You signed. Here is what happens next, stage by stage
Not a setup guide but a map of what happens after the yes: what gets installed, in what order, and how you know it is working. The examples come from Florence, the numbers from Keyword Planner and from Google's own documentation, read on 5 September 2026.
Cittago runs this as a service — see google ads agency.
There is a claim that turns up in every conversation about automated bidding, and almost nobody checks it: that you need a certain number of conversions before Performance Max can work. We went and read the documentation in three languages on the same day. Google publishes a threshold for seven campaign types. Performance Max is not one of them.
This is not for everyone. If the restaurant has a product problem, a staffing problem or a location problem, advertising makes it more visible rather than smaller. There is a whole stage below about the cases where the right answer is not to buy traffic, and it is there because it applies more often than is comfortable.
What you get out of reading: you know what will be asked of you and when, you know what to ask when somebody tells you to wait for more data, and you can recognise the moment a campaign stops being a cost and starts being a measurement.
- Which conversion thresholds Google actually publishes, for which campaign types, and for which one it publishes none.
- The five search markets every restaurant has, with monthly volumes from Florence.
- Two opposite calendars in one city: English demand peaks in August, Italian in November.
- Why the famous thirty conversions is a rule for reading results, not a rule for starting.
Start with the sentence you will hear at some point, usually in week three: we need more conversion data before we can switch on the automation. It is often true. What is not true is the number that tends to come attached to it, because for the campaign type most people mean, Google does not publish a number at all.
What Google actually publishes, and what it does not
On 5 September 2026 we read the official Google Ads pages for both automated bidding strategies in English, Italian and Romanian, looking for published thresholds. The set was defined by a rule rather than by convenience: the same two help pages, same identifiers, requested in three languages on the same day, in a real browser. The Target ROAS page carries a requirements list broken down by campaign type — seven rows: Search and Shopping, Display, App, Demand Gen, Video Action, Hotel, Travel, each with a written threshold in conversions and days.
“Search and Shopping campaigns: At least 15 conversions in the past 30 days at the conversion tracking level.”
Performance Max is not in that list — not in English, not in Italian, not in Romanian. The product name does appear seven times on the same page, but in other contexts: portfolio strategies are not available for it, and the average target ROAS reporting column is supported for it. Nowhere in the eligibility requirements.
So any figure presented as the minimum for Performance Max is a professional convention rather than a platform requirement. It may well be a good convention. It is not a rule, and the difference matters when somebody uses it to justify a delay.
That Performance Max works with no data, or that thresholds do not exist internally. It establishes something narrower and more useful: that the published requirements list names seven campaign types and does not name this one. If somebody quotes a number, the fair question is where they read it.
The short answer, before the detail
A restaurant's Google Ads campaign does not start with ads. It starts by defining the action that counts — a phone call, a booking, driving directions, an order — then checking the website and installing measurement, then choosing one of the five search markets every restaurant has, and only then running the first Search campaign. Automation comes last, when there is data you can trust.
What happens before that — the research, the conversation, the proposal — is in the article on how a Google Ads agency works with a restaurant. This one starts the day the proposal was accepted.
Stage 0: what a conversion means for a restaurant
A conversion is the measurable action you want somebody to take after seeing the ad. That sounds like a textbook definition, but for a restaurant it is a business decision: the possible actions are several and they are not worth the same.
- A phone call. It leads to a booking, so to a table filled at an hour you can plan for.
- An online booking. Same value, less work for the floor staff, but it needs a system that does not fall over on Friday night.
- Driving directions. The person is already moving; the intent is stronger than any click.
- A takeaway or delivery order. The only one that closes entirely online and the only one with a real money value.
- A menu view. Not a conversion. A useful signal of interest, a harmful target.
The choice is not to take all of them. If you treat a menu view like a booking, any automated bidding system will buy people who look and leave, because there are more of them and they cost less. What you call a conversion becomes, a few weeks later, what you buy.
Stages 1 and 2: the site has to hold, and measurement comes first
Paid traffic does not fix a weak website, it exposes it. The check is short and you can run it yourself: it opens in a few seconds on a phone on mobile data, the menu is text on the page rather than a PDF download, the phone number is tappable in the first screen, the opening hours are right including holidays, and the photographs are of your room rather than of a stock library.
