Marketing automation
You do not lose sales on the wrong channels. You lose them in the gaps in between.
Every email sent by hand, every abandoned cart left abandoned, every follow-up that depends on when you find the time: that is where the margin goes. It is not a budget problem, it is a structure problem. And today that structure is run by artificial intelligence: it reads the data, decides who to talk to and when, and improves on its own — under human control.
01
WHY
IT IS NOT “A NEWSLETTER WITH A TIMER”
Marketing automation is not automating sends. It is stopping the loss of customers you have already paid to acquire.
You spend to bring traffic in, and then let it evaporate through the cracks between one action and the next: the visitor who looks and does not come back, the full cart nobody follows up, the customer who bought once and whom you forget. Serious marketing automation is an ecosystem — flows, segments, triggers, scoring, content — that gets the right message to the right person at the right moment without you having to remember.
The difference today is that these are no longer “if-then” rules written once and left there: it is AI that holds the system together. It understands who it is dealing with, decides what to show them, learns from the results. It is the same shift we describe in our AI manifesto — value no longer sits in execution, which costs almost nothing, but in judgment: what matters, who to talk to, where to put the brakes. We keep that judgment in people’s hands. Since 2005, and as a benefit corporation.
“The greatest danger in times of turbulence is not the turbulence: it is to act with yesterday’s logic.”
— Peter Drucker
BEFORE AUTOMATING, FIND OUT WHERE IT BREAKS
Automating a broken process only means getting it wrong faster.
So we do not start from the tools, but from the uncomfortable questions — and AI helps us answer them with numbers, not feelings. A few real cases:
- A packaging ecommerce: Google was declaring +7%, real revenue was −1%. The problem was not the spend, it was retention: cohorts collapsed from 65% to 7%, 1,000 customers lost, €361,000 of history burnt.
- A biobank: a tracking error was distorting everything — declared cost per lead €16, real cost per qualified lead €1,041. For months you would have optimized the flows on the wrong number.
- A thermal spa: “Meta does not work” taken apart with the real data — ROAS on new customers 0.76x against Google’s 5.6x: Meta’s 2.8x was inflated by people who were already customers.
Reliable tracking, clean customer data, the funnel read one piece at a time. Because on dirty data AI does not perform miracles: it does damage faster. Do not wait for the results to tell you it is too late: by then you have already paid twice, in budget and in lost customers.
02
UNDER THE HOOD
NOT A TIMER. A SYSTEM THAT THINKS.
Behind a well-built automation there is no hand-written rule: there is an AI that reads, decides and corrects itself — and a person holding the wheel.
01 — It reads everything, in real time
In one case the AI read 25,645 orders and 87,427 rows in a day, down to the real margin product by product (32.8% on 6 million in revenue). An analyst would have taken weeks — and would not have found that the best seller returns 18.8%, while it is the single-use line at 37.8% that pays the bills.
02 — It decides and personalizes
A single control room in plain language governs 67 ad accounts: it reads spend and results live and builds complete campaigns. A thousand decisions a day — who to put in which flow, what to recommend, when to write and when to keep quiet — that nobody would take by hand.
03 — It critiques itself and improves
On a plan covering 30 brands produced by 62 AI agents, another agent redid the arithmetic before a euro was spent: break-even recalculated from 222% to 317%. AI checking AI: that is how you hold quality when the quantity explodes.
And the decision that counts stays human. AI proposes and accelerates; you confirm and choose. Brakes in the right place: this is where an agency with method differs from someone tinkering.
03
THE FLOWS
EMAIL & SMS · THE RETENTION SYSTEM
Five flows that, once switched on, recover revenue while you sleep.
They do not fire at a set time: it is AI that decides, person by person, when and with which product it is worth writing.
01
Welcome
Anyone who signs up or buys enters a path that warms them up and takes them to the first (or second) purchase. The first contact is not wasted.
02
Abandoned cart
The most profitable recovery there is: someone who was a step away from buying is brought back with the right product, at the right moment.
03
Browse abandonment
They looked and did not add to the cart? The intention was there. We catch it before it cools.
04
Post-purchase
The moment of maximum trust: that is where you build the second order, the contextual cross-sell and the review.
05
Reactivation (winback)
Dormant customers are not lost: they are the cheapest pool to bring back, if you know when and how to talk to them.
→ THE ADVANTAGE
The right message, always
Every customer gets the right message. Without you having to remember.
FLOWS SWITCHED ON · REAL RESULTS
It is not funnel theory. It is revenue that used to leave by the back door.
- A furniture and decoration ecommerce: working on cart and cross-sell, −40% cart abandonment and +60% cross-sell. The two most typical automation levers, measured.
- A B2C beauty ecommerce: conversion from 1.2% to 4.8% in six months, CPA −60% and average order value +35% — more customers, spending more, at a lower acquisition cost.
