Service, platforms, and three kinds of proprietary.
Two we have always run for you. Three we built ourselves.
Delivery and tech enabling, then proprietary technology, proprietary frameworks and custom data. Five elements, one proposition.
Two of these you could buy elsewhere. Three you could not.
For most of our first decade the answer to "what do you do" had two parts: we do the work, and we run the platforms you already pay for. Both are still true and still the biggest part of the business.
What changed is the three underneath. We started building the tools nobody sold us, wrote down the methods we kept re-inventing, and sourced the data the market would not hand over.
Service delivery
120 specialists who run the media, build the sites and ship the work, every day. More than fifty services.
Alone: fast execution, slow proof.
Tech enabling
Salesforce, HubSpot, Klaviyo, Google, Shopify and the rest. We implement, connect and operate them.
Alone: the same toolkit as everyone else.
Proprietary technology
Software we wrote, running in production. Not a licence we resell, not a wrapper with our logo on it.
Alone: a clever read nobody acts on.
Proprietary frameworks
The methods, written down and testable. Each keeps a record of what was claimed and what happened.
Alone: a very tidy slide.
Custom data
Market and competitive data we go out and source ourselves, engineered into something usable.
Alone: a warehouse nobody queries.
Most agencies have an AI slide. We have products in production.
A fair objection: what we own is our interest, not yours. Agreed. It earns a page only because it changes three things you do care about: what we can answer, how fast, and whether you can check it afterwards.
One question. Five layers. One answer you can defend.
The elements are not a menu. A real question travels through all of them, and each adds something the next cannot get alone.
The data makes the question answerable. The framework decides what would count as an answer. The technology reads at a speed a human cannot. The crew makes the call. And it lands in the platforms you already run, because an answer that never reaches an account is an opinion.
Three real questions, three routes through the same five layers. The named assets above are the ones that actually carry each one.
Take one layer away and the question still gets an answer. It just stops being one you can take to a board.
The largest element is still the work itself.
Everything on this page sits on top of a full-service operation. More than fifty services, grouped into the areas below, run by 120 specialists in one house rather than assembled from a freelance network.
This is what What we do covers in full. It is listed here because the other four elements are worth very little without it: somebody still has to run the media, build the site, write the query and ship the work.
The proprietary layers were built inside this operation, not beside it. Every product on this page started as a problem one of these teams was already being paid to solve, which is why they arrive as capability rather than as software you have to adopt.
We build our own. We also run yours.
This part is not proprietary and we are not going to pretend it is. It is the licensed stack most companies already own and under-use, and operating it well is a large part of what we get paid for.
- Google Ads
- Google Marketing Platform
- Meta
- TikTok
- OpenAI Ads
- YouTube
- X
- OpenAI
- Anthropic
- Gemini
- Perplexity
- Mistral
- Salesforce Marketing Cloud
- Salesforce Sales Cloud
- Salesforce Data 360
- HubSpot
- Klaviyo
- Shopify
- WordPress
- Webflow
- Adobe
- Lovable
- Google Cloud
- BigQuery
- Looker Studio
- Power BI
- Whatagraph
- Google Workspace
- Google for Education
Whatever else is already in the building. We certify where certification exists, and say so where it does not.
Operated by the people who work in them every day.
Nothing above is part of our proprietary stack. It belongs here because our own software has to land somewhere: a read that cannot be executed where the money moves is a report, not a decision.
We do not ask you to leave the stack you bought. We operate it and connect our own products to it, which is a different sentence from owning it.
Software we built, not software we resell.
Ten named products, eight in production, one in beta and one in build. They sit inside a pipeline of more than thirty in total, so roughly twenty others are still in design and development. Each exists because we needed an answer the market did not sell.
Every figure below is graded: how to read the numbers on this page.
Most are not a login we hand you: we run them, you get the output. Horizon and WAVE add a reporting view your team can have seats on.
WAndrome In production
Creative intelligence for MetaTells you which creative will win, before it goes live. Scores any image or video from 0 to 10, names the audience it will activate, and hands back a ranked fix list.
85% land in the performance band predicted, +16% before and after, 15% less wasted media on Meta. Internal benchmarks. Full section below.
WAVE In production
Wise AI Visibility EngineTells you what AI says about your brand, why, and what to change so it says something better. Five metric families, four engines, a dated history of every answer.
