Not so long ago, marketers focused on radio, television, print and billboards. The objective was still similar to digital and modern marketing: incentivize a potential customer to purchase a product with the best return on investment (ROI) possible. What’s changed is both businesses’ and customers’ relationship with technology. In a data-based world the limitations of traditional marketing efforts were quickly exposed.
The paradigm changed from product placement and brand awareness to strategy based analytics, numbers and customization. Traditional marketing was incapable of efficient targeting, a prime time radio campaign might reach every single person in an area to share a discount on onions but what if everyone doesn’t want to hear about sales on onions? What if they did, but they already bought some onions yesterday and want to buy lettuce today? Not to mention, traditional mass market media is usually expensive, making lots of different commercials for different segments isn’t usually a possibility. What about results? How do I improve my efforts or adapt them if it’s hard to gather real metrics to analyse market traction?
This example is over-simplified but the point is to illustrate how inefficient these efforts can be: They are not targeted, difficult to personalise and most importantly give little insight or metrics into traction, impressions or information that might help you adapt and personalise future efforts.
The data ecosystem has become a part of Marketing and vice versa. Today we can identify users through devices, we can target, retarget and format and adapt messages. Better yet, we can learn through insights and study all types of metrics such as impressions and clicks. This is why AdTech and the topics of today’s blogpost are more relevant than ever. A great review in depth of this process can be found in this paper.
AdTech is a broad term in digital marketing which refers to platforms, tools, software and infrastructure used to purchase, sell, track or/and analyse ads and campaigns.
For what purpose? Well we can’t speak for all marketers but as we mentioned on some other posts it usually has to do with a market and context which has grown increasingly complex with time. Adtech comes in to simplify the landscape, streamlining processes.
The bottom line? Taking care of Return On Investment (R0I), optimising spending and allowing for planning and strategy.
Before we show you how deep the rabbit hole goes there are some basic AdTech concepts we should revise (you can get more technical on almost any of them in this paper). Here we go…
We´re going to focus on Real Time Bidding (RTB) Systems which have three main actors: those who want to sell ads, publishers; those who want to buy them, advertisers, and a market where they meet, the exchange. In general advertisers and publishers don’t actively buy or sell ads but rather operate through a partner/platform. Beyond these main players other platform within the RTB ecosystem include:
When you’re on this end, you’re shopping. DSPs are platforms where advertisers or ad agencies go to buy ads. If a marketing team needs ads to reach a specific audience and they have a defined budget in mind they will go to a DSP to buy inventory. In short, advertisers submit their campaigns into a platform in order to bid for the most suitable ad-slot.
SSPs are the other side of the coin, which is to say, Supply Side Platforms are for publishers what Demand Side Platforms are for marketers. Publishers can sell their ad inventory here, but just like marketers in DSPs want to make sure their ads are performing as well as possible, publishers use SSPs to make sure they are selling at the best possible prices. At the end of the day, both sides need to optimise and maximise their revenue.
Like the stock exchange, Ad exchanges have buyers and sellers: Advertisers and Publishers. Think of it as a marketplace where DSPs and SSPs can be used to process ads both ways. Unlike stocks (where bids and asks are continuously displayed), Ad Exchanges’ sell their inventory by auctioning one ad at the time. These ads can be found by industry, verticals or types.
We covered both sides of the coin, but how do we reach a win-win? After all, both players are looking for the best possible outcome. The solution: Real Time Bidding (RTB) Mechanisms.
Before we get into RTB and auctions we should clarify: Normally ads were sold in bulk through fixed CPMs or Ad Platforms where the user would set a campaign budget and target and the platform would decide on the best manner to spend that money in order to reach that target. The issue with this method that ultimately led to auctions was that traditional methods did not allow for impression/auction level control.
As its name points, RTB mechanisms auction ad-slots on the fly, bringing together SSPs, DSPs and exchanges thousands of times per minute to distribute the ads slots as they become available. But in the context of online marketing, not any auction process takes place: exchanges implement the one specified in the Open RTB protocol as defined by the IAB.
Let’s take it back to step 1: a user scrolling a feed or playing a game. Before a certain point in the feed or a certain level in the game the device will send a message: unbeknownst to the user there will be an open spot for ads on their screen.
The SSP redirects this message to the AdExchange which, in turn, sends this message to several DSPs. DSPs usually have a variety of potential candidates, so they run an internal process that decides which campaigns in their inventory seem relevant, are in line with their budget-spend and selects the best fit among them.
Ever noticed different banners for the same campaign? This is also determined here. DSPs will choose the creative. The final step? The DSP will submit a bidding price to the exchange. The next step is the auction itself.
Small disclaimer: We’re breaking this down in steps to explain it but there is actually a speed requirement - a very high speed requirement. Bids must be submitted in the order of milliseconds, and also the whole process should be this fast: you don’t want an interstitial load to take forever in your application’s screen. This is to say, all of this happens, right before an ad appears on your screen. We’re talking about an almost imperceptible timeframe in the eyes of the user.
Yes, this is only a small part of it. But before we go any further let’s get into bids.
So let me get this straight, advertisers make payments to DSPs which in turn make payments to exchanges which in turn make payments to the publishers? Who and what is actually being paid? Lets focus on the demand side.
The most early form of payment was the CPM (cost-per-mile): advertisers paid a fixed amount for a thousand prints. This didn’t last long as limitations soon became evident: my Bay area delivery app does not get much benefit from impressions in the midwest.
Shortly after, we got CPC (cost-per-click): advertisers get charged for the clicks on the ads shown. This made a bit more sense, in the Bay App example, only people from that area would have an incentive in clicking my ad.