- A B2B medical devices ecommerce: one single segment, the audience with a VAT number, accounts on its own for 46% of email revenue. Proof that segmentation is where the money is.
The same customers, the same traffic. The only change is that now nobody falls through the cracks.
A DIFFERENT SHOP FOR EVERY VISITOR
Why show everyone the same products, when AI can tailor the experience to each one?
Contextual product recommendations, an internal search that understands what you mean even if you spell it badly, homepages and collections that reassemble themselves around whoever is looking, upsell and cross-sell at the exact moment it makes sense. Shopify estimates that good recommendations can lift conversions by up to 30%, and smart internal search is worth another +15–30% — but the real number is the one we find in your data.
And personalization is not only “which product”, it is also what and when you communicate. A few real examples:
- A medical ecommerce: an autopsy of 37 newsletters over a year → discovered that a deep discount is anticorrelated with revenue. The discount myth, taken apart with statistics.
- Same store: an editorial and email plan built on real seasonality indices, category by category. You send what people are about to buy, when they are about to buy it.
- A fashion accessories brand: six months of content (24 posts, 24 reels, 20 newsletters) planned on 24 months of sales data, with one stated constraint: no new photo shoots.
- The images in the flows: 4 photoreal scenes for the newsletters generated and checked for €0.30 in total, instead of a shoot costing thousands.
The same principle for customer care: an assistant that resolves most of the repetitive requests on its own — opening hours, shipping, returns, sizes — and leaves people the conversations that count. The goal is not to remove the human: it is to remove the work that does not need to be human.
Marketing automation
Automation applied to an efficient operation will magnify the efficiency. Applied to an inefficient operation, it will magnify the inefficiency.
04
HOW
METHOD, NOT AN OFF-THE-SHELF PACKAGE
It starts with the audit, the quick wins go on, everything is measured.
It starts with an audit of data and platform, the use cases with the fastest impact on revenue are chosen, the quick wins go live. First evidence in 2–4 weeks, full rollout of the flows in 6–12. No big bang: every automation is reversible, measured, under human control.
And we do not use “an AI” on its own: we have several agents working in parallel and put one to check another’s work before delivery — as when a raw advertising audit of 214,000 characters is stress-tested by a second agent and boiled down to a few operational moves. We choose the tools on your case and your CMS, with clean data upstream. Automation on dirty data is worse than no automation.
TOOLS
Klaviyo
Clerk.io
Klevu
ReConvert
Shopify Flow
Shopify Magic
ZipChat AI
Prisync
Polar Analytics
GA4 / server-side tracking
THE PROOF · AI READING YOUR DATA
Before automating, AI tells us the truth the reports hide.
- A footwear ecommerce: a complete Google Ads set — 77 keywords, 11 ads, 165 headlines, negative list — ready to import in 45 minutes (before: one or two days of a senior media buyer).
- A thermal spa: around 60 creatives reviewed one by one — the best at €1.62 per lead, the worst at €18.10 — and an already saturated audience (seen 13 times each) caught early.
- A multi-brand footwear ecommerce: 10,000 product pages localized into two languages, with a brand glossary and SEO rules locked in, to sell abroad without a translation agency.
And then, without even stopping
A contact funnel in 4 languages orchestrated across messaging and email · an online wine shop with the SEO audit turned into a sales narrative · a library of 8 advertising videos adapted at zero cost · a competitive analysis across 25 brands in fashion · a commission model that recalculates itself · a multichannel report with its own peer review inside it · a connector built on the spot when the data was not exposed · a migration verified on the live site.
Automation is worth nothing if it runs on the wrong data. It is worth a great deal when it runs on the right data — and today, it is AI that reads it.
—
EVIDENCE
TWENTY YEARS ARE NOT TOLD. THEY ARE SHOWN.
Real clients, real numbers.
Medical supplies · B2B
CFS
5,000+ orders a month on a platform that could not cope. Rebuilt and made fast, with the B2B processes automated.
+120% conversion 1.14% → 2.51% · −70% maintenance
Luxury leather goods · Florence
Via Verdi
A store built from scratch that had to sell abroad and raise the order value. It did.
+111% average order value · organic +320%
Garden · B2C
Peragashop
Opening up Germany and France while carrying enormous volumes. Result in the first quarter.
+126% revenue · ROAS 7.3
CASE STUDY · AN INTERNATIONAL FASHION ACCESSORIES RETAIL BRAND
“We felt we had unexpressed potential.”
The owner told us so. For fifteen years they had grown on a platform that had become dead weight. We rebuilt the ecommerce and worked on the purchase journey with targeted tests and microcopy, reducing abandonment without raising the advertising budget by a euro: +28% conversions, +40% on mobile.
His words at the end: “it is as if the work were the same, but we finally do not have the friction on us anymore.” With the friction gone, the most important question remained: where to put the advertising money?
