# Automated bidding: keeping control when the algorithm decides

> Performance Max and Advantage+ decide targeting for you. What can still be steered in 2026, and why the quality of your conversion data weighs more than your settings.

- Original page: https://dotsland.com/en/automated-bidding-keep-control/
- Language: en
- Published: 2023-04-07T17:45:59+02:00
- Updated: 2026-10-19T18:19:37+02:00
- Section: Digital Marketing
- Publisher: Dotsland, https://dotsland.com/en/

---

## In brief

- **Automated bidding has absorbed targeting.** Performance Max at Google and Advantage+ at Meta decide to whom, where and when your ads are shown. You no longer choose the keywords, the placements or the bid amounts.
- **The quality of the data you send drives performance more than any setting.** With a mediocre conversion signal, the algorithm will diligently buy useless volume.
- **Six decisions remain yours:** the value of conversions, exclusions, the product feed, creatives, budgets and account structure, audience signals.
- **The most frequent trap:** a campaign showing a spectacular return because it captures searches for your brand, which you would have got without paying.

On an e-commerce account we recently took over, a Performance Max campaign targeting a neighbouring country had spent several hundred euros in three weeks for a return of around 0.25: about €4 spent for €1 of revenue. No alert. As far as Google was concerned, it was doing fine, since it was spending its budget.

The cause was neither the bid, nor the budget, nor the creative: final URL expansion had been left on with no exclusions, and Google was sending that country’s traffic to the site’s French-language pages. The algorithm was doing what it was asked, with the data it was given.

The question in 2026 is no longer whether machine learning will enter online advertising. It is there, it decides, and it no longer asks your opinion.

## What has changed: you no longer buy targeting, you fund a decision

Until the early 2020s, managing a campaign meant choosing keywords, audiences, placements and bids. That work still exists, but in formats the ad platforms no longer push: automated bidding has become the default setting, and broad targeting with it.

Google converted its Smart Shopping campaigns into Performance Max in 2022: a single format that distributes your ads across Search, Display, YouTube, Gmail, Discover and Maps, with no arbitration possible on your side. Meta launched Advantage+ Shopping the same year, then made automation the default path in its ads manager.

The platforms’ reasoning holds up: they observe billions of behaviours and know better than you which user is close to buying. What they do not know is what has value for you: a sale at 30% margin and a sale at 4% look identical to them.

## What the algorithm really optimises: the signal you send it

A bidding algorithm does not seek to grow your business. It produces as many of the events you designated as conversions as possible, at the cost you accepted. This mechanism explains most of the disappointing campaigns we audit: the setting is not at fault, the declared target is wrong.

*For the same budget, the result depends first on the event declared as a conversion. No bidding strategy makes up for a badly chosen signal.*

| The signal you declare | What the machine optimises | What you get |
| --- | --- | --- |
| Purchase, form, call and sign-up counted together | The cheapest event to produce | Sign-ups with no intent, very few sales |
| Purchase, value equal to revenue | Revenue, margin being invisible | Growth on your least profitable products |
| Purchase, value equal to gross margin | The real contribution to profit | Less volume shown, profitability recovering |
| Add-to-basket as the primary conversion | Intentions, not purchases | A real cost per sale that nobody knows |

A good signal is useless if it arrives incomplete. Ad blockers, browser restrictions, the disappearance of third-party cookies: part of your sales never reaches the platforms, which optimise on a truncated picture of your business. We detail the mechanism and the check in our article on [tracking that lies to you in silence](https://dotsland.com/en/tracking-server-side-lying-in-silence/). Making conversion reporting reliable is a matter of [performance measurement](https://dotsland.com/en/expertise/analytics-and-performance/) before it is a matter of advertising.

Transparency about our ecosystem: Datafirefly Limited, our agency’s sister company, publishes [a server-side tracking platform](https://server-side.datafirefly.com/) (free plan up to 10,000 requests a month) that compares your real orders every day with those recorded by the ad platforms.

> **What to remember**  
> The quality and value of the data you send drive performance more than any bidding parameter. Fixing a conversion definition has more effect than three weeks of adjustments.

## Who decides what, in practice

Here is the table we set out at the start of an assignment.

*Split of decisions on a Performance Max or Advantage+ campaign, at 1 August 2026.*

| Decision | Who decides | Your room for manoeuvre |
| --- | --- | --- |
| Which user sees the ad | The algorithm | Indirect: audience signals and exclusions |
| On which network and at what time | The algorithm | Almost none, unless you isolate the lever in a dedicated campaign |
| The amount of each bid | The algorithm | You set the cost or return target, not the bid |
| What counts as a conversion and what it is worth | You | Total: the main control lever |
| What is excluded (brand, customers, countries, products, URLs) | You | Total, and rarely used |
| Creatives, product feed, landing pages | You | Total: the algorithm’s raw material |

