Algorithmic bidding is when software, not a person, decides how much to bid for each individual ad impression or click. Instead of setting one bid for a keyword or audience and leaving it, the system predicts how likely this particular person, at this particular moment, is to convert, and bids accordingly. It happens in milliseconds, millions of times a day, across search, social and programmatic display.
I’ve spent a large part of my career managing campaigns on Display & Video 360, The Trade Desk, Meta and TikTok across APAC, EMEA and North America. The platforms differ, but the logic behind their bidding algorithms is close enough that once you understand one, the others make sense.
How algorithmic bidding decides a bid
Every time an ad opportunity comes up, the platform runs an auction. Before bidding, the algorithm looks at the signals it has about that impression. On Google Ads, according to Google’s Smart Bidding documentation, those signals include device, location, time of day, remarketing list membership, the search query, browser and operating system, and for Shopping, product attributes and price competitiveness.
The model compares those signals against past outcomes: which combinations led to conversions, and how valuable they were. It then estimates the chance this impression converts and sets a bid that serves your goal. A returning visitor on a laptop at 10am who searched a very specific product name gets a higher bid than a first-time mobile visitor at 2am searching something vague.
A human with a spreadsheet can adjust bids by device or hour. No human can adjust them for every combination of dozens of signals in every auction. That’s the advantage, and it’s real.
The main bidding strategies and what they optimize for
| Strategy | What it does | Use it when |
|---|---|---|
| Maximize conversions | Gets as many conversions as possible within the budget | You’re starting out and every lead is roughly equal in value |
| Target CPA | Aims for an average cost per conversion you set | You know what a lead or sale is worth and need cost control |
| Maximize conversion value | Gets the most total value within the budget | Orders or leads vary in value and you pass that value back |
| Target ROAS | Aims for a return on ad spend you set | Ecommerce or any business tracking revenue accurately |
Google phased out Enhanced CPC for Search and Display campaigns in 2025, as Search Engine Land reported. That pushed many smaller advertisers from semi-manual bidding onto fully automated strategies, which makes understanding them more important than it used to be.
Algorithmic bidding in programmatic
On demand-side platforms such as Display & Video 360 and The Trade Desk, the same idea applies to display, video, audio and connected TV inventory. You choose a goal, such as conversions, cost per action, viewable impressions or completed video views, and the platform bids per impression in real-time auctions across many exchanges.
Programmatic platforms also allow custom bidding. You define what a valuable impression looks like, for example weighting a purchase more heavily than a newsletter signup, or scoring certain site sections higher, and the algorithm bids against your definition rather than a generic one. Done well, custom bidding is what separates experienced traders from the rest. Done badly, it teaches the algorithm the wrong lesson at scale.
What makes algorithmic bidding work, or fail
The algorithm is only as good as what you feed it. In practice, four things decide results:
- Accurate conversion tracking. If your tracking double-counts, misses conversions or counts page views as conversions, the model optimizes toward noise. Fix tracking before anything else.
- Enough conversion volume. Models need data to learn from. A campaign with a handful of conversions a month gives the system very little to go on. Consolidating small campaigns, or optimizing to a higher-volume action like a form start while volume builds, often helps.
- Realistic targets. Setting a Target CPA far below what the account has ever achieved doesn’t make results better. It makes the campaign bid so cautiously that it barely spends.
- Patience during learning. After a strategy change or a big budget or target change, performance usually wobbles for a while. Making another change every two days resets the learning each time.
A simple example
Say a hotel in Dubai sells direct bookings to guests from the UK, Germany and India, and in this example UK bookings are worth more on average because those guests stay longer. If the hotel passes the booking value back to Google Ads and switches from Maximize conversions to Maximize conversion value, the algorithm starts favoring the searches and audiences that bring longer, higher-value stays. The number of bookings might stay flat while revenue rises. That shift is only possible because the value data exists. Without it, the algorithm treats a one-night booking and a ten-night booking as equal.
Where humans still matter
Algorithmic bidding handles the auction. It doesn’t choose your offer, write your ads, fix your landing page, decide which markets to enter or notice that a competitor just halved their prices. The best-performing accounts I’ve seen pair automated bidding with people who obsess over those inputs.
If you’re new to paid search, start with this plain-English guide to PPC advertising. If you’re running campaigns and want someone to check whether the bidding setup is helping or hurting, our paid media team can review it.