Introduction
Value-based bidding (VBB) has taken the digital world by storm. In our testing we’ve found, when used thoughtfully, it’s a top choice for achieving high-performance – but there’s a catch. Adoption can be slow on a client by client basis and results can be unpredictable.
The main culprit: an obsession with finding “the right” business metric to feed into the black box of automated bid strategies (the bidder), and then assuming that Google’s bidder will figure out how to drive our objectives, and get the performance we want.
Instead, we need to think differently. When it comes to VBB, it’s not just about defining your goals; it’s about understanding how the bidder actually works and feeding it metrics that help it make smarter predictions in real time.
This report lifts the lid on automated bidding strategies, with a look at how Google’s bidder might work, why metric choice matters more than you think, and how to craft strategies that actually pay off. Our recommendations focus on working with the machine, rather than trying to tame it. The goal? Practical, data-backed strategies that turn the bidder into your best asset.
Meet the Bidder
First, let’s start by sharing how Google explains the value-based bidding:
In bidding, machine learning algorithms [the bidder] train on data at a vast scale to help you to make more accurate predictions across your account about how different bid amounts might affect conversions or conversion value.
Now for Brainlabs’ take: what exactly is the “bidder”? Aside from the fact that the bidder is a set of predictive algorithms created through machine learning, let’s think of it as Google’s own black-box algorithm trained on a massive repository of historical auction data – all designed to predict what bid to enter into any given auction. Below are our assumptions on how the bidder is created (trained), used during the auction to determine bids, and improved (through feedback), which will be instructive in how we can feed it appropriate metrics for VBB.
Our explanation on how the bidder might work
The Training Ground (1)
The bidder starts with a machine-learning model trained on historical auction data from across Google’s ecosystem (1a). In this training, the bidder is adjusted to make accurate predictions of the probability of conversion, and estimated value from that conversion given a bid in a particular auction. Once trained, this parent model is fine-tuned on your specific account data to create a version tailored to your bidding strategy (1b).
The Bidder and In-the-Moment Auctioning (2 & 3)
Each time a user enters an auction (3a), the bidder pulls in real-time signals, estimates the probability of conversion, and assesses potential conversion value (2 & 3b) and from these predictions, a suitable bid given the account objectives is selected.
The Bid Outcome and Feedback Loop (4)
Every auction outcome – whether a conversion happens or not – feeds back into the model to sharpen future predictions (4a & 4b). The more it learns, the better it gets, evolving with each round.
How do we pick conversion objectives which enable the bidder to work most effectively?
Recommendation 1: Enable fast feedback
For VBB to thrive, fast feedback is everything.
Why?
We’ve seen in our diagram how your account data is used to update the bidder. The quicker you can feed conversion data back to the bidder, the faster it adapts to changing conditions, be that conversion rates, competitor changes or sales.
There’s another nuanced point worth considering, albeit speculative. If the main parent model is trained on data that gradually feeds into the accounts – like most accounts do with real-time web tags – then it’s possible that large batch uploads could lead to less accurate predictions from the models.
Practical suggestions
To tackle delayed conversions or values needing offline calculations, predict earlier touchpoint values in real time or use approximate values for those instances where the true value might take time to be calculated. If you’re worried about how this might not be accurate enough, see recommendation 3 below.
Recommendation 2: Maintain consistency in values
Stick to consistent conversion values across similar conversion journeys.
Why?
In our diagram, we see how relationships between auction signals and conversion outcomes are used to train the bidder. When values range too broadly, the bidder struggles to learn, impairing the model’s ability to predict – and that’s where performance stalls.
Practical suggestions
Don’t leave the algorithm to figure out wildly varying conversion values. If your values swing by orders of magnitude for the same journey or set of auction signals, consider tightening that variance a bit. Keep things in check, and your bidding becomes far more stable.
Recommendation 3: Forget perfection
Precision is overrated in this game. Recognise that the bid itself comes from an imperfect prediction process by the bidder – there’s no single “right” value.
Why?
As we’ve explained, the bidder makes predictions based on conversion outcomes and auction signals at scale – in other words, it generalizes.
Your values don’t have to be 100% accurate – directional is fine (for instance, uploading 80 vs 82 as a value won’t make a difference). The bidder can struggle to predict outcomes for certain VBB metrics, and there’s no way to know which ones are less than ideal until you try optimizing towards it.
Practical suggestions
Think twice before investing tons of time and resources into crafting the “perfect” VBB metric (looking at you, life time value). Instead, take small, incremental steps toward a goal that aligns with your business needs and the principles above. And if things don’t go as planned, be ready to step back, recalibrate, and try again.
Turning insight into action
To summarise, when it comes to optimising value-based bidding, there’s no silver bullet. Automated bidding is a game-changer, but only if you understand the rules. The magic isn’t in chasing down the “right” metric from a business point of view; it’s in crafting a metric that works with the machine but still drives business performance. That’s where real value lies.
Embrace value-based bidding, but stay agile – refine, test, and let the data lead. The results? Consistently high-performance without the guesswork.
