How automated bidding works
Every time someone searches, or a placement becomes available, an auction runs. Automated bidding sets your bid in that instant using signals the platform holds and you do not: the device, the time of day, the location, the exact wording of the query, the browser’s previous relationship with your site, and how similar situations have converted before. A manual bid cannot vary at that granularity, because you are setting one number for a keyword and the platform is setting a different number for every auction.
You steer it by choosing a strategy and a target. Maximising conversions spends the budget on the most actions; a target cost per acquisition aims each bid at a cost you name; a target return on ad spend bids for value rather than count. Whichever you choose, the system is predicting the probability and worth of a conversion, so the quality of your conversion data decides the quality of the bidding.
Why automated bidding matters
It is now the practical default on the major platforms. Manual bidding cannot see auction-time signals, and on most accounts that gap is wide enough that manual management loses on volume, on cost, or on both. For smaller advertisers it also removes a job nobody was doing well anyway, which was guessing keyword-level bid adjustments from thin data.
The trade is control for reach. You stop deciding individual bids and start deciding what a conversion is worth to the business, which is a far better question for an owner to be answering.
Common mistakes with automated bidding
Feeding it the wrong goal is the expensive one. If the conversion action counts every form submission including spam, or counts a page view as a lead, the system will faithfully buy more of that. Automated bidding does not judge quality; it optimises towards whatever you told it to value.
The second is impatience. Every change to strategy, target or tracking restarts a learning phase during which results are unrepresentative, and adjusting the target every few days keeps a campaign permanently in that state.
The third is thin data. These systems need conversions to learn from, and in a smaller market like Nepal a niche campaign may not produce enough for the model to work with. When that happens, optimising towards an earlier and more frequent action can give the system something to learn on, provided that action genuinely predicts sales rather than merely being common.
How to act on it
Fix measurement first. Confirm what each conversion action counts, remove the ones that are not worth money, and where you can, send a value back rather than a count so bidding optimises for revenue instead of volume. Importing offline conversions is worth the effort for businesses whose sales close on the phone or in person.
Then set a target you can justify from your own margins, change it in small steps, and give each change enough time to settle before judging it. Keep watching search terms and lead quality, because the system will find the cheapest route to whatever you defined as success. Getting that definition right is exactly what accurate conversion tracking is for.
