Wise BI Foundryown the truth.
The first of the Wise Pirates methodologies · built for retail and omnicommerce
One system that turns your customers' data into media decisions you can prove. A platform that processes the data, the Foundry, and a method that makes it speak.
Every platform tells its own story. Your business has only one. We build it from the transactional truth you already own, and we prove every decision back to it.
What it is, in one minute.
The rest of this page goes deep. If you read only one part, read this: Wise BI Foundry turns your own sales data into decisions you can trust, and feeds them straight back into your marketing.
Brings your data into one clear picture
It gathers the data from your sales, your stores and every ad platform, cleans it every night on its own, and hands it back ready to use.
One version of the truth, not five
Every platform takes credit for the same sale. The Foundry shows the one version that really happened in your business, so you stop paying twice for the same result.
Decisions you can prove
Clear answers to the questions your board keeps asking, campaigns fed by what actually sells, and every number traceable back to where it came from.
In one line: your data already knows what works. We make it say so, every night, and turn it into better marketing.
Every platform claims the same sale. Your business made it once.
Before any of the science, one simple picture. If you add up what Meta, Google and the rest report, you are counting the same order two, three, even four times. Meanwhile the one place it truly happened, your own sales, sits unclaimed.
Platforms claim the same revenue
Each attributes the same order to itself. Summing the platform reports is counting the same revenue twice, or four times, by construction.
Analytics went from census to sample
Cookie consent hides a growing slice of buyers. Deciding on analytics alone is deciding on part of the film.
Data lives scattered, reports age badly
Each tool keeps its piece. Every copy-paste is a chance to be wrong and a week of delay between the question and the answer.
The question changed
It stopped being "what does the platform say?" and became "what does the customer's business say, and how do we prove it?"
The problem has a size, and it has sources.
Independent, market numbers, not ours. Every claim below was audited adversarially before it went on this page: confirmed, corrected, or refuted and dropped.
of data professionals name trust in the data as their number one concern, up from 66% a year earlier.
dbt Labs, State of Analytics Engineering 2026conversions claimed by platforms versus real orders in multi-channel programs, up to 4x by construction.
C3 Metrics Data Lab 2026overstatement of observational methods versus real experiments, documented across large RCTs.
Gordon et al., Marketing Science 2019of browsing blocks third-party cookies by default. Whoever decides on analytics decides on a fraction.
StatCounter 2026overnight swing in reported conversions when Meta cut attribution windows, rules changed twice in eight weeks.
AdBeacon 2026of data governance and analytics initiatives will fail through 2027. The discipline fails before the technology.
Gartner 2024And the upside is just as measured: platforms' own figures show −17.8% cost per result (Meta CAPI) and +5% to +17% conversion rate (Google Enhanced Conversions) from first-party signals, with up to 2.9x revenue uplift from mature first-party data (BCG and Google). The Foundry delivers this as part of the system, not as a tool on the side.
The truth lives in your transactional data.
Everything in the system measures against one fixed point: your own sales, orders and customers. The rest are witnesses and samples, useful but never the verdict.
First-party data
Your sales, orders and customers are the fixed point of the system. Everything else is measured against them, never the other way around.
Media platforms
Platform reports are interested readings. They enter the analysis after they are reconciled, never as the verdict.
Site analytics
On-site behaviour is directional and valuable, but it is never the billing, and the system never treats it as such.
Our posture: the system does not replace your tools, it disciplines them. The methodology is ours; the data is yours and works for you. On exit, the data and the definitions always stay, with a documented handover.
A factory and a discipline. One without the other does not serve.
Wise BI Foundry is the complete system. Foundry alone is the platform; the discipline is the analytical method that runs on top of it.
Foundry
The data platform: it extracts from every client source, cleans it in layers, gives it back to the platforms and serves clean data by API. It runs every night, with no hands. It is code, and it replicates per client.
The method
The analytical practice: the Ruler decided with the client, the arithmetic that closes, validation against finance, the clean test, the narrative. It is people trained to work on top of the factory.
The thesis: platform without method is expensive dashboards. Method without platform is artisanal consulting that does not scale. The value is in the complete system.
Proven foundations, assembled uniquely.
We did not invent the pieces. Each is a proven discipline with its own literature. What is new is the assembly: packaged as an agency methodology, with financial reconciliation, a frozen ruler and media activation.
