
channel
spend 0 for organic outreach
reach sends, impressions, or requests
step_1 accepted, clicked, or installed
step_2 replied, signed up, or activated
step_3 meeting booked, or trial started
step_4 closed, or paid
STRONGLY RECOMMENDED
segment the audience or ICP slice
creative_or_angle which message or creative
geo
signal_type which trigger brought them in
For LinkedIn outreach campaigns the four steps map to:
reach connection requests sent
step_1 requests accepted
step_2 replies received
step_3 meetings booked
step_4 deals closed
For a paid campaign into a B2B funnel:
reach impressions
step_1 clicks
step_2 form fills or sign-ups
step_3 qualified, or meeting booked
step_4 closed
For a mobile app, the original shape:
reach impressions
step_1 installs
step_2 activation event
step_3 trial or first key action
step_4 paying
The workflow does not care which. It cares that the four steps are consistent across every row and that each one is defined somewhere.
The two columns people leave out and then wish they had: signal_type and creative_or_angle. Without them, step 4 can tell you a campaign changed but not which part of it.
targets.md
What good looks like, with numbers. Written before you look at the data, not after.
# Targets
## The headline number
What we are trying to move this quarter, from X to Y, by when.
## Per stage, what healthy looks like
reach to step_1:
step_1 to step_2:
step_2 to step_3:
step_3 to step_4:
If you do not know, write UNKNOWN. Do not guess, because the
workflow will treat a guess as a benchmark.
## Cost ceilings
Maximum acceptable cost per step_3:
Maximum acceptable cost per step_4:
What a customer is worth, and over what period:
## The ceiling that is not money
For outreach: the daily send budget, ACCOUNT-WIDE.
For paid: the daily spend cap and who approves raising it.
## Kill criteria, set in advance
What failure looks like, with a date.
The kill criteria field is the one people fill in afterwards. Deciding what failure looks like after seeing the numbers is not a decision, it is a justification.
For outreach specifically, the healthy benchmarks:
reach to accepted below 15% means something is broken
25 to 30% on a cold signal-led list is
healthy
accepted to replied below 10% means the first message
replied to meeting varies by offer, but a drop here is
usually the ask, not the list
events.md
What each step actually means, in your system. One line each.
# Events
## step_1
The exact event name in the platform:
What it means in plain English:
When it fires, and whether it can fire twice for one person:
## step_2
[same]
## step_3
[same]
## step_4
[same]
## Known gaps
Anything not tracked, named here rather than discovered later.
## Attribution window
How long after reach does a conversion still count?
"Can it fire twice for one person" is the field that prevents the most wrong analysis. A step that double-counts makes a funnel look better at that stage and worse at the next one, which sends you investigating the wrong thing.
The attribution window matters more than people expect in B2B. A 7-day window on a 90-day sales cycle attributes almost nothing correctly, and the analysis will tell you every campaign failed.
The known gaps section is not optional. If meeting data is not recorded, the workflow needs to say "stage 3 cannot be assessed" rather than reporting zero as a finding.
What this workflow does not do
It does not fix attribution. If your lead source says "LinkedIn" rather than naming the campaign, step 4 cannot tell you which campaign produced the revenue, and no amount of analysis recovers it.
It does not replace a decision. It produces a queue of four lists. Someone still has to pick.
It does not change anything. No budget moves, no campaign pauses, no spend shifts. It reads and reports.
It does not work on one week of data. Step 3 compares cohorts at equal maturity, which needs at least two complete cohorts. On a 30-day cycle that is 60 days of data minimum.
Steps 1 to 3: Check, Map, Compare
Steps one to three. Validate, map, compare.
Nothing here produces a recommendation. That is deliberate: finding things before you have checked the data is how you end up confidently wrong.
Step 1: check the data
The step people skip, and the reason most campaign analysis is wrong.
---
name: 01-data-check
description: Validate the export before any analysis. Fail closed
on anything that does not reconcile. Lowest effort.
---
READ campaigns/export.csv and campaigns/events.md.
CHECK AND REPORT:
MISSING FIELDS
Which columns have blanks, how many rows, and which
campaigns they cluster in.
A column that is 90% empty is not a column, it is a gap.
DATE PROBLEMS
Rows outside the stated reporting period.
Missing days. A campaign with no Tuesday is either paused or
the export is broken, and those need different responses.
Timezone mismatches between sources.
NUMBERS THAT DO NOT RECONCILE
Any row where a later step exceeds an earlier one.
