AI integration,built as a system.

Inbox

A client, by e-mailTuesday, 08:52

Spring opening of our second store

Hi, we open our second store on 12 May and want to make some noise around it: social, a bit of print in the neighbourhood, and a small opening event. Budget around 8,000 €. Can you send an offer by Friday?

Thanks!

1A brief arrives. As it is, in your inbox.

Record

Filed

ClientA retail brand

WorkOpening campaign: social, print, an event

DueOffer by Friday

MissingWho signs off the offer

MissingBrand assets, store address

MissingBudget with or without media?

2Filed as a job. What the brief leaves out is flagged.

Draft

Thanks for the brief. Before the offer we need three things: who signs it off on your side, your brand assets and the store address, and whether the 8,000 € includes media. With those by Wednesday, the offer is with you on Friday.

3A reply, drafted from the rules you set.

Approval

Waiting for your yes

Reply to the client, spring opening.

Rules kept · checks passed · logged 08:57

Yes, sendChange

4Nothing leaves without a person’s yes. Every step is logged.

AI is a tool. It lies.

It states things that are not true, with full confidence. It has confirmation bias: it looks for what confirms its first answer, and agrees with what you already believe. On its own it is not reliable.

It becomes reliable inside a system: written rules, guards, workflows, checks, and a person’s yes wherever a mistake would cost something. That system is what we build.

One rule at every size: nothing goes out without a person’s yes.
The bigger the company, the more gates.

We run on it first.Our own work runs on a system we built. One request, start to finish. We show it working, in a meeting.

Each client is sealed in its own workspace, and a check refuses to cross between them. Nothing goes onto a live account without a person’s yes for that one item. Our newest check was attacked five times by a second reviewer before we switched it on.

Every project starts with an audit and ends with your people able to run it.

Seven steps. At each one, something is yours to keep.

AuditYou holda written report
DesignYou holdthe system on paper
BuildYou holdthe rules, written
Run togetherYou holdthe numbers
EducationYou holdyour team, trained
HandoverYou holdeverything, in your folders
AfterYou holdthe 90-day review

What happens at each stepThe seven steps in full.

01AuditWe sit with you and the people who do the work: how it runs now, where the time goes, which tools and data are involved, which AI is already used and how, what must never be automated. You get a written report, and it is yours whether or not you continue.
02DesignThe system on paper before anything is built: what each part does, what it reads, what it may change, who approves, how it is checked, and when we stop. You approve it.
03BuildOn your accounts, in your name. The rules are written before the first agent runs. Tested on your own real cases, and against attacks, before anyone relies on it.
04Run togetherIt runs beside your team, measured against where you started. What fails is fixed; if it does not earn its place, it stops.
05EducationYou and your teams learn how it works, where AI gets things wrong, how to check it and how to change a rule. This step is part of every project.
06HandoverEverything documented and in your own folders. Our access is removed, and we confirm it.
07AfterA review at 90 days against the starting numbers. If you want, we stay on to look after it when the AI companies change their tools.

What we build at each size.

One person

Your inbox, notes and documents in one system.

Your inbox, calendar, notes, documents and research in one system that remembers your work and drafts what you approve.

A small business

Up to about 30 people: a shared workspace with rules your team follows.

A shared AI workspace with rules your team follows, and the work that eats the week: replies, offers, reports. Nothing leaves without a person’s yes.

A larger company

Departments, each with its own systems: workflows across them, connected.

Workflows across departments, connected to the systems you already run, each connection with its own permissions. One department proves it before the next.

A corporation

A programme built with your IT, security and legal teams.

An AI integration programme we build with your IT, security and legal teams: a policy, a register of every AI system and who answers for it, approval gates, pilots in your own infrastructure, then the build across the company. We bring in specialists where the job needs them.

Security is part of every build.Your accounts, your access, your data. Nothing sends, books or changes data without a person’s yes.

Yours, alwaysYour accounts, your access, your data. We work from an access you give us, the least that does the job, and it is removed at the end.
Where your data goesThe design names which AI company reads what, where it is stored and whether it leaves the EU, before anything is built. The accounts are yours; nothing goes to a provider you have not approved.
One switch per connectionEvery connection to your tools has its own permissions and can be switched off on its own.
A person’s yesA person approves every outgoing item. If you later decide that one routine step may run on its own, you name that step and we test it first.
Tested, logged, cappedAttack tests before handover and after every model change, a log of every action, spending limits on your own accounts.
What your staff already useThe audit checks which AI tools your staff already use, and with what data.
An outside auditor where the law asksWe hold no security certification. Where you need a statutory security audit, we say so and bring in a certified auditor; for a penetration test, a specialist who does that work.

What we never do.Sell a ready-made bot as custom work. Promise percentages. Leave you a system nobody knows how to switch off.

No ready-made botsWe never sell a ready-made bot or a template as custom work.
No AI acting aloneWe never let AI speak to your customers or act on your data without a person’s yes. A routine step runs alone only once you have chosen it and it has passed its tests; and never without telling people it is AI where the law asks.
Nothing in our nameWe never put your AI accounts, access or automations in our name, or resell you subscriptions.
No percentagesWe never promise percentages. We measure where you start and report against it.
No AI on peopleWe never build AI that reads employees’ emotions, screens job applicants, monitors staff or hands out their tasks, or scores people for credit.
Your data stays yoursWe never keep or reuse your data for our own work.
Always a switchWe never leave you a system nobody in your company knows how to switch off.

A fixed price for the audit. A quote per phase after it.Each phase after the audit is quoted before it starts; nothing begins without your yes on the quote.

The audit: a fixed priceThe audit has a fixed price, agreed before it starts.
The build: per phaseThe build is priced per phase, from the audit’s design, never by open-ended hours.
How longThe audit takes about 2 weeks; a first build 4 to 6.
Education: its own lineEducation is part of every project, and shown as its own line in the quote.
Afterwards: optionalLooking after the system afterwards is optional, monthly, with capped hours.
AI tools: billed to youAI tools are billed to you by the companies that make them. We never resell them.

The law where you are.Where you are decides which rules apply. The audit flags the legal questions; your lawyer has the last word.

Where you are decides which rules apply. For clients in the EU we work under the EU AI Act and GDPR. For clients in the US, under the federal and state rules that apply to them. The audit flags the legal questions; your lawyer has the last word.

When the audit finds a gap that belongs to Origin, Scale or Merch, the report says so, and only then.

Tell us what eats your week.

A few lines on the work and the tools it runs in are enough to start. The first conversation is free.

hello@umbrella-co.eu