What Is Programmatic Advertising? A Plain-English Guide

Programmatic advertising is the automated buying and selling of ad space. Instead of a human negotiating with a publisher over a rate card, software decides in the time it takes a page to load which ad to show, to whom, and what that impression is worth.

That’s the whole idea. Everything else is plumbing.

Plumbing is worth understanding anyway, because programmatic is the only major ad channel where a meaningful share of your budget can disappear without anybody doing anything wrong on purpose. In search or social you’re buying from one company that owns the inventory. In programmatic you’re buying through a chain of intermediaries, from publishers you’ve never heard of. Difference between a well-run buy and a badly-run one is usually not the creative or the targeting. It’s whether you know where the money went.

What actually happens when a page loads

An auction runs, and it finishes before the page does. Here’s the sequence:

1. The publisher’s ad server notices it has an ad slot to fill and sends a request out.
2. That request reaches an ad exchange (a marketplace) carrying details about the slot. Site. Ad size. Roughly where the reader is. What device they’re on.
3. Advertisers’ systems evaluate the request and decide whether this particular impression is worth bidding on. And how much.
4. Highest bid wins, the ad is returned, and it renders.

All of it inside roughly 100 milliseconds. This is real-time bidding, or RTB. It’s the mechanism most people mean when they say “programmatic”.

Two things follow from that speed. Both matter operationally.

Nobody looks at the impression before it’s bought. There’s no human in the loop deciding this page is appropriate and that one isn’t. Whatever rules you set before the campaign started are the only judgement being applied. If your exclusion list is empty, you haven’t made a decision. You’ve declined to make one, and the auction will fill the gap on your behalf.

Every decision is made on the data in the bid request. Request carries what it carries. If a publisher passes a vague or spoofed domain, your platform is bidding on a description rather than a page. That’s why the transparency plumbing further down this article exists at all.

The four acronyms worth knowing

Most programmatic explainers drown you in initialisms. You only need four.

DSP, Demand-Side Platform. The buyer’s tool. Where an advertiser sets budgets, targeting and bid rules. When you “run programmatic”, you’re almost always sitting in a DSP.

SSP, Supply-Side Platform. The publisher’s side. Puts their inventory up for sale and works to get the best price for each impression.

Ad exchange. The marketplace where the two meet and the auction runs.

DMP / CDP, the data layer. Where audience information lives, so a bid can be based on who is likely seeing the ad rather than only where it appears.

A useful way to hold it: the DSP is the buyer, the SSP is the seller, the exchange is the market floor, and the data layer is what tells the buyer whether to raise their hand.

The thing worth noticing is that every one of those is a business with its own margin. Not a scandal. Market makers charge for making a market. But it’s the reason a rupee of budget and a rupee of publisher revenue are different numbers. And why “what did we pay” and “what did the publisher receive” are two separate questions you should be able to answer.

Ask which auction you are actually in

Before you tune a single bid, find out whether the exchange is running a first-price or a second-price auction. Correct bidding behaviour is opposite in each.

In a second-price auction the winner pays one increment above the second-highest bid, so bidding your true maximum is safe. You rarely pay it. In a first-price auction the winner pays exactly what they bid, so bidding your true maximum means paying your true maximum every single time. Industry moved largely to first-price. But “largely” isn’t “entirely”, and different supply paths inside the same DSP can behave differently.

This is a question your platform rep can answer in one sentence. Ask it. Second-price bidding habits applied to a first-price auction never surface as an error anywhere. They surface as a CPM that’s quietly higher than it needed to be, indefinitely. Which is honestly the most expensive kind of mistake because nothing ever flags it.

Where the money actually goes

This is the part that gets skipped. And it’s the part that costs the most. There are four leaks. All four are knowable, and none of them are reported to you by default.

1. Not every impression is a person

Invalid traffic is a real line item, and the industry has a formal vocabulary for it that’s worth borrowing, because it tells you what you can and can’t fix yourself.

Media Rating Council splits it in two. General Invalid Traffic (GIVT) is the routine, identifiable kind. Crawlers and spiders. Straightforward bot traffic. Data centre traffic. Detectable through ordinary filtration. Sophisticated Invalid Traffic (SIVT) is the kind built to evade detection: hijacked browsers and devices, adware, incentivised or deceptive clicking, malware-driven traffic using obfuscation (MRC Invalid Traffic Detection and Filtration Guidelines Addendum).

Practical consequence: GIVT filtering is table stakes and you should assume your platform does it. SIVT is where the money actually goes. It requires a vendor or an exchange that invests in detection. And it’s why verified inventory costs more per impression and usually costs less per outcome. When you choose between a cheap CPM on an unverified path and a higher CPM on a verified one, you aren’t choosing between expensive and cheap. You’re choosing between a known price and an unknown one.