It maps your public footprint the way models ingest it, simulates real buyer prompts by market and language, and stores every response before scoring it, so drift shows up as a change rather than an opinion.
Measured across ChatGPT, Gemini, Claude and Perplexity, plus AI Overviews and AI Mode. The honest limit: measurement is reliable, influence is indirect. Nobody can reach inside a closed model.
Wise Horizon In production
Market intelligenceThe noise of your market as one signal per product, weekly, scored from minus one to plus one. Headwind or tailwind, with the reason attached.
Competitive pressure fuses a price gap and an auction gap into one factor. When the two disagree, that disagreement is the news, and an analyst validates before anything enters the index.
57 factors across 8 families, 47 competitor profiles, running weekly in banking. Permutation testing and correction for multiple comparisons, so it says "not distinguishable from noise" when that is the truth.
Feed'em In production
Product feed intelligenceThe algorithm is not the lever, the feed is. Every product bid on for what it is worth to the business, not what one channel attributed to it last week.
It joins Analytics, Ads and Merchant data on normalised GTINs into one transactional value per product, then places each against a cutpoint the algorithm calculates rather than a threshold somebody typed.
6% to 12% better ROAS across our own deployments. Internal benchmark, unaudited, and it moves with catalogue size, margin and season.
Wise Shield Beta
A firewall for LLM apps and agentsPut your AI in front of customers, not your data in front of the model. It sits between your users and the model, filters what goes in and what comes back, and stops credentials and customer identifiers leaving your building.
Four layers. Input filter, scored in memory in 28 written languages. Classification, a second opinion on the grey zone only. Output filter, scoring the answer and reading refusals. Output validation, is this safe to send to a customer.
84.8% correct decisions and 3.3% false positives on a 200-prompt battery we wrote and scored ourselves, median of four runs. The remainder are misses, so roughly 12% of hostile prompts get through today. That is the number that matters and it is why this is in beta, not in production.
Coherence In production
Brief to creativeConfirms, before anything is produced, that the brief serves the business objective. The cheapest moment to kill a bad idea is before the shoot.
It measures the link between brief and finished asset, across brief, moodboard, storyboard and script, and runs as a go or no-go gate in front of production.
Pairs with WAndrome, which scores the asset once it exists. Newer than the rest, so: a current capability, not a track record.
Wise Sonar In production
Listening and trend radarWhat people are saying, where, and what is about to matter. Early enough to act on this week rather than report next month.
Listening platforms are a mature market and we are not pretending otherwise. Ours is the layer on top: the mapping of the exact communities where your category gets argued about.
Tool-agnostic, so it sits on whatever listening licence you already hold rather than replacing it.
SCOUT In build
Google Ads operations, watched dailyEvery Google Ads account we manage, reviewed every day, across eight fronts. Nothing changes without a person approving it.
Running on our own managed accounts while it finishes. A failed read is reported as a failure, never as silence. Full section below.
Vektora In production
Image and video generation, governedTwenty three image and video models in one app, on a per-person credit balance. A price router sends each generation to the cheapest provider that has the model and records what it really cost.
Up to 15% less production cost at an average quality level 7% higher. Internal benchmark, unaudited. Full section below.
GOV Pulse In production
Public-sector signalsPublic-sector and regulatory signals in real time, for organisations whose market moves the week a rule moves.
It watches the sources that decide those markets, official gazettes, consultations, tenders and regulator communications, and turns movement into something a commercial team can act on rather than a legal team can file.
Public sources, dated and citable, with the same rule as Horizon: an analyst validates before a signal is published.
None of these needs your ERP, your CRM or your customer records. Beyond that they differ, so here is each one.
- WAndrome and WAVE. Nothing of yours. A creative file, or a public footprint anyone can see
- Wise Horizon. One aggregated export of your acquisition and auction signals. No records, no identifiers
- Wise Sonar. Reads through the listening licence you already hold, on credentials you issue and revoke
- Feed'em. Read permissions on advertising and analytics accounts you already own, revocable in a click
- Wise Shield. Sits in front of your model. It sees the traffic it is there to filter, and nothing else
- SCOUT. The one with write access, on the Google Ads accounts we already manage for you, and only after a person approves each change
- Vektora and Coherence. Only the brief and the asset. No account, no platform, no records
- GOV Pulse. Public sources only. Nothing of yours at all
Wise BI Foundry goes deeper, because reconciling a number against finance is the point of it. The data section covers what leaves and how.
