Copilot — AI for performance marketing

How the Copilot works

Analyze. Decide. Prepare. You approve.

The same loop a good performance marketer runs every week — only with a data base that no human and no platform dashboard has: every order, every call, every order value, booked server-side.

The copilot loop. The fourth step is always yours.

STEP 01

Analyze

Campaign types, keywords, placements, audiences, creatives — measured against your real first-party revenue, not against the numbers the platform issues to itself. Including keyword research with search volume and price ranges.

STEP 02

Decide

Every finding comes with evidence and a consequence: which campaign type burns money, where the budget should go, which keyword is missing — sorted by impact in euros, not by gut feeling.

STEP 03

Prepare

The implementation is ready and waiting: budget change inside the safety corridor, new campaign or ad always created paused, every write operation validated against the Google API in a dry run first.

STEP 04

You approve

One click applies the proposal, one click discards it. Not a cent moves without your consent. Every approval lands in the audit log — who, what, when, and what the follow-up measurement showed.

New · The Copilot page

One page. Finding, proposal, effect, log.

Everything the Copilot finds lands in one inbox: findings from five sources — autopilot, budget advisor, anomalies, click price, search terms — ranked by effect in your reporting currency. Where an action fits, “Create proposal” sits right next to it; the numbers are always calculated by the server, never by the browser. Below it: the open proposals, the effect of executed changes and the log.

Example with made-up numbers: a finding from the search terms becomes a proposal, with your click an executed change, and two weeks later a measured effect.

Five sources, one inbox

Autopilot, budget advisor, anomalies, click price and search terms come together; duplicate findings on the same campaign are merged.

Ranked by money

What moves the most comes first: the effect per finding in your reporting currency, level “high” on top.

Up to ten at once

“Select all” approves up to ten proposals in one go. Each one still passes the dry run, the drift check and the log.

Keyboard instead of mouse

j and k move through the findings, Enter triggers the row’s action, x discards. Approving always asks once more.

Measured effect

Before against after, per day. An interim result after two weeks, the verdict after four.

Every executed change is re-measured — cost, purchases and ROAS per day, after against before, from your own orders instead of platform attribution. The first three days after the change don’t count, and for purchases that arrive late the verdict waits another week.

Example with made-up numbers. Before and after per day from your own orders, not from platform attribution. For an increase, “works” means at least 5% more purchases per day at a ROAS above your break-even.

Four verdicts

“Works”, “no effect”, “worse” or “too little data” when fewer than five purchases are available. A pause works when the campaign costs almost nothing afterwards.

Undo is a proposal

If the measurement says “worse”, one click prepares the way back as a counter-proposal. Same chain: dry run, your click, log.

Sum instead of single case

The effect of all changes in the last 30 days stands as a sum on the Copilot page and in the weekly report.

A watchdog for the measurement

If an effect measurement is overdue, the alerter reports it. There are no silent gaps.

Action catalogue

What the Copilot prepares. And what you say to it.

Every action comes from your audited numbers, runs as a dry run first and then waits for your click. None of it moves money before you have seen it.

Shift budgetDaily budgets by real purchases per campaign — not by the carts Google counts as conversions. The basis is the budget as set, in account currency, and how much of it is actually spent: a campaign that doesn’t spend its budget gets no “more”. Google Ads and Meta, at most ±20 % per step, shared budgets are refused. You approve.
Pause a campaignCampaigns that cost money and sell nothing, as a pause proposal — Google Ads and Meta. The Copilot can propose pausing, never activating. You approve.
Add negative keywordsSearch terms with clicks and cost but no purchase in 30 days, as exact-match negative keywords on the campaign — at most 20 per proposal, with your purchase truth alongside, not just Google’s number. You approve.
Adjust the bid targetTarget ROAS, target CPA or CPC cap only on an existing target and by at most ±20 % — never a first target, never on portfolio strategies. You approve.
Exclude locationsRegions with cost and no purchase as an exclusion on the campaign, at most ten per proposal. Can only lower spend. You approve.
Create a campaignShopping and Performance Max campaigns with ad group and product group in one go — always paused. You approve.
Draft adsResponsive search ads from the creative brief as a paused proposal, character limits checked server-side; Meta catalogue ads likewise paused. You approve.
Exclude devicesTablet and connected TV set to bid adjustment zero — can only lower spend, never raise it. You approve.
Fill audiencesEleven segments from purchases and carts, daily to Google and Klaviyo, consent only. Runs after a one-time approval.

What never happens: activating a campaign, setting a first bid target, changing audience targeting. And none of it without your consent — by default one click per proposal.

New · Steering via AI chat

Tell the chat. Approving is one click.