Progressive data layers
Refining data in progressive layers, from raw to ready to use, is the industry standard. What the standard does not impose, and we do, is one rule: only the citable layer can be cited.
Analytics engineering
Every transformation, test and definition is versioned and travels with the data. In the Foundry it runs every night, so no number is born outside versioned code.
Warehouse-native activation
The client's warehouse is the single source and is activated back into the platforms. First-party signals lift performance, and the warehouse stays the source of truth, not a vendor's black box.
Experiments, not observation
Controlled experiments show observational methods overstate the gain three to five times. In eBay's own experiment (Blake et al., Econometrica), a paid-search return that attribution put above 1,400 percent came out at minus 63 percent once measured causally: the sign was wrong, not just the size. That is exactly why nothing scales here without a clean test.
The gap: every piece exists in the market as technology. No one packages them as an agency methodology with financial reconciliation, a frozen ruler and media activation. That is where this framework lives.
From raw source to citable data: five layers.
Now it gets technical. Five layers, each with a job: the faithful copy (Raw), the mirror (Base), the domain (Structured), the citable data (Serving) and the business reading (Reporting). One rule the industry standard does not impose: only the citable layer can be cited. A sale is a sale, whether it comes from a CRM or an ERP.
An identity, not a rebuild
One configuration file defines the client; the warehouse, the names, the access and the API are born from it. Installing on another client changes the identity, not the system.
Generic core, specific edges
The engine is the same for every Foundry. Each client's reality, which sources, which taxonomy, lives in adapters and configuration, never in the core.
Total isolation per client
Each Foundry is an independent instance: data, credentials and infrastructure never cross between clients. Everything is rebuildable from the repository.
Proven in production
Two live Foundries today for two clients that look nothing alike: an ecommerce with an online-sales CRM, and a physical retailer with an ERP and stores. Completely different sources, the same machine.
The night it all happens: wakes, runs, delivers, sleeps.
The processing infrastructure is born at the appointed hour, installs and checks itself, extracts every connected source in parallel, rebuilds the layers, activates, and then disappears until the next night. You pay the night, not the day.
The warehouse is not an archive. It gives data back to work.
Build everything, activate by destination. Each output only fires to the platforms the client has connected. Turning one off breaks nothing; turning one on needs no new code.
First-party, from the base
Segments built from the transactional data, behaviour, category, recency, lifecycle, with identifiers protected by hashing, prepared in each platform's format and loaded automatically.
Ranked by what actually sold
Every product is ranked against the others by what it truly sold, with store sales counting alongside online. Two independent rankings, one per reality, are written back into the ad catalogue every night, so a product that holds up in the stores stops being bid as if only the site existed. This is what makes campaigns know what the business knows.
The future re-enters
Revenue-per-channel forecasting models, continuously validated against the realized, and the result flows back into the layers as just another source.
Served to whoever decides: for humans, a visual query builder with no SQL; for tools, stable sources with classified metrics; for AI agents, an llms.txt that describes the data live, so the warehouse is ready for the age of agents. The API auto-discovers whatever serving layer exists, so a new source appears in the API the next day, with no new code.
Every principle exists twice: in code, and in discipline.
The factory makes the data reliable without daily effort. The discipline makes sure what you conclude from it is useful truth. Four principles, in order.
Define the metrics first. Only then measure.
- A per-client catalogue of definitions: what is revenue, a new customer, a cost, written before the first calculation
- Dated, signed, versioned and frozen between releases, so the same question always has the same answer
- Stored inside the layer contracts, not in a document on the side
No number enters a report without a path back.
- Every value is born from versioned transformations on the warehouse, no copy-paste
- Re-derivable in minutes, from source, weeks or months later, by anyone on the team
- The arithmetic closes; a step that fails stops and warns, never publishes as complete
A number without a second source is a well-formatted opinion.
- Revenue against the client's finance, cost against platform billing, behaviour against the transactional
- Differences chased to their cause: order state, window, tax, taxonomy. "Close enough" is not a cause
- Measurement gaps are quantified and declared, not left to contaminate in silence
The analysis only ends when it becomes a campaign.