If something is missing and cannot be fixed inside the available budget, there is an option few people offer because it is work: build one dedicated page, fast, written for a single action and measurable end to end. The website stays where it is. It costs less than a rebuild and it can go live in days.
Then measurement goes in, and it goes in before the first bought click. That is a condition rather than a preference: what was not measured cannot be recovered afterwards. The vocabulary matters too — these are conversion signals and audiences, not cookies. Under how consent works now, what reaches the account is an aggregated signal with estimated parts, not a list of visitors.
Stage 4: a restaurant has five search markets, not one
This is where the diagnosis lives, and it is the part that separates a campaign that was thought about from one that was copied. Before buying traffic you decide which demand you are buying. A restaurant has five markets at once, and they behave differently on volume, price, competition and season.
A restaurant does not have one search market, it has five
Its own name, the category, an attribute, a district and a language. A representative term for four of them, from Keyword Planner on Florence. The fifth is the restaurant's own name, and we are not printing anyone's.
Source: Google Keyword Planner estimates, extracted manually by Cittago from the Google Ads web interface; Florence, 5 September 2026; window August 2025 – July 2026 · cittago.com
| searches per month | |
|---|---|
| Category: best restaurants florence | 22,200 |
| Category: best pizza florence | 6,600 |
| Category: restaurants in florence | 5,400 |
| Language: restaurants in florence italy | 2,400 |
| Intent: where to eat in florence | 1,900 |
| Proximity: florence restaurants near me | 590 |
| Market | Example | Monthly volume | What you are really buying |
|---|---|---|---|
| Brand | the restaurant's own name | not published: these are real businesses | demand that already exists; the cheapest route to a visit |
| Category | best restaurants florence | 22,200 | generic demand, large and undecided |
| Category, narrower | best pizza florence | 6,600 | the same demand with a shape attached |
| Language | restaurants in florence italy | 2,400 | visitors who will never read an Italian menu |
| Intent | where to eat in florence | 1,900 | a decision being made, not a place being looked up |
| Proximity | florence restaurants near me | 590 | somebody in the street, minutes from sitting down |
Notice the last row. florence restaurants near me is the smallest term in the set at 590 searches a month, and it is the one closest to a decision: somebody standing in a street with a phone is not browsing. The largest term, at 22,200, is a person on a sofa three weeks before a trip. Both are worth buying and they are not worth the same.
The fifth market is the restaurant's own name, and we are not printing volumes for it because those are real businesses. It is worth saying what it behaves like though: brand demand is the cheapest traffic in any account, because the person was already convinced by somebody else, and it is the one thing a competitor can quietly take from you.
Two opposite calendars in one city
We pulled the Florence corpus from Keyword Planner on 5 September 2026, over the window from August 2025 to July 2026, and looked at the monthly series. The result is unusually clean.
English demand for Florence restaurants peaks in August and bottoms in December
Twelve months of one keyword. The shape is the one you would guess for a tourist city — and that is exactly why it is worth checking, because the Italian-language version of the same question does the opposite.
Source: Google Keyword Planner estimates, extracted manually by Cittago from the Google Ads web interface; Florence, 5 September 2026; window August 2025 – July 2026 · cittago.com
| searches per month | |
|---|---|
| Aug | 33,100 |
| Sep | 33,100 |
| Oct | 22,200 |
| Nov | 14,800 |
| Dec | 12,100 |
| Jan | 14,800 |
| Feb | 14,800 |
| Mar | 22,200 |
| Apr | 22,200 |
| May | 22,200 |
| Jun | 18,100 |
| Jul | 18,100 |
English demand peaks in August and September at 33,100 searches a month and bottoms in December at 12,100. The Italian-language version of the same question does the reverse: 27,100 in November, 8,100 in July.
The Italian peak lands in the English trough, and the other way round
Two keywords, one city, the same twelve-month window. “ristoranti firenze” peaks in November and bottoms in July. “best restaurants florence” does the reverse. A restaurant that budgets against one of them is budgeting against half its year.