WAndrome. Better performance on Meta, before you spend a cent.
WAndrome is where all five elements land in one product, which is why it gets a section of its own: custom data underneath it, a framework deciding what counts as better, our own software doing the reading, our crew acting on it, inside the platform you already run.
Scores the asset before the spend
Any image or video, rated 0 to 10, with the audience it will activate and a ranked fix list. Around three minutes for a read that used to take weeks of live testing and the budget to pay for it.
Decomposition, then the first seconds
Visual structure, the semantic tags the platform is likely to attach, the audience clusters those signals cross into, and the opening seconds scored against the exact drop-off point.
A corpus, not a prompt
A coordinated team of specialised agents reading vision, language and context together, over a tagged corpus of more than 10 million creatives that we built and maintain.
It does not replace the live test. It replaces the blind one. Every read is probabilistic, the figures above are our own and unaudited, and the method is on the table in the first session.
Every account, reviewed every day. In build
SCOUT is finishing, and it already runs daily on the Google Ads accounts we manage. Eight fronts, read account by account, with the result stored and comparable over time.
It is the difference between finding a problem tomorrow morning and finding it at the month close.
Campaigns that should be spending and stopped, including ones the platform suspended on its own.
The hour each budget runs out, day by day, so a cut shows up before the day ends.
Products that proved they sell, then lost exposure, brought back into traffic.
Broken, redirected or slow ad links, caught before they waste clicks.
Everything that changed, by whom and when, separating the team's work from the technology's.
Keyword proposals built from what converts, pauses for what does not. Nothing enters without review.
Uniform tracking across the account structure, so the analytics reading adds up.
Investment, month pace and conversions on one screen, with the historical comparison.
How it works. Four of the six steps are technology. The two that change anything are not.
Every account, every day, at a fixed hour.
Alert by email, chat or in the app, to whoever runs the account.
A simulation of what would change, without touching the account.
A person sees exactly what would be done.
The acceptance is recorded, with a name and a date.
Every change keeps the before and the after.
The three principles
A failure is a failure. If SCOUT could not read an account it says so, never "all fine" by omission.
Nothing changes without a person accepting it. The technology flags and proposes; the decision is the team's.
Everything has an owner and a trail. A named owner per account, approved access, every action in the audit log.
What it changes for you
A correction instead of a write-off, because the problem arrives the next morning rather than at the month close.
Decisions stay with people, and the operation is auditable: there is a record of everything done in the account.
SCOUT is developed in-house and the criteria, rules and sources behind each check are not disclosed, here or anywhere else. What is on the table in a first meeting is the output: an account read, and the list of what SCOUT would have flagged.
The best image and video models in one app, paying only for what you generate.
Vektora gives a creative team 23 models, 16 for image and 7 for video plus an Auto mode, served by five providers, on a credit balance per person. Built for our own creatives first, which is why it is the clearest proof of how we build.
The interesting part is not the models. It is that every generation is governed, priced and recorded.
The same model can cost close to seven times more at one provider than another. The router orders them cheapest first. If one fails, for any reason, the next takes it and the creative never notices.
- 01The credits leave firstOne transaction checks the balance and debits it. If it is short the request is refused there, before any provider is called.
- 02It is impossible to go negativeThe row is locked during the debit and the database refuses any balance below zero. Concurrent spends on the same balance are in the test suite.
- 03The real cost is recordedThe price of the provider that actually served it, times the real parameters, so the number matches the invoice, to the cent on models billed per image.
- 04If it fails, the credits come backEvery provider failing, a cancellation, or four minutes on an image: the generation is marked failed and the credits return.
- 05The ledger only adds linesAllocations, generations, refunds, adjustments. Nothing edited, nothing deleted. That is what makes it governable.
- 06It leaves labelledDownloads carry the AI Act Article 50 label: AI-GENERATED on the image, the notice in the first three seconds of the video.
Two cost bugs were found in August and both are on the record. A provider was silently billing a premium tier at around twenty times the credited price. Recording the real cost of every generation is what made it visible in a day, and that discipline is where the saving comes from, not a cheaper model.
The two figures above are our own and unaudited. The 15% is the recorded per-generation cost of our own production work against what the same briefs cost before Vektora. The 7% is our creative leads scoring paired outputs from the same brief, blind, old workflow against new. Both move with the brief, the format, and how much of the job is generative in the first place.