Any row where spend is present but reach is zero.
Any row where step_4 is present but step_3 is zero.
Totals that do not match the platform's own reported total.
DOUBLE COUNTING
Check events.md for any step that can fire twice per person.
Flag every campaign where that step exceeds a plausible
ratio of the one before it.
SEGMENT CONSISTENCY
The same segment named differently across rows.
"SMB", "smb" and "Small Business" are three segments to a
spreadsheet and one to you.
OUTPUT
A table of every issue: what, where, how many rows, severity.
Then one of three verdicts:
CLEAN proceed
USABLE proceed, with the affected campaigns excluded
and named
NOT USABLE stop. Fix the export first.
FAIL CLOSED on NOT USABLE. Do not proceed to step 2 and do not
produce partial findings. An analysis built on a broken export
points confidently at the wrong campaign.
The reconciliation check against the platform's own total is the one that catches export errors. If your CSV says 1,200 sends and the platform says 1,450, something is filtered and you do not know what.
USABLE is the common verdict. Most exports have two or three broken campaigns, and naming them and carrying on is better than either ignoring them or stopping entirely.
Step 2: map the funnel
---
name: 02-funnel-map
description: Follow the spend, or the sends, all the way through.
Medium effort.
---
READ the validated export and campaigns/targets.md.
BUILD THE FUNNEL, overall and per campaign:
reach → step_1 → step_2 → step_3 → step_4
PER STAGE REPORT:
The absolute number
The conversion rate from the previous stage
The rate against the target in targets.md
The cost per unit at that stage, where spend exists
EVERY PERCENTAGE SITS BESIDE ITS ABSOLUTE. Up 40% from 5 to 7
is a different sentence from 500 to 700, and a report showing
only the percentage has said almost nothing.
THEN THE DIAGNOSTIC. Work in order, stop at the first stage
below target:
1 reach to step_1 below target
The entry point. For outreach, the connection note or the
list. For paid, the creative or the audience.
NOT the later stages. Do not report them.
2 step_1 fine, step_2 below target
The first thing they receive after entering. The message,
or the landing page.
3 step_2 fine, step_3 below target
The ask is mistimed, or the offer is wrong.
4 step_3 fine, step_4 below target
Not an acquisition problem. Price, fit, or sales process.
5 All stages fine, volume low
Not quality. The cap, the budget, or the list size.
OUTPUT the failing stage, the evidence, and nothing else.
DO NOT REPORT FIVE STAGES WHEN THE FIRST IS BROKEN. A stage 4
recommendation delivered while stage 1 is failing is why
analysis gets ignored.
IF A STAGE CANNOT BE ASSESSED because the data is missing, say
so by name. A diagnosis on partial data points at the wrong
stage, which is worse than no diagnosis.
Stage 5 is the one people never check. Every rate healthy and volume low is not a quality problem, and tuning the creative when the real issue is a daily cap wastes a month.
Step 3: compare the cohorts
The step that separates a real change from noise.
---
name: 03-cohort-compare
description: Compare users at EQUAL MATURITY, not at equal
calendar dates. High effort.
---
THE PROBLEM THIS SOLVES:
A cohort that entered 7 days ago has had 7 days to convert. One
that entered 60 days ago has had 60. Comparing their conversion
rates directly makes the recent cohort look worse every single
time, and that is an artefact rather than a finding.
Most campaign analysis makes this mistake and concludes that
performance is declining when nothing has changed.
THE METHOD:
1 Group by entry period. Weekly or monthly, whichever matches
your cycle length.
2 For each cohort, measure conversion AT THE SAME NUMBER OF
DAYS SINCE ENTRY.
Cohort A at day 30. Cohort B at day 30. Not cohort A at day
60 against cohort B at day 14.
3 Report the comparable windows only. If the newest cohort has
only 14 days of maturity, the comparison is day 14 across
every cohort, and you say so.
4 For anything with more maturity, report it separately as
context rather than mixing it in.
OUTPUT per cohort:
Entry period and size
Conversion at day 7, 14, 30, 60, 90, where available
Which windows are comparable across all cohorts
WHICH WINDOWS ARE NOT, named explicitly
THEN: is there a real difference between cohorts at equal
maturity, or is the apparent difference entirely explained by
time?
RULES
Never compare a 14-day-old cohort's total conversion against a
90-day-old cohort's total conversion. State that you will not,
and why.
