Broad match and smart bidding: when to trust the machine
Where automation earns its keep, where it burns money, and how to tell the difference from your own data.
Google wants you to use broad match with smart bidding. Every recommendation in the account pushes you there. Sometimes it is right. The question is how to know.
What the machine is good at
Smart bidding is a prediction engine. Given a search, a user and a moment, it predicts the chance of a conversion and bids accordingly. With enough data it is better at that than any person. It sees signals you cannot: device, time, location, browser, past behaviour, and hundreds more.
Broad match feeds it more searches to predict on. With a good target and good conversion data, the combination finds customers on searches you would never have written as keywords.
What the machine cannot do
It cannot know your margin unless you tell it. It cannot know that a lead from a student is worthless. It cannot know that the sale it just claimed would have happened anyway on brand search. It optimises to whatever conversion you give it, and it will find the cheapest way to hit that number, whether or not that helps the business.
The three conditions
Broad match with smart bidding works when all three of these are true.
1. The conversion is real. A sale, a qualified lead, a deposit. Not a page view, not a click on a button, not an “engaged session”. If the conversion is a step rather than an outcome, the machine will buy cheap steps.
2. There is enough of it. Around thirty conversions a month per campaign, and more is better. Below that, the predictions are noise. Below fifteen, use manual or enhanced CPC and fix the volume problem first.
3. The target comes from the business. A target ROAS or CPA calculated from margin and payback, not picked because it looked good last quarter. If you cannot say where the target came from, the machine is optimising to a guess.
If any of the three is false, broad match will spend money on searches that cannot pay back, and the reporting will make it look fine.
How to test it on your own account
Do not switch the whole account. Run an experiment.
- Pick one campaign with more than fifty conversions a month and a real conversion action.
- Create a campaign experiment with a fifty per cent split. Keep the control on phrase and exact. Set the trial to broad match with the same smart bidding target.
- Run it for at least four weeks, or until each arm has a hundred conversions.
- Compare cost per conversion and, more importantly, conversion quality. Check the CRM or the order data. Broad match often finds more conversions of lower value.
- Read the search terms in the trial arm. If a third of the spend is on searches you would never want, the machine is not ready for broad on this campaign.
Where broad match usually fails
- Brand campaigns. Broad match on brand terms buys generic and competitor traffic at brand prices. Keep brand on exact and phrase.
- Low-volume B2B accounts. Not enough conversions for the predictions to work.
- Lead generation with no lead quality feedback. The machine finds cheap leads. The sales team finds them useless. Connect the CRM and feed qualified leads back as the conversion before going broad.
- Accounts with weak negatives. Broad match needs a strong negative list. Without one, it wanders.
Where it usually works
- Ecommerce with accurate revenue tracking and a margin-based target.
- High-volume lead generation where lead quality is fed back offline.
- Mature accounts with two years of conversion history and stable targets.
The short version
Trust the machine when you have given it the truth: a real conversion, enough of them, and a target from the business. Until then, it is optimising to whatever you got wrong.
Talk to Joseph about your account
A thirty-minute call. You describe the account, Joseph tells you what he would look at first. No deck, no pitch.