The providers are American and the infrastructure runs in us-central1, so this is not a European-residency setup today, and there is no monitoring or alerting yet. Both are on the list and neither is hidden.
More than thirty products in the pipeline, and a way of getting there.
Ten named products is what you can see. The pipeline is more than thirty in total, so around twenty others are still in design and development, many already running as alpha or beta MVPs on real problems.
Almost all of it is agentic by construction: agents doing the repetitive reading, the classification and the first pass, with people deciding what any of it means.
Idea to productionNot every idea survives. The gates are the point of having a framework.
Why it is fast: the teams that build are the teams that run client work, so a product starts from a problem somebody is already being paid to solve. The formal framework is what stops that speed turning into twenty half-finished tools.
Agents on the repetitive half
Reading, tagging, classifying, first-pass drafting: the parts that are high volume and checkable. People decide what any of it means.
Fast, for a client or a sector
The same machine builds to order. When an industry has a problem the market has not productised, we can put an MVP in front of real work in weeks rather than quarters. Test that cheaply before believing it, which is what the read and the pilot are for.
Methods with a record, not slides with arrows.
Eight frameworks, built to one architecture on purpose. Four layers of proven public practice, openly credited, and a fifth layer that is ours: the part that writes down what was claimed, so somebody can come back later and check.
The first four layers are the price of entry and we credit the research they come from. The fifth is the one almost nobody writes down, which is why it is worth paying for.
| Framework | The question it settles | The layer that is ours | What you are left holding |
|---|---|---|---|
| Wise BI Foundry | Is the number true | A five-layer factory where only the citable layer can be cited, governed by the Ruler | The warehouse, the data and the definitions, with a documented exit |
| Wise Double Cockpit | What the board and the team each steer by | The Metrics Compass: one goalset and one owner per metric | Two coupled cockpits of 4 to 7 metrics over one governed base |
| Wise GEO Compass | Whether AI engines name you | Wise AI Radar, plus a quarterly audit against a frozen prompt set, run by people outside the delivery team | A pre-registered prompt space and a monthly weighted share of voice |
| Wise Agentic Helm | How much autonomy each workflow earns | The Autonomy Ladder, A0 to A5, with a pre-registered gate protocol | The Ship's Log: every agent action, approved and reversible |
| Wise Sidecar | What to do next, at senior level | The Logbook, with an adoption rate and a quarterly review | Every recommendation dated, attributed and measured against what happened |
| Wise Reddit | Whether you have earned the right to speak | Citizen Score, AI Echo Index and the Tide Gauge incrementality test | Community standing scored 0 to 100, per community, tracked monthly |
| Wise Mix | Where the next euro should go | Triangulation: the model graded against real incrementality experiments | The model, the code and the data, on your cloud, on open foundations |
| Competition Drone | What rivals ship, spend and say | Two lenses at once: one flown close on named rivals, one flown high on the market | A live benchmark, refreshed, and the war-gaming when a decision needs it |
The pattern worth naming: every framework hands its record to the client, not to us. The Ruler, the Ship's Log, the Logbook, the Metrics Compass exist so that a year in, somebody can ask "what did we say would happen" and get a straight answer. That is uncomfortable by design, and it is the best reason to trust the rest of this page, because it is the part that can be used against us.
Custom data, from our sources.
This is the part that gets misread most often, so we will be blunt. Every piece of data that makes our products worth having is data we sourced or built ourselves.
None of it is built from client data. Your data is yours, it is used for your work, and it never becomes an input to something we sell to somebody else.
Market and competitive, sourced
Central banks, statistics offices, regulators, the economic press, competitor pricing pages, auction presence. All public, all citable, all dated.
57 factors across 8 families and 47 competitor profiles, refreshed weekly, validated by an analyst before anything enters an index.
Corpora we built ourselves
More than 10 million creatives, read and tagged. Prompt spaces and AI citation histories, frozen so they can be compared over time. GTIN-level demand and benchmark price across a catalogue.
None of this arrives in a subscription. It is engineered by the Data Science Lab, and it is why the products can say something a licensed tool cannot.
Eight years of our own pattern
What actually happens, across 500 or more brands since 2018, in categories that behave nothing like each other. Response shapes, the concentration that keeps repeating in retail catalogues, the tests that failed.
Structural patterns, never client data. No account, campaign or customer record leaves the engagement it came from.