- A pattern with media value becomes a first-party audience the factory builds and loads
- A clean test proves it: one variable, criterion fixed before, verdict read in the transactional
- Only what proves scales, and the results return to the warehouse to sharpen the next reading
This is the division of labour: the factory guarantees the data is reliable; the discipline guarantees the conclusion is true.
The cycle almost no one closes.
Agencies report from the platforms outward: interested numbers, summed without reconciling, that die in a monthly slide. We close the cycle inside the business, in code.
They report outward
Interested numbers from the platforms, summed without reconciling, that die in a monthly slide and leave nothing behind.
We close inward
Audiences built from the base, sales returning to the warehouse, the reading sharpening at every turn, all in the client's own system.
"We don't sell dashboards. We sell, and prove, decisions with origin."
This is the ledger no dashboard keeps: the loop that turns your data into media and media back into data. It is our own methodology, running in production today, not a diagram on a slide.
Nothing scales until a clean test proves it.
The clean test is the gate between a reading and more money invested. It moves one variable at a time: we change the audience and keep the creative the same. The decision metric, the window and the win threshold are fixed before it starts, and the verdict is read in your transactional data, never in a platform report.
Change the criterion midway and the test is void. A three-day winner is noise in a hurry. Only what proves scales: the clean test is the only door between a reading and more money invested.
The machinery, for the people who will audit it.
This is the far end of the page. You do not need any of it to decide, and most readers stop before here. It is written for the data scientist and the engineer on your side of the table, the only people who can verify it, because they always ask and we always have the answer. Only what withstands verification is here.
Hierarchical, nothing fails in silence
A main orchestrator, a controller per stage, one process per source. If a source fails, the others carry on. Every extraction, stage and transformation is logged with its duration and result, and a step that fails stops and warns instead of publishing as complete.
Exists only while it works
The processing infrastructure is born for the run and gone after it. Each run is reproducible from zero, so last quarter’s numbers can always be rebuilt exactly as they were.
Contracts, not a document on the side
Definitions live as versioned contracts next to the data, not in a separate file. The definitions layer also decides what may be summed: reach, frequency and averages take a maximum, never a sum.
Discovers the data, never a hand-written list
The API has no hand-written table list. It discovers whatever citable layer exists that night, classifies each column as a dimension or a metric, aggregates inside the query, authenticates by key and logs every request. A new source appears the next day, with no new code.
Person-safe by construction
Identifiers are normalized, then hashed before they are stored: a phone in international format for one platform, digits only for another. Personal data is never stored in clear in the citable layer, and the API never returns it. Data does leave every night for activation, that is the point, and it leaves hashed.
Documentation that cannot go stale
The documentation for AI agents is generated from the same discovery on every run. It is never written by hand, so it is never out of date, and it never carries a key inside.
We treat a media decision the way a lab treats a trial: the metric, the window and the win threshold are written before a single euro moves, and changing the criterion mid-flight voids the test. Most of the market calls a dashboard reading measurement. This is a system your own team can audit, line by line.
Two omnichannel retailers, opposite realities, the same machine.
Two live Foundries today, for two clients that look nothing alike. We do not publish client numbers here, on principle: the answers exist and they are the client's. What we can show is the proof of the method working.
An ecommerce retailer
- A full semester of revenue reconciled to the cent against the client's back office, with the analytics gap from cookie consent quantified and now a permanent part of the reading.
- First-party audiences rebuilt from the transactional base every night and activated into media, with identifiers protected by hashing.
- The first clean test of the cycle beat the open public with the same creative, and was never turned off.
Unequivocal proof: a value questioned by the client was re-derived from source six days after delivery and reproduced to the detail, with the query in plain sight.
A physical retail chain
- Completely different sources from an ERP and store network, ingested through the same generic layer contracts: a sale is a sale, from a CRM or an ERP.
- Store and online performance unified into one product signal that travels back into the ad catalogue, so campaigns know what the business knows.
- Monthly closing reconciled across agency, platforms and finance, with every difference explained by its cause, not waved away.
Unequivocal proof: two Foundries, completely different fonts of truth, run on the same engine, differing only by an identity file and adapters.
Both cases run on the same principle: the verdict is always read in the transactional data, never in the platform report. Client identities and specific figures are shared under engagement.
The questions your board keeps asking, answered with your own data.