2. Viewability is not delivery

An impression can be served and never scroll into view. “Served” and “seen” are different numbers, and the gap between them is money.

There’s an actual standard here, which is useful because it gives you something to hold a vendor to. A display ad counts as viewable when 50% of its pixels are in the browser window for a continuous one second. For creatives larger than 242,000 pixels the threshold drops to 30% of pixels. In-stream video needs 50% of pixels for a continuous two seconds (Google Ad Manager: Overview of Viewability and Active View).

Read that definition carefully. It’s a floor rather than a promise. Half an ad, visible for one second, on a page nobody was reading, counts. Viewability is a hygiene metric. It tells you the ad had a chance. Doesn’t tell you it took it. Buy against it, because unviewable inventory is wasted by definition, but don’t report it upward as though it were a performance number. It’s the absence of one specific failure, not the presence of a result.

3. The supply chain takes a cut at every step

Between the advertiser’s budget and the publisher’s revenue sit the DSP, the exchange, the SSP and often several more parties. You can find out who they are. Most advertisers never ask.

Three specifications exist precisely so you can. ads.txt lets a publisher declare who is authorised to sell their inventory. sellers.json “provides a mechanism to enable buyers to discover who the entities are that are either direct sellers of or intermediaries in the selling of digital advertising”, and the OpenRTB SupplyChain object “enables buyers to see all parties who are selling or reselling a given bid request” (IAB Tech Lab: sellers.json). Together they let you trace a single impression back through every hand it passed through.

Operational version of this is supply path optimisation. Noticing that the same publisher’s inventory is reachable through four routes at four different costs, and buying it through the shortest one. It’s unglamorous work that nobody will thank you for. And it’s one of the highest-return things available to a mid-sized programmatic buyer, because it improves your economics without touching your creative, your targeting or your offer.

Question to put to your platform or your agency, in writing: what share of our spend reaches the publisher, and through which paths? You’re entitled to an answer. Quality of the answer tells you a great deal about who you’re working with.

4. Frequency compounds quietly

Without a cap, the same person sees the same creative repeatedly. You pay each time. Past a certain point each additional view does less than nothing. It irritates.

Reason this leak is so persistent is that frequency is invisible in every report you’re likely to be looking at. Impressions go up, cost goes up, the ratio between them looks stable, and nothing on the dashboard says this is a small group of people seeing it dozens of times rather than a large group seeing it once. You have to go and ask for the frequency distribution specifically. And the first time you do that on an uncapped campaign is usually memorable.

Cap it before launch. Uncapping later is a decision. Never capping is an oversight.

The order I actually work in

If I were opening a programmatic account tomorrow, this is the sequence. Order isn’t decorative. Each step is close to worthless without the one before it.

Step 1 is the one people skip. And it’s the one that determines whether the other five matter. Automated buying will optimise toward whatever you tell it to, at speed and at scale. Point it at the wrong number and it will hit that wrong number very efficiently. If you can’t yet say what a conversion is worth to you, the sheet that tells you whether your marketing makes money is the prerequisite, not the follow-up.

Step 5 is the one people violate without noticing. A campaign that has been live for four days has produced a small sample, and a small sample produces dramatic swings that are mostly randomness. Reacting to each swing feels like management and is actually noise amplification. Decide in advance how long you’ll leave it alone, write that date down, and honour it.

The weekly check that catches most of it

Once a week, pull the placement report. The list of actual domains and apps your ads ran on. Read it. Not skim it. Read it.

You’re looking for three things.

Sites you can’t identify. Open the domain. If it’s a content farm with forty ad slots and no discernible reader, it’s a made-for-advertising site. Exists to convert your budget into its revenue. Exclude it and move on.

Sites that surprise you. Sometimes pleasantly. A niche publication you’d never have thought to buy, performing well. That’s the genuine upside of programmatic and it’s worth harvesting deliberately. Take the good surprises and buy them directly next quarter.

Concentration. If a small number of domains are absorbing most of your spend, your “programmatic reach” is really a handful of sites bought expensively through an auction. That may be fine. But you should know it. And you can often buy those sites for less by going straight to them.

That report is where the four leaks above stop being abstractions and turn into a list of domains you can act on. Fifteen minutes a week. Honestly the highest-yield fifteen minutes in the channel.

The three things that have to be present

Programmatic only works when marketing, technology and data are all doing their job.

Marketing decides who’s worth reaching and what to say to them.

Technology executes the buy at a speed and scale no human can match.