Your first-party data is never an input to anything we sell to somebody else. It is not pooled with another client's, not used to train anything, and not shown to another client in any form.
Where a product needs a signal from you it is scoped, named on this page, and used only for your work. Horizon takes one aggregated export of your acquisition and auction signals; Feed'em reads advertising and analytics accounts you already own. Neither leaves your engagement.
With Wise BI Foundry the warehouse, the data and the definitions are yours and stay with you, with a documented exit protocol. Where hashed identifiers move for advertising activation they are still personal data, we treat them that way, and the legal basis and retention sit in the contract.
Exclusive. The products we run for you. We do not put the same one to work for two direct competitors in the same market, and the boundary is agreed in writing before you sign.
Shared by design. The market layer we source ourselves, productised into industry pools so one weekly signal serves several clients in a sector. Banking Signals Pool runs today inside Wise Horizon; Retail Price & Promo Pool and Pharma Demand Pool are in build. From our sources, never from client data.
Yours alone. Everything calibrated to your own signals. A pooled input is not a shared conclusion.
Two places this earns its keep in 2026: retail media, where the network grades its own homework and a reconciled base is the only way to check it, and clean rooms, where the question is whether your side is defined the same way twice.
Any one of these is a feature. Five together is a position.
Five elements do not add up, they multiply. Each pair does something neither half can do alone, and that is the argument for having all five under one roof.
- Delivery × TechnologyThe read reaches the live account the same day, by the people who already run it.
- Delivery × DataThe number in the meeting is the number in the warehouse, because one team owns both.
- Technology × FrameworksThe framework decides what the software is allowed to optimise for, and what would disprove it.
- Frameworks × DataA claim with a source, pre-registered before the test rather than chosen after it.
- Platforms × TechnologyOur software writes into the stack you already pay for, instead of asking you to leave it.
- Platforms × DataOne reconciled base behind Salesforce, Google and Meta, so the three stop disagreeing.
Six of the ten. The other four are the ones you would expect.
The honest test is who else could put the same five on the table. These are the three you are actually choosing between, and each is better than us at something.
The global network
Scale, balance sheet, continuity
- Better at multi-market governance, and at surviving its own staff turnover
- Its technology is group technology, so your competitor two floors up has the same
- Group media deals decide some answers before the question is asked
The specialist independent
Deep in one thing, close to the client
- Often the best in the market at its one discipline, and worth hiring for it
- Rents the same platforms as everyone else, so the edge is the people
- Judgement lives in one or two heads, and cannot connect to the rest of your funnel
Building it in-house
Context nobody outside will match
- Knows the business, the politics and the constraints better than any agency
- Learns from a sample size of one, and carries the full cost of the tooling
- Grades its own homework, with nobody outside to disagree
We are none of those three and we are not better at what they are best at. What we have is all five elements under one roof, and a commercial model that puts part of our fee where our claims are.
An easy solution, not a tech stack.
A wall of proprietary names makes people picture licences to buy, systems to wire in and teams to train. It is the opposite: you get the edge, we run the engine.
Learned in minutes
A read in plain language, with the decision it implies. The statistics stay under the bonnet. If it needs a training plan, we built it wrong.
Two exits, both written
Seven of the ten stop and nothing unplugs, because nothing was plugged in. Wise Shield sits in your request path and SCOUT writes to your ad accounts, so those two do have a removal step, and the Foundry hands over a warehouse you already own.
The risk you do take on
Smaller than a platform rollout, not zero. Scoped read access on four products and write access on one, a hashed activation path, a vendor to review. All of it answerable in writing.
Ours to maintain
Models we can swap, in private EU environments, with prompt-injection and output screening in front of anything that touches a client system. A named owner per product, maintenance on our side.
The commercial truth, asset by asset. We will not claim "no contract" where an annual licence exists, so here it is in the open.
- WAndrome
- Not a SaaS. A capability we run for you, under market exclusivity. No licence to carry.
- Wise Horizon
- An annual subscription, sized by users and scalable by area, proven in a closed-month pilot before any commitment. No technical lock-in, and cancelling breaks nothing.
- WAVE
- A one-off audit, a monthly tracking subscription, or an optimisation retainer. Start with the audit.
- Feed'em
- Project based. A free catalogue audit read by a senior specialist, then one category as a pilot, then rollout. You keep a system running in your own accounts.