FLAG small cohorts. A cohort of 11 people converting at 18%
versus one of 400 converting at 14% is not a finding. It is two
extra conversions.
FAIL CLOSED with fewer than two complete cohorts. Say the
comparison is not yet possible and when it will be.
The small-cohort flag matters more in B2B than anywhere else. At 20 connection requests a day, a weekly cohort is 100 people, and a few conversions either way swings the rate by several points without meaning anything.
The time-to-convert question is the useful output. If cohort A converts at 8% by day 14 and cohort B reaches 8% only by day 45, the campaigns are not performing differently. They are producing buyers with different cycle lengths, which is a targeting insight rather than a creative one.
Steps 4 to 6: Find, Decide, Brief
Steps four to six. Find the changes, prepare the decisions, write the brief.
These are the only steps that produce recommendations, and they only run on data that passed steps one to three.
Step 4: find the changes
---
name: 04-change-finder
description: Identify which campaigns, segments and angles moved,
and whether the move is real. High effort.
---
READ the validated export and the cohort comparison.
FOR EACH DIMENSION in turn, compare this period against the
last:
Campaign
Channel
Segment
Creative or angle
Geo
Signal type, where recorded
PER MOVEMENT REPORT:
What moved, at which stage
The absolute change, beside the percentage
The sample size on both sides
Whether the cohorts being compared are at equal maturity
Whether the movement is larger than normal week-to-week
variation for that campaign
THE LAST TWO ARE THE FILTER. A 20% swing on a campaign that
swings 25% every week is not a change. A 20% swing on a
campaign that has held within 5% for two months is.
RANK by how much the movement affects the headline number in
targets.md, not by the size of the percentage.
A 60% improvement on a campaign producing two meetings a month
matters less than a 6% decline on one producing forty.
OUTPUT no more than TEN movements. Ten is a list someone works
through. Forty is a document nobody opens.
FLAG SEPARATELY anything that changed because of something you
did: a budget change, a new creative, a paused campaign, a
changed list. Those are not discoveries, they are consequences,
and mixing them in makes the real findings harder to see.
FAIL CLOSED on any movement where the sample is under 30 on
either side. Report it as "too small to assess" rather than as
a finding.
The "larger than normal variation" check is what stops this becoming a weekly list of noise. Most campaigns move 10 to 20% week to week for no reason, and treating every swing as a finding trains people to ignore the report.
The consequences flag is the other one. If you raised a budget on Monday and the campaign grew on Wednesday, that is not an insight.
Step 5: prepare the decisions
Four lists. Every finding goes in exactly one.
---
name: 05-decision-queue
description: Turn findings into a queue of four lists. Medium
effort.
---
PER FINDING from step 4, assign exactly one:
SCALE
Performing above target, with enough volume to trust it, and
headroom to grow.
REQUIRED: the current number, the target, the sample size,
and WHAT THE CEILING IS. For outreach, the account-wide daily
limit. For paid, the budget cap and who approves it.
A scale recommendation with no stated ceiling is how an
account gets restricted.
INVESTIGATE
Something moved and you do not know why.
REQUIRED: the specific question to answer, and where the
answer would come from. "Look into it" is not an entry.
HOLD
Underperforming, but not yet enough data to act.
REQUIRED: how much more data is needed, and by when it will
exist. A hold with no review date becomes a permanent hold.
TEST
A hypothesis worth one experiment.
REQUIRED: the one variable changing, the success criterion
set in advance, and the duration.
ONE VARIABLE. Change three and the result tells you nothing
about which one worked.
RULES
No finding appears in two lists.
MAXIMUM THREE ENTRIES IN SCALE and THREE IN TEST. A week with
seven tests running produces seven unattributable results.
Everything in INVESTIGATE has a named owner, or it does not go
on the list.
IF A CAMPAIGN IS FAILING AT STAGE 1 of the funnel map, it
cannot be in SCALE regardless of its later numbers.
OUTPUT the four lists, ranked within each by effect on the
headline number.
The required fields are what make this a queue rather than a summary. A SCALE entry with no ceiling, or a TEST with no success criterion set in advance, is a note rather than a decision.
Three scale and three test is the cap for a reason. More than that and nothing gets properly attributed, which means next week's analysis cannot tell you what worked.
Step 6: write the weekly brief
---
name: 06-weekly-brief
description: The document someone actually reads. Under a screen.
Medium effort.