Ask the AI chat what could work better in your Google Ads account — or say straight out what you want changed. The Copilot checks the maths against your real numbers, prepares a finished proposal and has Google test it in a dry run. A card with before and after appears in the chat. Nothing runs until you click.

Example with made-up numbers. This is how a proposal appears in the dashboard’s AI chat; it only runs on the click of a signed-in person. Available for Google Ads; daily budget and pausing also for Meta.

AI chat with a proposal card (example data): question about an exhausted budget, table with ROAS and cost per purchase, card “Change daily budget” from 40 to 48 € with an “Approve and execute” button — This is how it looks in the dashboard: the real chat interface of the demo tenant, answer and card with example data. The card waits for your click.

Before, after, evidence

Every card shows the old and the new value, the change in percent and the numbers the proposal rests on.

Nothing goes stale quietly

If the account has changed since the proposal, it does not run. Open proposals expire after 24 hours.

A complete log

Who proposed what and who approved it, before, after and Google’s answer — on one page, for you and your team.

Limits nobody moves

Budgets at most ±20% per step, then a 24-hour cooling-off period. Campaigns can only be paused, never activated. You unlock every capability one by one — and switch it off again at any time.

Steerable via chat today: daily budgets, pausing campaigns, pausing or activating ad groups, excluding tablets and smart TVs, negative keywords, bid target and location exclusion in Google Ads — daily budget and pausing on Meta as well. The AI cannot approve anything itself, not even by a detour.

Without a shop

No shop? Then the Copilot works in cost per enquiry.

For trades, practices, consultancies and dealers who sell on enquiry, the Copilot judges every campaign by its cost per proven enquiry — a submitted form, a qualified call, a booked appointment. Clicks on phone or WhatsApp links do not count as enquiries, and neither does spam.

Example with made-up numbers: the same proposal as for shops, calculated in cost per proven enquiry instead of ROAS.

Your target

You decide what an enquiry may cost. Without a target the Copilot uses your account’s median — and only judges from 10 proven enquiries per campaign.

Spam doesn’t count

Mark an enquiry as spam or “not a lead” and it no longer counts anywhere — and is retracted in Google Ads.

The real order value

Once the job is won according to your CRM or invoice, the real value replaces the estimate. That is how Google learns which enquiries bring revenue.

Audiences from enquiries

Three lists built from your enquiries — everyone who enquired, won and still open — as audiences for Google Ads once you switch them on.

Copilot from €749 a month, counted by proven enquiries instead of orders — tracking included. Everything for lead businesses →

A real case · D2C brand, UK · Google Ads

23% of the budget, 6% of the purchases. Visible only once you break it down.

The platform reports looked unremarkable. Only the breakdown by campaign type — spend from the ad account, purchases from our own measurement — shows where the money really goes.

Purchases as we measure them ourselves (last paid click), spend from the ad account. This is the breakdown the Copilot runs every week.

14 days

a quarter of the money, a sixteenth of the purchases

Over 14 days one channel took almost a quarter of the budget and delivered every sixteenth purchase. The platform report gave no hint: there, generous view-through counting fills the gap.

1 click

to the reallocation

The Copilot proposed the shift into search, researched the keyword gaps (search volume, competition, click price) and placed the new campaign fully configured but paused into the account. Approval and activation stay human.

Week after week

measured, not believed

After the shift the same loop checks whether the decision holds — against real orders, not platform attribution. What does not work goes out again. As an evidenced proposal, too.

What's sitting in your account that the platform reports don't show? We'll look together on the call.

Book a demo

Runs alongside your existing tracking · compare first, then decide

Keyword intelligence

Classic SEO tools tell you the volume. We tell you which keywords buy from you.

The Copilot has direct access to the original data of the Google Keyword Planner — search volume, competition, cost-per-click ranges, per country. And it crosses them with something no keyword tool in the world has: your real conversions. On top, it shows who your competitors on Google are and which search terms they have that you are missing, marked as estimates.

Market knowledge from the source

Keyword research straight from the Google Planner instead of from estimate databases — the same numbers Google plans with. Included in the Copilot — instead of as an extra subscription to a classic SEO tool.

Crossed with your purchase truth

The decisive step: volume and cost per click say nothing about whether a keyword sells for you. The Copilot lays the market data over your real orders and finds the terms your profitable buyers are behind — and the expensive ones that only get clicks.

All the way to a finished campaign

The research doesn't end in a spreadsheet but in a fully prepared, paused search campaign with the terms it found — exactly how the campaign in the case above came about. Approval as always: your click.

Request a keyword opportunity analysis for your market →

YouTube

YouTube that brings enquiries.

Views alone don’t pay the bills. We show which video sends visitors to your website and brings enquiries or purchases, keep your videos up to date and set up new ones ready to go. Nothing changes on your channel without your approval.