- Wise Shield
- In beta, and priced as a pilot rather than a product. It starts in observation mode, blocking nothing, so you can see what it would have decided before it decides anything.
- SCOUT, Coherence and GOV Pulse
- No separate licence. They run inside the scope we already work in, and appear as capability rather than as a line item.
- Vektora
- Per seat, on a credit balance, billed on what is actually generated. Used on your work by our creatives; a client seat is possible and priced per seat.
- Wise Sonar
- Sits on the listening licence you already hold. If you do not hold one, that licence is the only new cost.
- The frameworks
- Advisory, side-car or full operation. The record stays with you in every model, and where we run the media we do not also get to judge it.
- Wise BI Foundry
- The methodology stays ours. The warehouse, the data and the definitions are yours, with a licence to the instance code and an assisted handover if you leave.
- How we are paid
- A fixed fee for the work, and on most performance scopes a component tied to agreed business metrics rather than hours logged. The split, the floor, the cap and who measures are agreed before the engagement, not after.
You are buying an outcome, not adopting a platform.
In a regulated sector these arrive before anything on this page does.
- ISO 27001 certificate number, and the scope it actually covers
- Sub-processor register, with locations
- Audit and inspection rights, for you and for your regulator
- Incident reporting and notification windows
- Exit and stressed-exit plans
- Key-person cover, replacement timelines and continuity
- Liability, professional indemnity and cyber insurance
- Our role classification under the EU AI Act, per product, not a blanket claim
We answer all of it in writing, per engagement, because it belongs in an annex rather than in marketing copy. Ask at the first meeting and it comes back with the proposal.
Every one of these starts with a read, not a contract.
The same four steps, whichever element you are curious about. We would rather lose an hour than sell you something that does not fit.
A read
45 to 60 minutes on one creative, one product line, one prompt set or one catalogue. No system access, no commitment, and you keep the read either way.
A pilot
One category, one closed month, one product line. Defined start, defined end, and a review with the people who have to live with the answer.
A split test
Half guided by the read, half exactly as you run it today. Same budget, same objective, compared on the metric you already trust.
Scale what won
Category by category, area by area. If the split test did not move, we say so and you stop, which has happened and will happen again.
Who this is for
Businesses with enough volume for a weekly signal to mean something, in a category with named competitors, and someone senior who already distrusts at least one number in their own reporting.
Who it is not for
Anyone below the volume where statistics beat judgement, anyone who wants a tool they log into and run themselves, and anyone who wants the answer before the read.
Step three is the only claim on this page you can break. Everything else asks you to believe something. A split test, on your budget, judged on your metric, does not.
That is deliberate. Thirty products in development are worth exactly nothing to you until one of them beats the way you run things today, measured by you.
Building this was a choice, and it cost us something.
Every hour spent writing our own software solutions was an hour of investment, not monetized services. The alternative was competing on price with agencies holding the same rented toolkit.
Three grades, and we label which is which rather than letting them blur. Independently held: ISO 9001 and ISO 27001, certified by Bureau Veritas, and the platform partner tiers. Certificate numbers and scope on request.
Internal benchmark: measured by us, on our own deployments, unaudited, and none of them run against a control group unless the sentence says so. That covers WAndrome's 85%, 16% and 15%, Feed'em's 6% to 12%, Vektora's 15% and 7%, Wise Shield's 84.8%, and the survey figures above. Coverage, not outcome: 57 factors, 47 profiles, 30+ products. They say how much, never how well.
If a number here matters to your decision, ask for the method. Where we cannot show it without naming a client, we say that too.
Independent, and it shows up in the work
No holding company deciding which platform gets the budget, no group media deal to protect. When we say a channel is not working, nothing internal punishes us for it.
The trade-off is ours to own: no group balance sheet behind us either.
Full stack on the platforms too
Google Ads Premier Partner plus Cloud and AI, and more than 100 platform, cloud and AI certifications across the crew, list on request.
We build our own and we run yours, which is the point of the first two elements.
Audited where it counts
ISO 9001 for quality and ISO 27001 for information security, both independently certified and available with scope.
On most performance scopes, part of the fee sits on agreed business metrics rather than hours logged.
One thing this page deliberately does not do is name client brands: most of this work sits under agreements that do not allow it. What we will do instead is arrange reference calls in your sector, walk through an anonymised split test including one we lost, and show the record a framework kept on a live account with the name removed. Ask at the first meeting.
