What Is Smart Bidding

The Spanish interface speaks of pujas inteligentes, though day to day the original name sticks. The decisive word is auction: the calculation is redone at the moment it is decided whether the ad appears and at what price, not once a day and not when the campaign is set up. Google Ads offers other automations that fall outside this category, such as Maximise clicks or enhanced CPC, because they do not optimise towards conversions using the same set of signals.

Four strategies make up the group. Maximise conversions seeks the largest possible number of conversions with the available budget. Target CPA (tCPA) pursues that same volume while holding an average cost per conversion as a reference. Maximise conversion value works on the amount rather than the count. Target ROAS (tROAS) adds an average return as a reference to that logic. The first two count conversions; the last two need every conversion to arrive with a value attached, which requires measurement to be properly set up beforehand.

The practical limit is not the strategy but the declared goal. The system optimises towards whatever has been marked as a conversion, so a definition that mixes actions of very different worth will be met to the letter and still fail to show up in the commercial result.

In every auction the system estimates the conversion probability of that specific impression and adjusts the bid accordingly. The signals it uses include device, location, time of day and day of the week, the actual query typed by the user, the remarketing lists they belong to, browser, operating system and, in Shopping, product attributes. These are combinations an advertiser could not reproduce by hand, because many of them exist only in that instant.

The model needs history. When a strategy is created or reactivated, when a setting is changed, or when campaigns, ad groups or keywords are added or removed, the strategy enters a learning state and can take up to three weeks, or the equivalent of one to two conversion cycles, to calibrate to the new goal. During that phase daily results fluctuate more than usual and are not a basis for decisions.

On data volume it pays to be cautious about the figures in circulation. Google's official recommendation is to assess performance over a long period, on the order of a month, and with at least 30 accumulated conversions, or 50 in the case of Target ROAS. These are measurement references, not activation requirements: a strategy can start with less data by leaning on the account's history. From that calculation you also have to subtract conversion delay, the time between the click and the conversion being recorded, which in long purchase cycles reaches weeks and always leaves the final days of any report incomplete.

The choice of strategy decides what the account optimises for, and it is made before any report is opened. A business with a uniform order value can be run on Target CPA without losing anything along the way. A catalogue with widely differing margins per product needs to work on value, because otherwise the system will buy the cheapest conversions rather than the most profitable ones.

The second decision is the pace of change. Every relevant modification resets calibration, so an account that is touched every week lives in permanent learning and never reaches a stable stretch that can be judged. The same goes for targets: cutting a Target CPA in one move shrinks the eligible inventory and usually reads as a drop in performance when in fact the brief has been changed.

The third affects measurement. Because the bid rests on the declared conversions, tracking quality stops being a reporting matter and becomes an operational parameter. A duplicated event, a conversion without value or a badly chosen goal does not merely distort the dashboard: it changes where the account puts its money every day.

• Start with Maximise conversions when history is thin and move to Target CPA or Target ROAS once the account has gathered stable data.
• Change one parameter at a time and wait for learning to finish before attributing any effect to the change.
• Move CPA or ROAS targets in small steps rather than jumps, so that a large share of eligible inventory is not dropped all at once.
• Check the conversion definition before choosing a strategy, because a badly declared goal is optimised towards the wrong result just as effectively.
• Assign value to conversions if the value-based strategies are to be used, and check that this value is closer to margin than to the invoiced amount alone.
• Subtract conversion delay when comparing periods and do not close conclusions on the most recent days of data.

• Adjusting the target every few days and blaming the strategy for an instability that the management itself created.
• Copying conversion thresholds from general articles without checking which strategy and which campaign type they referred to.
• Mixing conversions of very different worth under a single goal, for example a catalogue download and a closed sale.
• Splitting a small budget across many campaigns, so that none of them accumulates enough data to calibrate.
• Reading the learning state as a fault and reverting to manual bidding before the cycle has finished.