<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by Sarah Java Beans on Medium]]></title>
        <description><![CDATA[Stories by Sarah Java Beans on Medium]]></description>
        <link>https://medium.com/@sarahjavabeans?source=rss-c89d62ece11e------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/0*Rsx6UwxcXbrWQENt.jpg</url>
            <title>Stories by Sarah Java Beans on Medium</title>
            <link>https://medium.com/@sarahjavabeans?source=rss-c89d62ece11e------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Fri, 20 Nov 2026 02:57:11 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@sarahjavabeans/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[Introduction To Ads — A Software Engineer Perspective]]></title>
            <link>https://medium.com/@sarahjavabeans/introduction-to-ads-a-software-engineer-perspective-cf7980b8fdee?source=rss-c89d62ece11e------2</link>
            <guid isPermaLink="false">https://medium.com/p/cf7980b8fdee</guid>
            <category><![CDATA[advertising]]></category>
            <category><![CDATA[advertising-technology]]></category>
            <dc:creator><![CDATA[Sarah Java Beans]]></dc:creator>
            <pubDate>Fri, 10 Apr 2020 00:16:00 GMT</pubDate>
            <atom:updated>2020-04-10T00:16:00.882Z</atom:updated>
            <content:encoded><![CDATA[<h3>Introduction To Ads — A Software Engineer Perspective</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*s9MFLFjmGncfGH9NtCPUxg.png" /></figure><p>Overall the advertising and marketing space is huge and highly complex, so this is by no means comprehensive. You’ll find some quick insights into customer perspectives and a handful of technical challenges in the space, how to find your niche and distinguish yourself as an engineer in this space based on your interests and the myriad of challenges the space provides. I plan to write a series of articles in this space to share my knowledge and to spark discussions about motivations by role(s), how tech has shifted the landscape already and how it may continue to do so in the future, and how some general human propensities can lead to ethical challenges.</p><h3>Overview</h3><p>Advertising and marketing seem simple on the surface — a person or company has something they want to sell (products, services, information, etc) and they want to make the public aware of this in hopes of ultimately winning new business. Historically this was a simple model with fairly clear goals and participants. Like with all other industries the internet has dramatically changed the landscape in a number of ways, most importantly it has shortened the feedback loops within the space. By shortening the feedback loops the internet has allowed players in this space to “science” this space. While psychology and design concepts have always been a part of advertising, the internet has allowed companies to get an infinitely deeper insight into the “users” and draw direct correlations between that behavior and the types of ads/campaigns being run.</p><h3>Customers</h3><p>There are 3 main categories of customers — and again this is an oversimplification but it will give you a good starting point for understanding the customers. As you get deeper into the industry overall you will find that there are a multitude of different types of customers involved, but their desires generally fall into the categories below. Most of the customer complexity comes in where there are “people in the middle”. For example agencies may act on behalf of an actual advertiser, but when doing so they are still driven (at least in large part) by the goals of the advertiser they represent. Ad networks may aggregate many small publishers and may build out campaign management platforms and tools to help advertisers match to a subset of their inventory — imagine a multitude of influencer blogs that join an ad network to manage their inventory and find advertisers for them. In that case the ad network may behave like a publisher to their advertiser customers and may function similar to an advertiser from the perspective of their publisher partners. In these examples you can see that there may actually be a myriad of customers, but their ultimate motivations still generally fall in the categories below in some way or combination. When evaluating customer relationships here and features for those customers it’s important to understand how they fit into these categories and figure out where they make their money to understand their motivations.</p><h3>Advertisers</h3><p>People/companies who want to run an advertising or marketing campaign in order to influence user behavior or perception in some way.</p><p>Motivations:</p><ul><li>Sell a product or service</li><li>Direct — the advertiser wants people to see and interact with the ad such that they make a “conversion” (purchase, contract, install an app, etc)</li><li>Indirect — psychology studies have identified the “Mere exposure effect” or “Familiarity principle” (<a href="https://en.wikipedia.org/wiki/Mere-exposure_effect">https://en.wikipedia.org/wiki/Mere-exposure_effect</a>). In short this principle of human behavior exposes that humans develop an affinity for things they see more often. This understanding has led to an entire subset of marketing called “Brand Awareness” where instead of the goal of the campaign being to trigger (and measure) immediate conversions instead the goal is to simply keep your brand visible and familiar as a way to create a feeling of trust on the part of users viewing the ads.