order system for cafés · draft
Search ads.
One agent that sets up your Google advertising. It reads what the account has done so far, term by term, sees who else is in the auction, and decides what to bid on, what to block and what to test. Then it writes the ads, produces the material and builds the campaign in the account.
We read what
the money did
Before a word is written the agent pulls in everything the account has done so far. No screenshot from the dashboard — the numbers straight from the Google Ads API, per campaign, ad group, keyword and search term. And the one number most accounts never look at: how much of the auction we didn't even show up in.
- What each campaign cost, and what it gave back
- Impression share — and whether budget or rank capped it
- Cost per conversion, and which way it is heading
- Search terms that spent money and gave nothing back
- Where the funnel leaks: impression, click, page or form
Brand search
Non-brand search
Performance Max
Competitor terms
We see who else
is bidding
Then the agent reads the competition where it shows: in the auction. Who overlaps us on which terms, who outranks us, who has started bidding on our brand, and what their ads promise. Everything running on Google sits in the open in the Ads Transparency Center — the agent reads it so nobody has to.
- Who is in the same auctions, and how often they win
- Who bids on our brand — and what it costs us
- Which promise and which offer their ads lead with
- Sitelinks and pages they send people to
- The term nobody bids on — the gap we can take
Auction insights
We pick up
what's happening
At the same time the agent watches everything outside the account that can move the result. Season and demand, a click price that suddenly climbs, a competitor entering the auction, a rule change from Google — the things that make a term cheap this week and expensive the next.
- Search demand by week — when the category peaks and dips
- Click prices climbing on the terms that matter
- A new advertiser in the auction, or one leaving it
- Google's own changes: match types, defaults, policies
- Questions people started asking that no ad answers yet
Signal feed
Everything weighed
into one decision
This is where it all meets. The results, the auction and the signals are laid side by side, and the agent draws one conclusion from them together. It comes out as three calls: what to pause, what should get more budget, and what to test — never a fixed kill rule, always a question of sample size first.
- Which campaigns to pause, and why
- Which one is capped by budget and can take 20% more
- Which terms to add, which to block — from real search terms
- Which angle is worth its own ad group
- A hypothesis with a target that can be judged later
The decision
- Pause competitor terms: 3 conversions for SEK 2,400
- Non-brand +20%: capped by budget, 34% of the auction lost
- Test: "alternative to" gets its own ad group and page
- Judge by: non-brand cost per conversion 356 → under 320
We learn what
people actually type
Before anything is written, the agent goes through the words buyers use — from the account's own search terms, from sales calls, from competitors' ads — and sorts them by intent. Brand first, then the terms that mean "ready to buy", then competitors, then the problem-aware. Spend opens one rung at a time.
- Keywords from how buyers say it, not how we say it
- Four rungs: brand, ready to buy, competitor, problem-aware
- Phrase and exact first — broad only after 30 conversions
- A term with 50 searches and clear intent beats 5,000 mixed
- Terms we never bid on: free, jobs, course, "what is"
Keywords · by intent
We learn
the format
A search ad has no picture to hide behind. It is fifteen headlines of thirty characters, four descriptions of ninety, and Google decides which pair shows. The agent learns the limits, the customer's tone, and which words are never allowed — then writes inside all three.
- Fifteen headlines, thirty characters each, counted
- Four descriptions of ninety — one job per line
- Sitelinks, callouts and the path under the URL
- The page headline echoes the ad — message match
- Words that can never be used: guaranteed, cheapest, free
Responsive search ad
The right agents
are put to work
Now the right agents are brought in. One knows bidding, budgets and the auction, one writes the headlines and counts every character, one makes the landing page say what the ad promised, and one makes sure every click can be traced all the way to a sale. They work from the same brief and hand over to each other.
- Which agents the task needs — not all of them, every time
- One who knows bidding, budget and the auction
- One who writes the ad copy, and one who tightens it
- One who makes the page match the ad
- One who traces every click to a conversion, in the CRM
Agents on the job
Then the ads
get written
This is where the brief becomes finished ads. Fifteen headlines and four descriptions per ad group, each counted against Google's limits so nothing is cut off. Different enough from each other that Google has something to choose between — the same claim fifteen ways is one headline, not fifteen.
- Fifteen headlines with a real difference between them
- The keyword in at least three of them, the brand in one
- Every character counted: 30 and 90, never one over
- One promise per description — not everything at once
- Every ad group carries its hypothesis and its target
Writing the ads
Then the material
gets produced
Search needs no picture. Performance Max, Display and YouTube do — and if you don't give Google material, it stitches something together from your site, badly. So for every asset group the agent produces images in the customer's style, in every size Google asks for, all carrying the same promise as the search ad.
- Landscape, square and portrait for Performance Max
- 16:9 for YouTube, 9:16 for Shorts and Demand Gen
- One visual style across the whole asset group
- The same promise as the search ad, so the click matches
- Two styles to choose between — same job, different feel
Here everything stops
and waits for you
The build is finished. The ads, the material, the keywords, the negatives, the budget and the page sit ready exactly as they will land in the account — twelve drafts. Nothing goes live until someone says yes.
order system for cafés · draft
nord · draft
new ad group · draft
cafés · responsive display · draft

skippable · 15 s · draft
Skip Ad ▷|
Ad · 0:06
feed · 1:1 · draft

from the account's search terms · draft
from the terms that have converted · draft
per day · draft
headline mirrors the ad · draft
To the owners · at handover · draft
what we don't build · draft
We find the hours
that actually buy
Before the campaign is built, the agent goes through when the customer's own conversions happen. Not the industry's general best times, but the hours and devices that gave the cheapest results in this account — and sets the schedule and the bid adjustments from that.
- Conversions per hour and day, from the account's own data
- Which hours gave the cheapest results, and which burned budget
- Mobile and desktop don't convert at the same rate
- Bid adjustments per hour, never more than ±20% at a time
- New windows tested continuously, so the schedule improves
Conversions per hour
We check the account
before anything is built
There is no undo worth the name in an ad account. So the agent runs a pre-check first — that the tracking counts a real conversion, that the location setting means people actually in the country, that Search Partners and Display are off — and builds nothing until every line is green.
- The right account and currency — the wrong account costs most
- Conversion tracking tested with one real conversion
- Location "presence": people in the country, not fans of it
- Search Partners and Display off until they've earned it
- Brand excluded from Performance Max at account level
Pre-check
Then everything is built
in the account
The agent builds the campaign for real against the Google Ads API — campaign, ad groups, keywords, negatives, ads and assets, in that order. The negatives go in before a single click can happen. Everything is created paused, and nothing is activated by a machine. When you turn it on, the agent's job is done — from there a loop keeps it working.
- Campaign, ad groups, keywords and ads built in the right order
- Negatives applied before the first impression
- Manual bids until 30 conversions — then a target, not before
- Links get their tags, so every click can be traced
- Everything lands paused; you turn it on — and the agent is done
Building the campaign