Where each of these lives.
The full mechanism, the evidence, the limits and the caveats. Start with whichever question brought you here.
SCOUT and Vektora have their own sections on this page. GOV Pulse and the remaining frameworks have neither a section nor a page yet, and we do not link pages that are not live, on a page about not overstating things.
The awkward ones, answered first.
What exactly is proprietary at Wise Pirates?
Three of our five elements. Technology: eight products in production, Wise Shield in beta and SCOUT in build, inside a pipeline of more than thirty in total. Frameworks: eight documented methods, each with a fifth layer that is ours. Custom data: market and competitive data we source, corpora we built, and eight years of our own pattern.
Where does your custom data come from? Do you use our data?
Our sources, never yours. Public market data, competitor pricing and auction presence, corpora we built such as the 10 million creatives behind WAndrome, and our own accumulated pattern. Your first-party data is never pooled, never trains anything, and is never an input to a product we sell to somebody else.
What does it cost?
It depends on the element and the scope, and we quote after the free read or audit rather than before. The shapes are a fixed fee, an annual subscription for Wise Horizon, monthly tracking for WAVE, project fees for Feed'em and a pilot for Wise Shield. On most performance scopes part of the fee sits on agreed business metrics.
You say more than thirty products. How many are actually live?
Eight run in production today, Wise Shield is in beta and SCOUT is finishing. The other twenty or so are in design and development, many running as alpha or beta MVPs on our own work before any client sees them. We label the stage on every product rather than counting the pipeline as a portfolio.
Is there lock-in, and do we have to buy licences?
Mostly no licence on your side: the stack is a capability we run for you, with Wise Horizon on an annual subscription and WAVE on a tracking tier. On exit there are two shapes. The products stop and nothing unplugs, because nothing was plugged into your systems. Wise BI Foundry is different: you already own the warehouse and the definitions, and you get a documented exit protocol, a licence to the instance code and an assisted handover. Each element also stands alone, and most clients start with exactly one.
You promise exclusivity but you are also building shared industry pools. Which is it?
Both, on different things. The proprietary products are exclusive: we do not run the same one for two direct competitors in the same market, defined in writing. The market data pools are shared by design and built from our own sources. Anything calibrated to your signals is yours alone.
How do you build products this fast without breaking things?
Two gates and a rule. The gates are in the innovation framework, which kills more ideas than it advances. The rule is that nothing reaches a client before the crew has run it on our own work first, which is why Vektora and SCOUT were internal tools long before they appeared on this page.
Is this just a wrapper around ChatGPT with your logo on it?
No, and it is a fair thing to check. The value sits in the custom data and the method, not the model. WAndrome runs a coordinated team of specialised agents over a tagged corpus of more than 10 million creatives. Horizon adds statistical guardrails and a human analyst before anything enters an index.
How is a proprietary framework different from a methodology slide?
A slide asserts. A framework pre-registers what would count as success, runs the test, and records the verdict whichever way it goes, including "not distinguishable from noise" and recommendations a client declined.
What proof is there that any of this works?
Mixed, and labelled as such. ISO 9001 and ISO 27001 are independently certified. Figures like Feed'em's 6% to 12% ROAS range, or WAndrome's 15% less wasted media on Meta across six projects in the first half of 2026, are internal benchmarks from our own deployments, unaudited. The honest answer is the split test: half guided by us, half as you run it today, judged on your metric.
Who actually works on our account, and what happens if they leave?
A named team, with the senior people identified before signature and their allocation written into the scope. The frameworks exist partly for this: the record survives the individual. Key-person cover, replacement timelines and continuity go in the contract annex.
Is any of this a risk under the EU AI Act or GDPR?
The stack is built EU-first: no core-system access, no customer records inside the products, private environments and data in Europe. We state our role classification under the AI Act per product rather than claiming blanket compliance. Vektora is the exception we name on the page: its providers are American and it runs in us-central1 today.
How long before we see something?
A read takes 45 to 60 minutes and you keep it. A pilot runs a closed month or one category. Search and AI visibility take longer: first measurable moves in 60 to 90 days, structural movement in 6 to 12 months.
Bring the question. We will bring the stack.
One creative, one product line, one prompt set or one catalogue. 45 to 60 minutes, no system access, no commitment, and you keep the read whatever you decide afterwards.
Book the read →