---
STRUCTURE, in this order:
## The one thing
The single most important finding, in one sentence, with the
number.
## Data quality
One line. Clean, usable with exclusions named, or not usable.
If anything was excluded, say which and why.
## The funnel
The failing stage only. Not all five.
The absolute beside every percentage.
## Decisions for you
The four lists from step 5, with the required fields.
Ranked by effect on the headline number.
## What changed since last week
Including whether last week's one change produced anything.
If it is too early to tell, say when it will not be.
## This week's one change
ONE. With the reason and the success criterion.
## What I could not assess
Any stage or dimension where the data was missing.
RULES
UNDER ONE SCREEN. A brief that needs scrolling gets skimmed,
and the thing at the bottom is the thing that mattered.
EVERY PERCENTAGE SITS BESIDE ITS ABSOLUTE.
EVERY METRIC PAIRS WITH AN ACTION. If a number cannot answer
"so what do I do", cut it from the brief.
BANNED: impressions or reach as a headline number, any metric
with no action attached, and the word "optimise" without
naming what changes.
NEVER MORE THAN ONE CHANGE A WEEK. Change three and next week's
brief cannot tell you which one worked, and the week after that
you are guessing.
IF THE HONEST ANSWER IS "nothing significant moved", write that
in one line rather than padding. A brief that finds something
every week is a brief that is inventing findings.
The "what I could not assess" section is what makes the brief trustworthy. A report that never admits a gap is one where you cannot tell the difference between a clean stage and an untracked one.
"Nothing significant moved" is a valid brief. Most weeks that is the truth, and saying it is how the weeks where something did move get taken seriously.
The calculation checks
Run these against any number before it goes in the brief.
[ ] Does every stage total reconcile with the stage before it?
[ ] Does the export total match the platform's own total?
[ ] Is every percentage shown beside its absolute?
[ ] Is every cohort comparison at equal maturity?
[ ] Is any sample under 30 on either side of a comparison?
[ ] Has any step been counted twice for one person?
[ ] Does the attribution window match the actual cycle length?
[ ] Are segment names consistent across every row?
[ ] Is spend allocated to the period it was spent, not booked?
[ ] Does any stated ceiling match the real one?
Check 7 is the quiet one in B2B. A 7-day attribution window on a 60-day cycle reports that almost every campaign failed, and the analysis will be confidently wrong in a way that is hard to spot.
Check 10 applies specifically to outreach. A scale recommendation that would take an account above roughly 20 connection requests a day, account-wide rather than per campaign, is not a scale recommendation. It is a restriction waiting to happen.
Running all six as one job
Take ownership of this from start to finish.
Read campaigns/export.csv, campaigns/targets.md and
campaigns/events.md.
Run steps 1 to 6 in order. Each reads what the previous wrote.
STOP AND WAIT at two points only:
After step 1, if the verdict is NOT USABLE
After step 5, before the brief is written
Between those, keep working. Make intermediate decisions and
note them rather than checking in.
HARD RULES
- Never proceed past step 1 on a NOT USABLE verdict.
- Never compare cohorts at unequal maturity.
- Never report five funnel stages when the first is failing.
- Never recommend a volume above a stated ceiling.
- Never more than one change in the brief.
- If a stage cannot be assessed, name it rather than
estimating it.
End with the brief, and a one-line note on anything you were
not confident about.
Two stops. The data verdict and the decision queue are the only two points where a human needs to be in the loop. Everything else is reversible by re-running it.
The Worked Example, End to End
One week of a real-shaped B2B outreach operation, run through all six steps.
The numbers are illustrative, not measured. Run the workflow on your own export before trusting any benchmark in here.
The inputs
An agency running four campaigns across two LinkedIn accounts, plus one paid campaign.
TARGETS
Headline: meetings booked per month, from 12 to 20, by Q end
reach to accepted 28%
accepted to replied 12%
replied to meeting 25%
Daily ceiling 20 requests per account, ACCOUNT-WIDE
Kill criteria any campaign under 15% acceptance
after 200 sends, reviewed day 21
EVENTS
step_1 connection accepted. Fires once. Auto-detected over
14 days.
step_2 first reply received. Can fire twice if they reply to
two messages. KNOWN ISSUE.
step_3 meeting booked. Tagged manually. KNOWN GAP: tagged
inconsistently before 1 Sept.
step_4 closed won. From the CRM.
Attribution window: 90 days.