Source: Google Keyword Planner estimates, extracted manually by Cittago from the Google Ads web interface; Florence, 5 September 2026; window August 2025 – July 2026 · cittago.com
| searches per month | |
|---|---|
| IT · Jul | 8,100 |
| IT · Nov | 27,100 |
| EN · Dec | 12,100 |
| EN · Aug | 33,100 |
For a restaurant that speaks to both audiences, this is not trivia, it is the budget calendar. The weak months of one language are the strong months of the other, which means there is no single low season to plan around — there are two half seasons that interlock.
That Italians eat out in November and foreigners in August. Keyword Planner measures the language of the query, not the nationality of the person typing it. The shape of the two curves is a fact; the explanation is a reading, and it is worth keeping the two apart. It also holds for these two keywords and not for every English one: restaurants in florence peaks in May instead.
Stage 5: start on Search, and start with the brand
The first campaign is a Search campaign, for three reasons that have nothing to do with fashion: you see what the person typed, you control what you buy, and you find out which searches produce real actions. Nothing that follows works without that answer.
Brand demand goes in its own campaign. It is cheaper than everything else, it has the highest action rate, and if you mix it with generic demand it will make your averages look excellent while telling you nothing about the rest. That mixture is the most common way a report can be beautiful and meaningless.
Stage 7: Target CPA and Target ROAS, and the difference that decides things
These two get confused constantly, and the difference determines what kind of customer you buy. Google's own page, read on 5 September 2026, puts it plainly:
“Target CPA bidding is an automated bid strategy that sets bids for you to get as many conversions or customer actions as possible.”
That is more actions at an average cost you set, with every action counting the same. For a restaurant that wants bookings, it is usually the right choice.
Target ROAS is a different animal. The same documentation says the strategy predicts the value of a potential conversion on every search and bids up where the predicted value is high: more value relative to spend. To do that it needs real values attached to conversions, not just a count of them. A restaurant with no online orders carrying amounts has nothing to feed it.
And there is one line on the Target CPA page that contradicts a very widespread belief: advertisers can start using Target CPA with no conversion history. The thirty conversions figure does exist in the documentation, but it is a recommendation for evaluating performance over thirty days, not a condition for starting. Two different things, and confusing them has delayed a lot of campaigns for nothing.
The same pages carry a note in all three languages we read: from June 2026 Google Ads is relabelling the bidding strategies. “Maximize conversions with a Target CPA” becomes “Target CPA”, and the value version becomes “Target ROAS”. The underlying behaviour stays the same, the note says. If your panel shows one name and the guides show another, that is why.
| Term | What it means here |
|---|---|
| Conversion | The measurable action you defined as valuable: a call, a booking, directions, an order. |
| Conversion signal | What reaches the account about that action after consent — partly measured, partly modelled. |
| Target CPA | Automated strategy that seeks as many actions as possible at an average cost you set. |
| Target ROAS | Automated strategy that seeks value rather than count; needs real values on conversions. |
| Performance Max | The campaign type that serves across all Google surfaces at once, with limited control over each. |
| Top of page bid range | Keyword Planner's published estimate for the top of the page. Not an invoice. |
Stage 8: what a customer is worth, not what a cover costs
A restaurant does not live on single meals, it lives on people who come back. That changes the arithmetic completely, and it is where most budget conversations get stuck, because somebody compares the cost of a click with the price of a dish.
The right comparison is different: what it costs to acquire a new customer, against what that customer spends with you over a year. A click that brings somebody who returns monthly amortises differently from one that brings a tourist once — and both are good customers, they are simply worth different amounts and are bought at different prices.
Invented figures, shown so the mechanism is visible; not data from a client. Say twenty clicks produce one visit and a click costs half a euro: a visit costs ten euros of advertising. If the person comes once and spends twenty-five, you barely broke even. If they come back four times over the year, the same ten euros bought a hundred. Same campaign, same cost, opposite verdicts — and the difference does not show up in the ads account, it shows up in the dining room.
Stage 9: when Google Ads is the wrong investment
This stage appears in everything we write about advertising, because it is the only one that does not end with an invitation to talk. There are four situations where the honest answer is no.