## The six levers that remain in your hands

- **The value of conversions.** A single primary action per objective, valued in margin rather than revenue if your rates vary from one range to another.
- **Exclusions.** Brand queries, existing customers, countries outside the delivery area, out-of-stock products, staging, blog pages. Every forgotten exclusion is budget going out without anyone having decided it.
- **The product feed.** In these formats, the feed is the targeting: titles, categories, attributes, availability and price decide the queries on which you appear. Modules structure it on PrestaShop, WooCommerce and Shopware, including those of [the Datafirefly marketplace](https://www.datafirefly.com/), our sister company.
- **Creatives.** The algorithm runs the images and texts you supply, and no others. Three bland visuals make three bland combinations.
- **Budgets and account structure.** One campaign per market and per objective is the only way to stop the algorithm deciding for you between a profitable country and one that is not.
- **Audience signals.** Customer lists, product page visitors, recent buyers to exclude. They come from your [first-party data](https://dotsland.com/en/first-party-data-from-scratch/), and their quality sets the starting point for learning.

## The black box: what these campaigns do not show

An automated campaign returns an aggregated result. It does not say what share of the budget went to YouTube rather than Search, only provides an extract of the queries triggered, and does not separate the sales due to your brand awareness from those it generated. Readability is rebuilt through structure, not through reports.

- **Isolate your brand.** A search campaign dedicated to brand queries, excluded everywhere else. You then see what each one costs.
- **Read at account level.** Total revenue and advertising cost, week by week. A campaign that grows while the total stagnates is moving sales around, not creating them.
- **Test by geographic area.** Cut the campaign in two or three comparable regions for three weeks and observe the gap: it is the most accessible incrementality measurement without a tool.
- **Segment your asset groups** by range or margin level: a single group drowns any product-level reading.

> **The mistake not to make**  
> A campaign showing a 12-to-1 return while the account’s revenue has not moved is not performing: it is billing you for sales you already had. Searches for your brand name are the cheapest to convert, and therefore the first ones the algorithm grabs to meet its target. Exclude them, let it run for three weeks, compare the account total: if nothing drops, you were paying for your own brand awareness.

## Learning, and the mistake of tweaking everything every two days

A bidding algorithm needs volume and stability. Google long documented orders of magnitude around 30 conversions over 30 days to manage to a target cost per acquisition, more for a return-on-spend target. Below that, the machine learns from noise.

Any significant change to budget, target or structure restarts a learning phase of around two weeks. An account tweaked every two days never comes out of it: you are watching a permanent transition performance, and drawing false conclusions from it.

Our rule is deliberately rigid: one change at a time, at least three weeks of observation, never a judgement over a period shorter than your purchase cycle. This discipline applies to all uses of automation, which we cover in our piece on [artificial intelligence applied to marketing](https://dotsland.com/en/ai-marketing-sme/).

## What you can check this week

1. **Open the list of your conversion actions** and see which are marked as primary. A newsletter sign-up next to purchase, and you have your first problem.
2. **Assign a value to each conversion,** in gross margin if your ranges do not have the same profitability.
3. **Reconcile seven days of orders** between your back office and the ad platforms. A stable gap can be corrected; a gap that varies for no reason hides a measurement problem.
4. **Check the default settings:** geographic areas (they regularly revert to “all countries”), automatically added languages, URL expansion, staging and blog not excluded.
5. **Isolate the brand** in a dedicated search campaign, and exclude those queries from automated campaigns.
6. **Set an observation window** and stick to it: one change a week, no decision before three weeks.

These six points need neither budget nor tool: an hour and access to the account. They are the foundation of our [paid acquisition and visibility](https://dotsland.com/en/expertise/acquisition-search-advertising-social/) assignments.

## Frequently asked questions

### Do you still need classic search campaigns now that Performance Max exists?

Yes, at least on your brand queries and your high-intent queries. A dedicated search campaign guarantees control over the message and provides the point of comparison Performance Max lacks.

### How do you know whether your automated campaign brings sales you would not otherwise have had?

The return shown measures what the platform attributes to itself, not what it created. Cut the campaign in two or three comparable areas for three weeks, then compare their revenue with that of the areas left active. If the gap is small, so is the contribution.

### How long should you wait before judging an automated campaign?

Two weeks of learning, then at least a week of stable observation. If your purchase cycle exceeds two weeks, extend accordingly. Any significant change to budget or target resets the counter.

### My budget is modest. Does automation still work?

It works badly below a certain volume of conversions, for lack of data to stabilise learning. Below around thirty conversions a month, concentrate the budget on a single campaign and high-intent queries.

## Taking back control without taking back the bids

The debate is settled: nobody will go back to manual bidding at scale, and that is not desirable. The machine calculates better and faster than we do.

The work has shifted, it has not disappeared. Defining the objective, sending accurate and valued data, excluding what has no business in the campaign, letting time do its work. These four points are worth more than days spent in the interface.

**Unsure what your automated campaigns are really buying?** In a first meeting, we look at three things: your conversion actions and their value, the gap between your real orders and those seen by the ad platforms, and the share of budget that goes on your own brand. [Let’s talk](https://dotsland.com/en/contact-us/).