Two things already visible in the inputs. step_2 can double count, and step_3 has a tagging gap before a date. Both get handled in step 1 rather than discovered in step 4.
Step 1 output: the data check
DATA CHECK, week ending 3 October
ISSUE WHERE ROWS SEVERITY
Missing creative_or_angle Campaign C 31 Medium
No rows for 28 Sept All campaigns - High
step_2 exceeds step_1 Campaign B, 2 days 2 High
Segment named 3 ways "SMB"/"smb"/"Small" 44 Low
step_3 blank pre-1 Sept Campaigns A, B 60 Known
RECONCILIATION
Export total sends: 1,240
Platform reported: 1,240 MATCH
FINDINGS
28 Sept missing across every campaign. Not a pause, an export
gap. The week is 6 days, not 7, and every daily average in
this analysis reflects that.
Campaign B shows step_2 above step_1 on two days. This is the
known double-count in events.md: two people replied to two
messages each. Deduplicated to 4 unique repliers.
Campaign C has no angle recorded on 31 of 44 rows. Step 4
cannot assess creative performance for Campaign C.
VERDICT: USABLE
Proceed with Campaign C excluded from angle-level analysis.
All daily averages calculated on 6 days.
The missing day is the finding people miss. Without catching it, every daily average is understated by a seventh, and a campaign looks 14% worse than it is.
Step 2 output: the funnel
FUNNEL, week ending 3 October, 6 days of data
STAGE ACTUAL RATE TARGET STATUS
Sends 1,240 - -
Accepted 223 18.0% 28.0% BELOW
Replied 31 13.9% 12.0% above
Meetings 7 22.6% 25.0% slightly below
Closed 1 14.3% - -
DIAGNOSTIC: stop at the first failure.
STAGE 1 FAILS. Acceptance at 18.0% against a 28% target.
223 accepts from 1,240 sends. At target that would have been
347, so this stage cost roughly 124 accepts this week.
The later stages are reported above for completeness and are
NOT the finding. Reply rate at 13.9% is above target and
meetings are close to it. Tuning either would be work on a
stage that is already fine.
WHERE TO LOOK, in order:
1 Is a connection note attached on every campaign?
2 Is it specific and dated, or could it go to anyone?
3 Does it contain an ask?
4 Is the list actually the ICP?
5 Is the profile optimised?
Check 5 is the one nobody runs. Every request drives a profile
visit, and the accept is frequently lost there rather than in
the note.
NOT ASSESSED: closed-won. One deal is not a rate.
Step 3 output: the cohorts
COHORT COMPARISON, at equal maturity
COHORT SIZE DAY 7 DAY 14 DAY 30 DAY 60
Aug week 1 780 1.8% 3.1% 4.4% 5.1%
Aug week 3 810 1.9% 3.3% 4.6% -
Sep week 1 840 2.1% 3.4% 4.5% -
Sep week 3 790 2.0% 3.5% - -
Oct week 1 1,240 1.6% - - -
COMPARABLE ACROSS ALL COHORTS: day 7 only.
NOT COMPARABLE: everything beyond day 7 for Oct week 1.
THE FINDING AT EQUAL MATURITY
Day 7 conversion to meeting has held between 1.8% and 2.1%
across four cohorts, then dropped to 1.6% in the newest.
That is a real drop at equal maturity, not a time artefact.
It is also only 0.2 points below the lowest previous cohort,
on a larger sample, so it is worth watching rather than
acting on.
WHAT THIS RULES OUT
Someone looking at total conversion would see Oct week 1 at
1.6% against Aug week 1 at 5.1% and conclude performance had
collapsed. It has not. Aug week 1 has had 60 days to convert
and Oct week 1 has had 7.
That comparison is the single most common error in campaign
reporting and it produces panic rather than insight.
TIME TO CONVERT
Every cohort reaches roughly 3.3% by day 14 and 4.5% by day
30. The curve shape is consistent, which means the cycle
length is stable even where the entry rate moves.
That last paragraph is the useful output. A stable curve shape with a moving entry rate means the problem is at the top of the funnel, which agrees with step 2.
Step 4 output: the changes
MOVEMENTS, ranked by effect on meetings booked
1 CAMPAIGN A ACCEPTANCE
24% to 15%, on 420 sends versus 390 last week.
Normal weekly variation for this campaign: +/- 3 points.
This is a real change.
EFFECT: roughly 38 fewer accepts, so 1 to 2 fewer meetings.
