How a signal works
Every time someone searches, the bidding system has a fraction of a second to decide whether to enter the auction and how much to bid. It makes that decision from whatever it can see about the moment: the exact words typed, the device and operating system, the browser, the time of day, the day of the week, the person’s location and the location they seem interested in, the language they browse in, and any audience lists they belong to. Each of those inputs is a signal.
Signals are not settings. In an automated strategy you do not write a rule saying that mobile users get a higher bid; the system reads the whole combination live, produces one bid for that one auction, and discards it. Two people typing the same phrase a minute apart can be valued very differently.
Some signals you supply. Customer lists uploaded as first-party data, conversion values imported from your CRM, and the audience signal attached to a Performance Max asset group are all inputs you control. The rest come from Google’s own account and network history, and you never see them itemised.
Why signals matter
The quality of an automated campaign is mostly the quality of what it is fed. A bidding model with accurate conversion data, sensible conversion values and a clean customer list will separate a serious buyer from a casual browser. The same model with a broken tag, a conversion action that fires on every page view, or values that treat a small order the same as a large one will bid confidently on the wrong people.
This matters more in a small market than a large one. In Nepal the daily volume in most niches is thin, so the system sees fewer examples and takes longer to find a pattern. Sending a strong signal — a real purchase or a qualified lead rather than a form view — shortens that period considerably.
Where signals go wrong
The most common failure is a signal that lies. A conversion counted on a thank-you page visitors can reload, or a lead event that fires when a form is abandoned, teaches the system that low-value traffic converts. Nothing in the interface flags this. The campaign simply gets worse at finding customers while the reported numbers look healthy.
The second is confusing a signal with a restriction. An audience signal tells the system who a good customer probably resembles. It does not stop the campaign showing to anyone else, and advertisers who assume it does are surprised by where the spend lands.
How to act on it
Start with the data you send. Verify that each conversion action counts something a business would recognise as valuable, that values are attached wherever a sale has a value, and that duplicates are excluded. Then add what only you have: a customer list, offline outcomes, refunds removed from the value you report.
After that, resist the urge to tune. Signals need volume before they mean anything, and every edit resets the pattern the system was building. Change one input, let it run through a full sales cycle, and judge it on cost per acquisition rather than on the movement of a single day.
