use case · analyze
When CPA climbs, Neo finds the root cause and proposes one fix.
Neo first reads the change history and the brand’s events, then drills down from the store to the channel, the campaign, the metric and its trigger. One sentence names the cause, one fix follows, with the date to judge it.
4,000 advertisers and agencies already use Creaboost.
A verdict, a root cause, one fix.
Camille asks in the chat, the way she would ask an expert. Neo reads Meta, Google Ads, Shopify sales and Maison Lune’s memory, then answers in the order of a diagnosis.
Why has my CPA been climbing for a month?
Camille Durand · Tue, Oct 13 · 09:12
Neo · 09:21
- change history
- memory · events
- 26 weeks · weekly view
- conversion lag corrected
diagnosis of Oct 13 · performance_diagnosis · Sep 12 → Oct 11 · Meta, Google Ads, Shopify · 7-day click · Google sync 06:10
The rise comes from Meta prospecting: concept 04 has been fatiguing since the week of Sep 21.
Blended CPA, computed on Shopify orders, went from $33 to $39 (Sep 12 → Oct 11, against the previous 30 days). Spend barely moved; orders fell because paid traffic fell. A dip this size calls for a single fix.
- blended CPA
- $39
- $33 · +17%
- ad spend
- $121,000
- +4%
- Shopify orders
- 3,110
- −10%
- store sessions
- 183,000
- −9%
root causeProspecting France is underperforming because its CPA rose 22% in 7-day click, driven by a 19% drop in link CTR, which was caused by the fatigue of concept 04, “before/after,” starting the week of Sep 21, as shown by its frequency (2.1 → 3.8) and its hook rate (−31%) three weeks in a row, at steady spend.
the fix
- 01
Three new hooks on concept 04’s body, in its ad set, and nothing else for now.
rule A moderate dip gets one isolated fix, so its effect can be read. A fatiguing winner keeps its body and changes its first 3 seconds: Meta counts a new hook as a new ad.
check Oct 27 · hook rate > 26%- expected impact · blended CPA back toward $36
- time to see it · 7 to 10 days
- who acts · the creative team, then Camille
- watch · Prospecting France link CTR
the drill, level by level
- storesessions −9%
Sessions are down; conversion rate and AOV are holding.
Revenue = sessions × conversion rate × AOV. Conversion rate stays at 1.7%, mobile included, despite the new checkout of Sep 22. The drop comes from traffic.
- conversion rate 1.7% · steady
- checkout of Sep 22 · ruled out
- channelsMeta
Meta carries 78% of spend and the change; Google stays within its usual swing.
Google Ads is read with brand and non-brand apart. Brand CPC is up 22% since Oct 2, on 4% of spend: too small to explain the rise.
- Google · within its 90-day swing
- Brand CPC · 4% of spend
- campaignProspecting France
Of the top five Meta campaigns, only one moves beyond its usual swing.
Prospecting France: CPA +22% since the week of Sep 21, when its usual week-to-week swing is 8%. No budget, audience or optimization event change on that date.
- CPA +22% · usual swing 8%
- catalog and retargeting · steady
- metriclink CTR
CPA = CPM ÷ (1,000 × CTR × conversion rate): CTR carries the rise.
Link CTR −19%, post-click conversion rate steady. CPM is up 11% since Oct 1, as it was in the same weeks last year: a seasonal rise that needs no fix.
- link CTR −19%
- CPM +11% · seasonal
- triggerconcept 04
Concept 04, “before/after,” shows the fatigue signature.
It carries 38% of Meta spend. Frequency up, reach flat, hook rate and link CTR down, while its spend holds. The pieces it shows are in stock.
- 38% of Meta spend
- frequency 2.1 → 3.8
- hook rate −31%
- measurementruled out
The attribution change of Sep 24 only explains the gap with the CPA Meta reports.
Prospecting France moved from “7-day click and 1-day view” to “7-day click.” Read in 7-day click everywhere, the rise is still all there. Meta purchases track Shopify orders day by day: the pixel is healthy.
- attribution changed Sep 24
- healthy pixel
Every level read over 26 weeks in a weekly view; the last two days corrected with the 7-day click ÷ 1-day click ratio stored in memory.
not checkedGoogle Ads change history can’t be read here: Neo dated the changes from creation dates and the events in memory. No target CPA in Economics, so the reference is the median of the last three months, $35.
Saved in Docs · diagnosis · Maison Lune · 10/13/2026
- Export to Google Docs
- Schedule the performance watch
how Neo gets there
Three rules before reading a single number.
- memory · events
First, what changed.
Neo reads each platform’s change history (budgets, bids, statuses, new ads, tracking), then the events in memory: promotions, launches, stock-outs, site releases. Many drops start on the day of a change.
- data · 90 to 180 days
A wide window, in a weekly view, corrected for conversion lag.
One day or one week on its own shows noise, so Neo reads 90 to 180 days week by week. Recent days are corrected: Google by conversion time, Meta with the 7-day click ÷ 1-day click ratio. Meta and Shopify are read live, Google Ads synced several times a day.
- method · performance_diagnosis
A drill from the store down to the trigger.
Store, channel, campaign type, campaign, ad set, metric, sub-metric, trigger: Neo only goes down where the change sits. When the cause is outside the ad accounts, it says so and stops: nothing to change in the accounts.
A diagnosis, one fix, then a watch every week.
The answer lands in the chat. Keep it in Docs, then let Neo watch: the performance watch runs a new diagnosis only when a metric moves beyond its usual swing.
Explore Docs →chat
The answer, in the order of a diagnosis.
Verdict, root cause, fix, drill, not checked. Ask a follow-up question: Neo keeps the brand’s context.
docs
The diagnosis, saved and dated.
Earlier diagnoses are kept. Shared by link, exported to Google Docs or Slides in the brand’s look.
boards
A board to track the fix.
Concept 04’s hook rate, Prospecting France link CTR and blended CPA, week after week.
tasks · slack
The performance watch, every week.
A light check of the business signal and the top five campaigns. Neo only writes when a metric moves more than its usual 90-day swing.
go further
More use cases, and the product pages behind them.
questions
What people ask us about diagnosis.
Does Neo change my campaigns?
Neo reads, analyzes and proposes. It has no write access to your ad accounts, and each fix is written for your team, who applies it in Ads Manager.
Why a single fix?
So its effect can be read. On a moderate dip, Neo proposes one isolated fix, with the expected impact, the time to see it, who acts and what to watch. On a drop that puts the business at risk, it proposes several and warns that their effects won’t be separable. It fixes without rebuilding: a rebuild resets learning.
Are Neo’s numbers the same as my dashboards?
Yes. Neo computes metrics with the formulas of the Creaboost dashboards, and every answer states the attribution window it used. Meta and Shopify are read live; Google Ads and TikTok Ads are synced several times a day, and the answer says up to when.
What if I haven’t connected Shopify?
Neo then works from ad platform conversions and lists it under what it could not check: without a connected store, there is no blended CPA, no MER and no sessions to start the drill.
Can Neo tell a real trend from noise?
Yes. A change counts when it exceeds the metric’s usual week-to-week swing over 90 days. Neo never concludes from one day or one week alone, and under thirty or so purchases, it flags the gap as noise.
Ask the question on your own accounts.
Book a demo: we connect your accounts, and Neo explains your latest CPA swing in front of you.