CASE STUDY · ECOMMERCE DATA ANALYSIS WITH AI
From 390,000 raw orders to an investment plan, in a single session.
An international fashion accessories retail brand, around €5.3M annual revenue, 391,679 orders in its history. The objective: understand the real margin by brand and by month, and use it to decide where to invest. The problem was that the data existed, but it was scattered and misleading: prices including VAT, incomplete catalog costs, attribution inflated by two platforms, broken UTM tracking.
In a single conversation, our AI system extracted 52,772 variants and roughly 84,563 order lines from 67,734 orders. It brought prices back to net of VAT, netted the returns line by line, and validated everything to ±2% against the native reports: €5.19M net, gross margin 34.8%. Then it cross-referenced each brand’s margin with the real ROAS and found campaigns split by prestige, not by margin — with a brand at 45% margin left on low priority. Finally it exposed €268k of “phantom” revenue from double counting between Google and Meta.
Not a chatbot answering questions: an analyst that extracts, cleans, validates, cross-references three platforms, finds the hidden error and reaches the decision — with the rigour of a senior, at the speed of a chat. Final output: a monthly budget plan, two-tier ROAS targets by margin, and a 10-slide presentation ready for the owner.
—
THE NUMBERS
THE NUMBERS OF THE WORK
One session. Three platforms. One plan.
391,679
orders analyzed (the whole history)
52,772
product variants · ~84,563 lines (12 months)
€5.19M
net analyzed · validated to ±2% vs native reports
18+
brands × 12 months × 8 categories of granularity
3
deliverables: 2 Excel files + 1 PowerPoint deck
—
FINDINGS
THE FINDINGS A MANUAL ANALYSIS WOULD HAVE MISSED
Six hidden errors, found before a euro was spent.
PRICES INCLUDING VAT
Margins inflated by 22%
Without the correction, every margin would have been overstated by 22%.
DOUBLE COUNTING
€268k of “phantom” sales
Google + Meta overlap, exposed with a single-channel benchmark.
CAMPAIGNS
The right brand, in the wrong group
A brand at 45% margin left on low priority: profitability burnt in silence.
TRACKING
Broken UTM, traced to the cause
Click IDs stripped and a default UTM contaminating the data.
COST COVERAGE
42% of variants with no cost
But 99.8% on what actually sold → the analysis holds, with the framing stated honestly.
COMPLIANCE
A possible foreign VAT error
A local rate different from the Italian one: flagged, not hidden.
The value in one sentence
~400,000 raw orders across 3 platforms turned into verified margin by brand, an investment plan and a presentation. In hours, not weeks.
AI · THE WHOLE PICTURE
What you see here on ecommerce, we are doing in every corner of marketing.
Marketing automation is one piece of a much bigger change. In a few weeks we counted almost eighty real uses of artificial intelligence in our work — analysis, campaigns, content, creative, localization, operations, even software built from scratch. Not forecasts: a running account. If you want to see how far it has already gone, it is all written in our manifesto.
THE UGLY TRUTH (WHICH WE PREFER TO TELL YOU)
Not everything that can be automated is worth automating. And AI without brakes is a risk, not a superpower.
If your data is a mess, it gets put in order first — you do not build on top of it. If the volume is still small, a few essential flows are worth more than an over-engineered ecosystem nobody maintains. And over-engineering is a cost: tools that talk to each other badly, lock-in, complexity that slows you down instead of freeing your time.
The same goes for AI: left without control it gets things wrong confidently, and on sensitive data it is also a privacy and compliance problem. We put the brakes where they are needed and keep the last word with people. And if your numbers say you are better off starting small — or waiting — we will be the first to tell you. Twenty years of craft are also for saying no.
—
QUESTIONS
WHAT YOU WANT TO KNOW BEFORE STARTING
The questions we always get.
Where do we start, concretely, and how long does it take? +
An audit of data and platform, choosing the use cases, the quick wins. First evidence in 2–4 weeks, rollout of the flows in 6–12.
Does AI let me cut the advertising budget? +
Indirectly yes: by recovering sales you were already losing (carts, retention, dormant customers), you lower your dependence on paid acquisition.
How do you keep AI under control? +
Several agents checking each other, human confirmation on the decisions that count, clean data upstream, brakes where they are needed. AI accelerates; the responsibility stays ours and yours.
What are the risks and how do you manage them? +
Lock-in, dirty data, over-engineering, privacy and compliance. They are managed by choosing the tools for the case, starting from the quick wins and keeping human control over the flows.
How do you measure success? +
Numbers that count: conversion uplift, AOV, customer lifetime value (LTV), reduced CAC, carts recovered, customer satisfaction. Not vanity metrics.
Find out whether your ecommerce is really using its potential.
Or how much of it you are leaving behind.
We start from an unfiltered picture: where you lose sales today, which flows to switch on first, where AI would earn you margin and where it would not. In a few days, numbers in hand. From there the choices are yours.