Example data. Views from your YouTube channel, visits and enquiries per video from your own tracking (every link in the video carries its own tag), value from your conversions. Changes to your channel only go to YouTube after you confirm them.

Measured per videoEvery link in a video carries its own tag: visits, enquiries and their value per video, from your own data. Plus how far videos on your website are watched.
Videos kept up to dateThe video check finds past dates in titles, broken tags and missing links to your website. Titles, descriptions and tags come as a proposal with before and after, you confirm.
Set up ready to goNew videos arrive on your channel with chapters, subtitles, AI labelling and a link to your website, not public at first. A video only goes public once you approve it. On request we also produce product videos.

Control

Why the Copilot never flies alone.

Ad budget is real money. That's why every write operation is locked inside a safety frame — not as a limitation, but as the condition under which you can hand a machine access to your account.

Four gates between a proposal and your money. The third one is you.

Budgets only inside the safety corridor Changes of at most ±20 % per step, with a lock against accidental double-apply and a read-back check after every write.
Anything new is created paused Campaigns and ads are created but never activated — the dry run against the API comes before every real write, and the activation comes from you.
Every recommendation names its source No recommendation without the number behind it — and that number comes from your database, not from a language model's gut feeling. If the data isn't there, the Copilot says so.
Complete audit log Every proposal, every approval, every rejection, every write operation with a timestamp — for you and, if you're an agency, for your client.
Learning phases are left alone If Google or Meta reports a bidding strategy still learning, or it was changed only recently, there is no recommendation. The first matching rule wins.
You choose the mode Default: every change by your click. If you wish, small steps run after a veto period — budget up to 10 %, negative keywords, location exclusion, together at most 20 % in seven days. Every step is announced by email beforehand, one click stops it, and the period can be switched off again at any time.

Pricing

From €749 a month, net. Tracking included.

The Copilot builds on the full uncoverflow tracking — it's part of the price. Here are both tiers side by side, so you can see what the difference buys.

Tracking package · from €299/month net

The measurement foundation: server-side capture of orders, calls and forms · delivery to every connected advertising and analytics platform · returns dashboard with ad spend against first-party measured revenue · AI weekly report · data chat with your own numbers.

Copilot package · from €749/month net

Everything in the Tracking package — and on top of it: weekly account analysis down to keyword level · keyword research from Google's planning data · proposals ranked by impact and backed by numbers · prepared implementation for budget, pause, campaign, ad and audience · steering via AI chat with one-click approval · measured effect of every executed change · Meta steering for daily budget and pausing · management report.

The difference between the two packages is €450 a month in the first tier (up to 500 orders; all prices net plus VAT) — and it buys you the right-hand column: analysis, decision and ready-prepared implementation, week after week. All tiers →

Included

Weekly account analysis down to keyword level · keyword research on original Google Planner data (without an extra SEO tool subscription) · evidence-backed proposals ranked by impact · prepared implementation (budgets, pauses, campaigns, ads, audiences) · steering via AI chat · measured effect per change · management report as a PDF · the complete uncoverflow tracking.

Who it's for

Shops and companies from roughly €10,000 monthly ad budget without a full-time performance marketer of their own — and agencies that want to scale analytical depth without burning hours per client. Partner terms on request.

Getting started

Same as with the tracking: prove first, pay later. You get the first analyses and proposals during the parallel run — you see on your own account what the Copilot finds before you decide.

Book a demo We'll run the loop live on a running account

Frequently asked

Answered briefly.

Does the Copilot replace the Google Ads Editor?

No, and that's deliberate. The Editor is an excellent tool for building — but it doesn't know what's true. The Copilot provides the eyes: what really converts, what really burns money, where the budget really should go. It prepares the most common consequences for one-click approval; the fine work you keep doing in the tool of your choice.

What's the difference between the Copilot and the AI weekly report?

The weekly report (part of the tracking) tells you what happened. The Copilot goes two steps further: it decides what follows from it and prepares the implementation in the ad account — budget shift, pause, new campaign. Analysis is included; acting is the Copilot.

Can the AI spend my ad budget on its own?

No. There is no code path that moves money without your consent: the default is one click per proposal, budgets only within the ±20 % corridor, campaigns and ads are created paused only. The one exception you switch on yourself: the veto period for small steps, which are announced beforehand and only run if you don’t object — for new contracts only after a written addendum. That's not a setting you can flip by accident, it's architecture.

Does this work for Meta, TikTok and others too?

On the analysis side, yes — the Copilot measures every connected channel against your first-party revenue. On the implementation side, Google Ads is integrated most deeply (budgets, pausing, negative keywords, bid target, locations, campaigns, ads, audiences). On Meta it steers daily budget and pausing per campaign with a dry run and creates paused catalogue ads. Further channels follow as clients need them.

Read on