Data tells you whether any of it worked, and feeds that back into the next bid.

Weaken any one and the other two stop compounding. Good data with no offer is a well-measured failure. A great offer with no measurement is a guess you keep repeating. And strong marketing with weak technology is a good idea being bought at a bad price. Which is the failure mode programmatic specialises in, because everything still runs, the reports still populate, and nothing announces that it’s going wrong.

When programmatic is the wrong tool

Programmatic rewards scale and patience. It’s a poor fit when:

Your addressable audience is small enough to reach by hand. If your entire market is a few hundred companies, a list and an email client beat an auction. Auction’s advantage is finding people at a scale you can’t enumerate. If you can enumerate them, you’re paying for machinery you don’t need.

You need results this week. Learning period is real, and cutting it short means paying for data you then throw away. Programmatic is a channel you commit a quarter to, not a fortnight.

You can’t yet measure what a conversion is worth to you. Covered above, and the most common of the four.

Demand for your category doesn’t exist yet. Programmatic is very good at putting a message in front of people. It isn’t good at making people want something they’ve never considered. That’s a brand problem, and the point at which it becomes the binding constraint is when performance marketing stops working and brand is the only lever left.

If you’re early, comparing channels and trying to decide where to start, programmatic is almost never the answer. Google Ads and Meta Ads are a better first classroom. One counterparty, a much shorter supply chain, and mistakes that surface faster and cost less.

The honest summary

Programmatic isn’t a channel that works or doesn’t. It’s an execution layer. It makes whatever you already do faster and larger, including the mistakes. Get the offer, the measurement and the audience right first. Then automation is leverage rather than an expensive way to be wrong at scale.

The four leaks aren’t a reason to avoid the channel. They’re the reason to treat it as an operational discipline rather than a setting you switch on. Invalid traffic, viewability, supply path and frequency are all knowable. And the buyers who bother to know them are buying the same inventory as everyone else at a materially better price. That gap is the entire opportunity.

If you want the practical version of how this fits into a wider paid strategy, the performance marketing playbook covers how the channels sit together, marketing metrics explained covers the numbers you’ll be judged on, and what advertising actually is and why digital took over explains how an auction ended up deciding all of this in the first place.

FAQs

What is programmatic advertising in simple terms?

It’s the automated buying and selling of ad space. Instead of a person negotiating with a publisher over a rate card, software decides in the time it takes a page to load which ad to show, to whom, and what that impression is worth. Everything else, including the platforms and the acronyms, is plumbing around that one idea.

What is the difference between a DSP and an SSP?

A DSP, or demand-side platform, is the buyer’s tool: where an advertiser sets budgets, targeting and bid rules. An SSP, or supply-side platform, is the publisher’s side, putting their inventory up for sale and working to get the best price for each impression. The ad exchange is the marketplace where the two meet and the auction runs.

Is programmatic the same thing as real-time bidding?

Not quite, although the terms get used interchangeably. Real-time bidding is the auction mechanism: a request goes out, advertisers’ systems evaluate whether the impression is worth bidding on, the highest bid wins and the ad renders. All inside roughly 100 milliseconds. RTB is what most people mean by programmatic, but programmatic is the broader term for automated buying, and it also covers direct deals executed through the same pipes without an open auction.

Where does programmatic budget actually leak?

In four places. Not every impression is a person. Invalid traffic, both the routine kind and the kind built to evade detection, is a real line item. An impression can be served and never scroll into view, so “served” and “seen” are different numbers. The supply chain between your budget and the publisher’s revenue takes a cut at each step, and you can trace those steps using ads.txt, sellers.json and the SupplyChain object. And without a frequency cap you pay repeatedly to irritate the same person.

What counts as a viewable impression?

For display, 50% of the ad’s pixels visible in the browser window for a continuous one second. For creatives larger than 242,000 pixels, 30% of pixels. For in-stream video, 50% of pixels for a continuous two seconds. Treat that as a floor rather than a promise. It means the ad had a chance to be seen, not that anybody saw it.

Should I bid differently in a first-price auction?

Yes, and it’s the first question to settle. In a second-price auction the winner pays just above the second-highest bid, so bidding your true maximum is safe. In a first-price auction you pay exactly what you bid, every time, so your maximum becomes your average. Ask your platform which one applies on which paths before you tune anything else.

When is programmatic the wrong tool?

When your addressable audience is small enough to reach by hand. When you need results this week (the learning period is real, and cutting it short means paying for data you throw away). When you can’t yet measure what a conversion is worth. Or when demand for your category doesn’t exist yet. Automated buying optimises toward whatever number you give it, so a wrong target gets hit very efficiently.