2 ACCOUNT 2 ACCEPTANCE
26% to 19%, across both campaigns on that account.
Normal variation: +/- 4 points.
Real, and it is account-level rather than campaign-level.
3 SEGMENT: SMB
Acceptance 21% against 15% for mid-market, on comparable
samples. Held for three weeks.
EFFECT: consistent, not a change. Context rather than a
finding.
4 CAMPAIGN D, PAID
Cost per meeting 340 to 410.
Sample: 3 meetings versus 4. TOO SMALL TO ASSESS.
CONSEQUENCES, not discoveries
Campaign C volume up 60%. We raised its daily limit on
Monday. This is the expected result of a decision, not a
finding.
EXCLUDED
Campaign C angle analysis. No angle recorded on 31 of 44
rows. See step 1.
Movement 2 is the one that matters and it is only visible because the dimension list includes account. Both campaigns on one account dropping together points at the account or the profile rather than at either campaign's copy.
Step 5 output: the decision queue
SCALE
SMB segment on Account 1.
Current: 21% acceptance, 3 weeks stable, n=620.
Target: 28%.
CEILING: Account 1 is at 14 of 20 daily requests across two
campaigns. Headroom is 6 a day. Account-wide, not per
campaign.
ACTION: move 6 a day from mid-market to SMB. Do not raise
the account total.
INVESTIGATE
Account 2 acceptance dropped 7 points across both campaigns.
QUESTION: did the profile change, did the list source
change, or is the account restricted?
WHERE: account health check, then the profile, then the
list source for both campaigns.
OWNER: [name]
BY: Wednesday
HOLD
Campaign D paid cost per meeting.
NEEDED: 8 more meetings for a readable rate. At current
volume that is roughly 3 weeks.
REVIEW: 24 October.
TEST
Campaign A connection note.
ONE VARIABLE: the note only. Nothing else changes.
HYPOTHESIS: the current note is generic and acceptance fell
when the list source widened.
SUCCESS: acceptance back above 22% on 200 sends.
DURATION: 10 days.
NOT IN SCALE
Campaign C, despite 60% volume growth. It is failing at
stage 1 with 16% acceptance, and a stage 1 failure cannot be
scaled regardless of later numbers.
Step 6 output: the brief
# Campaign brief, week ending 3 October
## The one thing
Acceptance fell to 18% against a 28% target, costing roughly
124 accepts and 1 to 2 meetings this week.
## Data quality
Usable. 28 September is missing from the export, so all daily
figures are on 6 days. Campaign C excluded from angle analysis,
no angle recorded on 31 of 44 rows.
## The funnel
Stage 1 is failing. 223 accepts from 1,240 sends, 18.0% against
28%.
Reply rate 13.9% and meeting rate 22.6% are both at or near
target and are not the problem this week.
## Decisions for you
SCALE 6 daily requests from mid-market to SMB on
Account 1. Headroom is 6. Do not raise the total.
INVESTIGATE Account 2 dropped 7 points across both campaigns.
Check the account, then the profile, then the list
source. [owner], Wednesday.
HOLD Campaign D paid. 3 weeks to a readable rate.
Review 24 October.
TEST Campaign A note. One variable. Success is 22%
acceptance on 200 sends, over 10 days.
## What changed since last week
Last week's change was raising Campaign C's daily limit.
Volume up 60% as expected. Acceptance unchanged at 16%, so the
change produced volume rather than quality. That was the
expected trade and it is worth noting Campaign C still fails
at stage 1.
## This week's one change
The Campaign A connection note.
Reason: the only stage failing is acceptance, and Campaign A
accounts for the largest part of the drop.
Success: above 22% on 200 sends by 13 October.
## What I could not assess
Closed-won. One deal is not a rate.
Campaign C creative performance. No angle recorded.
Anything beyond day 7 for the October cohort.
One change, not four. The queue has four entries but only one is a change to the system this week. The others are a reallocation inside an existing ceiling, an investigation, and a wait.
Three things to remember
- Step 1 is not admin. A missing day, a double-counted event or a mismatched total will produce confident analysis of the wrong thing, and you will not notice.
- Compare cohorts at equal maturity. Comparing a 7-day-old cohort's total conversion against a 60-day-old one makes performance look like it collapsed when nothing changed.
- One change a week. The queue can have four entries. The brief has one change, or next week's brief cannot tell you what worked.
The analysis finds it. Prosp supplies the numbers 💜
