Services
Two service lines, one method
Installing a tool can make the numbers add up, but it will not tell you whether those numbers are right. We use a method that tells you what is normal and what is actually broken — and, once the data holds, how to stop redoing the same things by hand.
The frame
The data chain
Every service sits on one of four links. Break one and everything downstream fails, however solid it is. Marketing is the most visible symptom, not the only one: the same broken chain hits product, retention, inventory and management control.
/ 01
Collection
The data never arrives: consent, ad blockers, browsers restricting cookies, native apps, ERPs and CRMs generating data nobody collects.
/ 02
Structure
The data stays where it was born: every system keeps its own piece, no owned history, nothing gets joined up.
/ 03
Reading
Nobody actually reads it: different numbers for the same fact depending on who opens which tool.
/ 04
Decision
Decisions run on perception: budget, restocking and priorities moved towards whatever looks like it is working.
/ 01
Measurement
Measurement
"Which of my numbers is the real one?"
Meta says one thing, GA4 another, the ecommerce platform or the ERP a third. Part of that gap is normal: the work is saying how much of it, and finding the real fault hiding inside the rest.
Entry point: the measurement diagnosis.
Measurement diagnosis
Inventory of active sources, real orders compared against recorded transactions over 90 days, the gap split into normal and anomalous with the technical cause of each anomaly, and a first threshold hypothesis. Ten working days, one document, and the freedom to do nothing else.
Implementation
Priority anomalies fixed, server-side architecture with deduplication and consent routing, a documented event schema, reconciliation against the source of truth on orders, a defined threshold for every pair of sources and a 30-day verification cycle.
Offline reconciliation
For anyone generating the lead online and closing elsewhere: automotive, travel, B2B. Linking CRM or ERP to the ad platforms, offline conversion uploads, match rate measurement, lead → sale reporting per campaign.
Pipelines and data engineering
Ingestion from heterogeneous sources, modelling in a versioned transformation layer, data quality checks, orchestration and failure alerts. For non-standard environments: headless, custom, multi-platform.
Analytics Department
Ongoing cover: continuous data quality monitoring, monthly reports on deviations from the thresholds, fixes and maintenance included when platforms change the rules, regular check-ins with your internal contact. The data team you do not have to hire.
Training on the method
For teams and agencies that need to verify data work without depending on us: which sources to compare, how to read a gap, how to tell normal from broken. The procedure stays documented with your team.
/ 02
Automation & AI
Automation & AI
"How do I stop redoing the same things by hand, and how do I decide better?"
Every company produces data daily — orders, customers, products, communications — and in most cases it lives in separate silos. The result is time lost collecting information that already exists, and decisions taken in the dark.
Entry point: the business process assessment.
Business process assessment
A map of what is actually there, online and offline: analytics, advertising, ecommerce, CRM, ERP, recurring manual processes, archives and flows between systems. Output: digital maturity, high-impact use cases, timing and expected return for each intervention. The report is yours and stays valid with any supplier.
Reporting and dashboard automation
Ingestion pipelines from the sources in use, metrics defined once and valid for everyone, dashboards that refresh themselves and a report that distributes itself. With freshness checks: if a source stops updating, the report says so instead of showing an old number as if it were new.
Recurring process automation
Identifying high-volume, low-cognitive-value processes, checking compatibility and estimating the return before starting, then automated flows with human supervision only where a decision is genuinely needed. Testable pilot in 4–8 weeks.
Customer communication automation
A conversational assistant on web chat, WhatsApp or email, built on real data — catalogue, order status, internal procedures — with escalation to a human and context handover. Complex cases stay with people.
AI assistant on your knowledge base
Manuals, procedures, contracts, past quotes: the team asks questions in plain language and gets answers with the exact source reference. OCR on scans, granular access control, GDPR compliance.
Decision system
From marketing to management control: shared metric definitions, a margin model built on your real data, an executive dashboard and a monthly reading ritual. It answers the question no marketing system answers: what does a product, a customer, a channel actually margin.
AI training for teams
Interactive sessions, not lectures: what AI can do for your specific sector, examples built on processes the team already runs, materials that stay with the company. On longer paths, a lab on a real case.
Where the two lines meet
Data warehouse + control centre
As long as your reference number lives inside GA4 or inside Meta, it is neither yours nor a reference: it is one interested tool's reading. The warehouse is where that number becomes yours, and the control centre is where you look at it.
It makes sense in three specific situations, and it is fair to say so: multi-store or multi-market, where the consolidated total exists nowhere; conversions that close off-site; or when a business intelligence solution has already been evaluated and dropped over per-seat licence costs.
- +Data warehouse on BigQuery with versioned modelling: ingestion from ad platforms, analytics, ecommerce and ERP
- +Owned history beyond the native platform limits, which retain between 7 days and 14 months
- +Automatic reconciliation across sources against the source of truth on orders, recalculated at every refresh
- +Control centre: one dashboard with the reference number, the gap for each source and its status against the threshold
- +An alert when the threshold is crossed, with the cause already identified — not a generic warning
- +Extensions on top of the warehouse: chat with data, monitoring agents, predictive models on BigQuery ML
The method
APIC: four phases, and you can stop after any of them
Every project follows the same path, whatever its size. Each phase produces a verifiable deliverable before moving to the next. No framework agreements, no blind annual commitments.
A
Assessment
Analysis of your real stack and processes, not a generic workshop case.
→ Findings report
P
Pilot
One single, closed problem with measurable results in 4–8 weeks.
→ First verifiable result
I
Implementation
We build what the pilot proved is needed, and nothing more.
→ Production setup
C
Continuity
Monitoring, alerts and adjustments when the platforms change the rules.
→ Ongoing retainer
Rules of engagement
They apply from day one
On every service, without exception.
/ 01
Your agency stays where it is
We do not replace whoever handles media, creative or content. We only work on the underlying data.
/ 02
We do not buy media
Whoever spends the budget cannot also be the one checking whether that spend is measured properly.
/ 03
No black boxes
Warehouse, containers, accounts, code and documentation are all in your name. Change supplier and you take everything with you: the system keeps working.
/ 04
We work with your team
Training, feedback and adaptation are part of the process, not an extra.
/ 05
Two fixed contacts
One technical lead and one operational contact, always the same people, never a rotating generalist.
It is for you if
- +You have more than one source telling you how much you sold, and at least once you have wondered which one to believe.
- +Someone in the company has to answer for those numbers in front of a partner, a CFO or an investor.
- +There is at least one recurring process eating more than four hours a week of qualified work.
- +You can give us access to the tools quickly and have someone internal who answers technical questions.
- +You accept that the outcome of the analysis may be "your system works, no intervention needed".
It is not for you if
- −You want the ad platform numbers and your ecommerce numbers to match: they never will, part of the gap is structural.
- −You are after a guaranteed ROAS increase. We measure and diagnose: if better data leads to better decisions ROAS goes up, but that is not a promise we can put in writing.
- −You want the work to start before an analysis: on an undiagnosed system we cannot scope the project honestly.
- −You need someone to run the campaigns too. We do not, and there is a reason for that.
- −You are looking for "AI" as a standalone project: a model built on unvalidated data produces convincing, wrong answers.
How we think
Frequently asked questions
Where do we start?
How long does a typical project take?
Do I have to take the whole path?
Do you only work with large brands?
Can you work alongside our in-house team?
What if the diagnosis says nothing needs fixing?
Let's start
Want to know where to start?
Tell us about your current stack and processes: we'll tell you which of the two doors makes sense to open first.