Strip away the architecture and this is what it is for. Every one of these is answered from your transactional truth, with origin, ruler and proof, never from a platform report.
What does a new customer really cost, and how fast does it pay back?
Answered from the transactional data, reconciled with finance.
Average order value fell: was it price, discount, or the mix that changed?
Decomposed against the ruler, not guessed from a dashboard.
Which categories bring new customers, and which bring margin?
Read from owned sales, by category and lifecycle stage.
Can spend go up without losing efficiency? How far, and by what rule?
Answered with guard-rails defined in the ruler before scaling.
How much of what the platforms claim actually happened in the business?
Reconciled claim by claim against the transactional anchor.
Which customers justify their own audience, and what changes when it goes live?
Proven with a clean test, verdict read in the transactional.
The answers exist and they are yours, which is exactly why they are not on this page. The system guarantees each one comes out with origin, ruler and proof.
Dashboards show. They don't prove, and they don't decide.
A dashboard is a window. This is a factory, with a team deciding on top of it. The table makes the difference concrete.
| Capability | Dashboards / BI tools | Wise BI Foundry |
|---|---|---|
| Platforms side by side, with history | Yes | Yes |
| Reconciliation against the client's finance | No | Yes |
| A written, versioned, frozen Ruler per metric | No | Yes |
| Numbers re-derivable from source, auditable | No | Yes |
| First-party audiences built from the base and activated | No | Yes |
| Data served by API to humans, BI and AI agents | No | Yes |
| Reading interpreted and validated by humans | No | Yes |
And enterprise CDPs? They activate, but in a black box, with lock-in and the data at the vendor. Here the warehouse is the client's, the method is auditable, and any piece is replaceable.
A fixed rhythm, proof against haste.
The cadence is not bureaucracy. It is what separates measuring from watching.
The factory runs
Extract, transform, activate. By morning yesterday's data is already clean.
Signals and deviations
What moved, what needs action, read against the reconciled base.
Reconciled report
Numbers with origin and proof, reconciled with the business.
Deep-dive and Ruler review
Thematic deep-dive and a Ruler review, with a version note.
The Ruler only changes at the quarterly milestones, with a version note. A trend only exists when the measurement does not move. Most measurement failures in the market are definition drift, not data.
Five phases, one order.
No one reads before validating, no one validates before defining, no one defines before connecting. Skipping phases is buying conclusions you later return.
The Foundry instance is born
A single configuration, the client's infrastructure and warehouse. A reconcilable-from-zero instance.
Connect what the client has
The sources the client has connect by configuration; the layers fill every night.
The catalogue of definitions
The definitions decided with the client and stored in the layer contracts.
The past reconciled until it closes
History rebuilt and reconciled with finance until it closes: the foundation of the reading. A client with no media stops here, with full value.
The cycle starts turning
Reports with proof, audiences, clean tests: the loop starts and does not stop.
A methodology of our own.
This is not a way of using a tool. It is a proprietary system, built, tested and running in production, and it belongs to Wise Pirates.
The methodology, the layer contracts and the operating playbook are a proprietary system, owned by Wise Pirates and protected under contract. What is public is the shape of the system and its principles; the detail is shared with clients under engagement.
And the knowledge accumulates: every analysis, rule and test is logged in the client's own system. The agency can change people; the client never loses the memory.
The data foundation the other frameworks stand on.
Wise BI Foundry is the foundation the rest build on. A framework that multiplies the others, because they all need the same thing: the transactional truth, clean and reconciled.
Tell us your sources and your questions. We will build the foundation.
The Wise BI Foundry, answered.
What is the Wise BI Foundry?
Why can't we just trust the platform dashboards?
What is the Ruler, and why does it matter?
How is a clean test different from normal A/B testing?
What are the five layers of the Foundry?
Do we need all our media channels connected to use it?
Who owns the data and the code, and what happens if we leave?
Is the framework proprietary?
Where do the market numbers on this page come from?
Your data already knows the answer.own the truth.
Foundry processes the data, the method makes it speak, and every insight comes with an action. Tell us your sources and the questions your board keeps asking.
Start with Wise BI Foundry →See all methodologies →The warehouse, the data and the definitions are always yours, with a documented exit protocol and a licence to the instance code. Every number is re-derivable from source. You are never locked in.