- The website cannot be fixed and cannot be replaced with a dedicated page. With no destination that converts, budget buys visits that close.
- The margin does not carry the acquisition cost. If clicks on your terms start at real money and you have no way to bring the person back, the arithmetic above never closes.
- There is no brand demand yet and no way to create it. A restaurant open three weeks has something else to solve first: giving people a reason to search for its name.
- The problem is somewhere else. Not enough staff on Friday night, wrong opening hours, reviews describing an experience you would not want amplified.
Where you are, in three thresholds
Not a conclusion: a way of working out which of the three cases you are in right now.
- You have not defined what a conversion is. Then this is not a campaign problem, it is a decision problem, and an hour of conversation settles it.
- You have defined it but it is not measured. Then measurement is the priority, and every euro spent until it works is a euro you learn nothing from.
- It is measured and you have a few weeks of data. Then the question is no longer whether to raise the budget, but which of the five markets to raise it on, and in which month. That is where somebody who does Google Ads management for a living earns their fee — or anybody else, as long as they answer that question before quoting you a number. If you are still at the first threshold, you do not need anyone yet.
If you want to see all of this laid out on a real account, with the numbers and the part that did not work, the case studies are more useful than any services page, including ours.
We re-run the extraction above periodically, and the next season will move the two calendars more than anything else on this page. Ask us again in three to six months 😉
Questions with answers we think are useful
Some of these we asked ourselves while reading the documentation; the rest are the ones that turn up in the second week of a campaign.
How many conversions do I need before starting Performance Max?
Google publishes no conversion threshold for Performance Max. On the Target ROAS page, read on 5 September 2026 in English, Italian and Romanian, seven campaign types are listed with thresholds and Performance Max is not among them. Any number you are given is a professional convention rather than a requirement.
Is the thirty conversions rule real?
The figure exists in Google's documentation, but not as a starting condition: it appears as a recommendation to evaluate performance over the last thirty days including at least thirty conversions. The same page says advertisers can start using Target CPA with no conversion history at all. It is a rule for reading results, not a gate.
What is the difference between Target CPA and Target ROAS?
Target CPA seeks as many actions as possible at an average cost you set, treating every action the same. Target ROAS seeks value: it predicts what a potential conversion is worth on each search and bids higher where that prediction is high. It needs real values attached to conversions, so usually online orders with amounts.
When should a Florence restaurant launch a campaign?
It depends which language you are speaking to. In our Keyword Planner extraction of 5 September 2026, English demand peaks in August and September at 33,100 searches a month and bottoms in December at 12,100, while the Italian-language equivalent peaks in November at 27,100 and bottoms in July at 8,100.
Can I start without conversion tracking and add it later?
You can, but you permanently lose the earlier period: unmeasured data cannot be recovered afterwards, and any automated strategy you use later starts with a shorter history. It is the one stage where hurrying costs exactly what it saved.
Is a menu view a conversion?
No. It is a useful indicator, but if you call it a conversion, any automated system will buy people who look and leave, because there are more of them and they cost less than people who book. What you call a conversion becomes, after a few weeks, what you buy.
Should I bid on my own restaurant's name?
Usually yes, and it is the cheapest traffic in the account, because the person was already convinced by somebody else. Keep it in a separate campaign, otherwise it lifts your averages and you can no longer read the performance of anything else.
Are the bid ranges in this article what I will pay?
No. They are top of page bid ranges published by Keyword Planner: an estimate of the market on that term, not an invoice. What you pay depends on ad quality, on the landing page, and on who else is bidding that day.
Why start on Search rather than Performance Max?
Because Search shows you the query. You see what the person typed, you control what you buy, and you come out with two lists: searches that produced actions and searches that produced only spend. Automation is better at optimising than at telling you what your market is.
Last updated: 5 September 2026. Volumes come from our own Keyword Planner extraction on that date, over the August 2025 – July 2026 window; the quotations from Google's documentation were verified the same day in English, Italian and Romanian.
Sources: Google Ads: About Target CPA bidding · Google Ads: About Target ROAS bidding · Google Ads: About Performance Max · Google Ads: Conversion tracking · Google Ads: Keyword Planner