There was still room for improvement. Cue CPA (cost-per-action); advertisers pay if predetermined key events occur (e.g. installing the app or -for a ride app- asking a car). Campaigns have evolved to become more customizable. For example, acquisition campaigns focus on app installation; retargeting campaigns focus on engaging users who seem to have interest in an app or product.
Still with us? We covered pricing for advertisers and although broadly, the same rules apply for other players in the chain. There are some possible complications. For example, what happens when different links in the exchange charge differently in the same auction.
Imagine an exchange charges the DSP on a CPC basis but advertisers reach a deal on a CPA basis. DSPs have to estimate a CPC cost which is in line with the charged CPA. Plus a profit!
When paying for ads, many teams will wonder about the effectiveness of said ads, when doing so, the term attribution comes up. We covered the basics of attribution and it’s various models in a previous post and will keep it simple as it’s not the main topic of the post. The idea behind it is to be able to identify the channel, ad or specific action that generated a conversion. When paying for ads one might want to know which ad is responsible for a conversion.
If you print an add and a conversion occurs soon after, that conversion might be attributed to that ad. But, what if the conversion occurs the next day? Week? Month? What if the user clicked on a handful of ads for the same campaign within a period of time; which one was the “cause” of the conversion? The last one shown? Are all ads equally responsible?
We’ve covered the different components and players as well as bids. It’s time to get into how these bid auctions actually work.
We all imagine a packed room with people holding up their signs along with a smartly dressed auctioner banging a wooden hammer for each sale: “Sold to the bidder in the first row”. No? Maybe that’s just me then. This is not what happens with real time bidding (RTB).
In reality, auctions are carried out following the “closed envelope” or “sealed bid” methodology: every participant submits an offer simultaneously, without knowing how much the others offered. The exchange “opens the envelopes” and sorts the bids from lowest to highest - this last one wins the auction.
In First Price Auctions, bidders who win pay exactly the price they offered. It is quite intuitive but can be unstable: If I’m willing to pay 100, and continually win doing so, shouldn’t I bid 99? I will probably still win and I will pay less. When generalised this behaviour can make for erratic behaviour in the exchange.
For this reason, Second Price Auctions (introduced in a seminal paper by Edelman and Osotrvsky) gained popularity. In this case, the winner is still the person who submits the highest bid, but is charged the second highest bid plus a cent. In other words, the winner is charged the minimum amount over what makes that bid the highest offer. This solves the mentioned issue: lowering my bid does not have any impact on what I’m charged. If there is room to lower my bid the second price mechanism will do it for me.
RTB is not limited to mobile ads, it can be found anywhere online ads are displayed. One of the oldest examples, which is now boosted, is sponsored search. Search mechanisms are everywhere: your ecommerce website, delivery apps, text search, etc. The reasoning is similar for sponsored search but the context changes. Some sponsored results will be shown among organic (non-paid content) results. If ads are irrelevant to the search then this might hurt UX. Typically, in this context bids are considered along some scoring metric in order to decide who the auction winner is.
The reasoning behind this is that even if someone offers a lot of money, if the content is irrelevant or of poor quality, it shouldn’t be shown. Typically, more than one slot is being auctioned, since you are modifying listings there are several positions for sale.
This listing mechanism can improve Second Price Auction Mechanisms. One example is VCG (Vickrey-Clarke-Groves). In VCG, the paid price is determined not only by the bid immediately below the winner, but by weighting the effects that displaying this ad may have on the whole listing: either by replacing a competing ad or displacing an organic listing.
Don’t worry we will make it quick.
Mutt Data’s specialised team of Data Nerds knows a thing or two about Real Time Bidding (RTB). Our founders and a handful of the first team members at Mutt Data were working in AdTech before they decided to journey into Mutt.
This gave us a special perspective on market evolution and the whys behind it. After all, several Mutters helped shape this very market. From simple per-impression pricing, bundled ad-spaces to auctions, Real Time Bidding (RTB) and audiences. We have lived and worked through countless changes and progressions and in the process we have studied the strengths and weaknesses of each alternative solution.
This experience has prepared our team to offer strategic advice on designing the best solutions and architectures for each client. We focus on making these systems cost-effective and tailored to our client´s current and future needs through scalable architectures.
Moreover, the team has experience working for both sides: demand and offer. In short, working on both systems for the advertisers who pay for the ad-spaces, and systems for publishers who sell said space, and platforms implementing their auction mechanism. This led us to work on several different components of the RTB ecosystem. On the demand side of the things: creation of audiences, conversion models, bid optimizers and pacing mechanisms among others. On the supply side, for example, implementation of waterfall mechanisms that chose in which exchange a given available slot should be offered.
Last but not least, besides the classic banners or interstitials shown in ads in an app, the Mutt team has worked both with sponsored listings (paying for an item to be featured on the top of your ecommerce or delivery app) and sponsored search (paying a search engine for items to pop up as recommendations in relevant queries)
You may have gathered that we nerd out a bit when it comes to Adtech and Real Time Bidding. Guilty as charged, our experience took us on a journey through several verticals, fields and different RTB ecosystem components and we are enjoying the ride.
We hope you’ve found this post useful, and at least mildly entertaining. Curious about how Real Time Bidding could be an opportunity for your business? At Mutt Data our specialised team of Data Nerds can help you, contact us here.
Does a remote-first job appeal to you? Looking for a young multidisciplinary and dynamic environment? Got some mad dev skills, and like applying machine learning to solve tough business challenges?
We´re Hiring! Check out our Lever account for open positions.