</li><li>Convey information — an example of this would be political ads in a democratic society. Politicians want to leverage the Familiarity principle to create a feeling of trust, they may want to advertise specific political platforms to inform users of their position, and lobbying companies will likewise attempt to sway the perception of users by conveying some information about a politician or political party. There have also been informational campaigns for environmental issues — like the Smokey The Bear commercials of the 1980s, recycling information, anti-smoking and health-related informational ads, etc.</li></ul><h3>Publishers</h3><p>People/companies who have inventory on which to run ads in order to monetize their platform.</p><p>Motivations:</p><ul><li>Generate an income stream from their platform. If you look at all of the services that Google provides — email, maps, search, Google-fi, etc — stop to think about how it’s funded. They do have some direct revenue outside of advertising (say enterprise usage of the Google Cloud), but the company was founded on and still sees a massive income stream from their advertising.</li></ul><p>Additional considerations:</p><ul><li>Reputation — although publishers want to monetize, the owner of the platform with inventory generally wants to restrict the types of ads shown such that they protect their reputation in some way. For example they may not want sensitive content on their platform (like adult ads), or they may not want to run ads from competitors on their platform, etc.</li></ul><h3>End Users</h3><p>People who view and interact (maybe) with ads. Since we don’t have too much of a choice on if we are shown ads we’ll look more at the benefits and drawbacks ads may have on end users.</p><p>Benefits:</p><ul><li>Instead of having to pay money for an online platform or service, users may agree to be shown ads in order to gain the benefits of a platform. One significant example of this is that many people use Facebook without a subscription fee, but the operating cost (and profit) is instead driven by the ads.</li><li>May learn about new products/services/information.</li><li>May appreciate coupon ads that give them a further discount on some product or service.</li></ul><p>Drawbacks:</p><ul><li>Many platforms still do not offer users an ad-free option, so users are forced to consume the ads. Again, Facebook is a good example of this.</li><li>Psychology is used to optimize the impact of ads. This creates a very fine line between providing users with relevant marketing vs. manipulation. The problem here is one of education — marketers become experts in human behavior (in ways) but the users themselves are unaware/uneducated about this. I highly recommend reading Predictably Irrational (<a href="https://www.amazon.com/Predictably-Irrational-Revised-Expanded-Decisions/dp/0061353248">https://www.amazon.com/Predictably-Irrational-Revised-Expanded-Decisions/dp/0061353248</a>) as it will open your eyes to how the marketing industry knows more about your behavior than you do (we all like to think we’re rational and have free will, but…)</li><li>Privacy — users may not grasp how much or what kind of information they are agreeing to share when they interact with ads.</li><li>The depth of targeting that the internet has facilitated has allowed advertising and marketing to shift into the individual echo chamber. Without the fast feedback loop and rampant detailed user data available on the internet, in general ads used to be shown to more people. The onset of targeting optimization creates a vehicle for shady advertising behavior to go relatively unnoticed. If an advertiser buys a commercial during the Super Bowl their message is very visible to the general public. However if an advertiser wants to target 5 people on Facebook they can do just that — and convey messages to a select few that they don’t necessarily want to be shared to the broader public. This subtlety enables an advertiser to share information in a way that they don’t have to be accountable to the general public.</li></ul><h3>Systems</h3><h3>Campaign Management</h3><h3>Campaign</h3><p>A campaign can generally be thought of from the perspective of an advertiser. They set goals, set a budget, find inventory, and sign a contract with one or more publishers to serve their ads. Campaigns are usually confined to a specified timeline and have some kind of budget cap.</p><h3>Impressions</h3><p>When an ad renders in a place visible to the user, that is considered an impression. The user need not take any action for an impression.</p><h3>Interactions</h3><p>When click tracking became available to the industry, many people wanted to focus on (and pay for) clicks instead of impressions. A click means that the user clicked on the ad and viewed the next chunk of content provided by the advertiser. It’s a stronger signal that the user was aware of and had the ad present in mind. This is measured and reported on as a “click through rate” — the number of clicks that happened in relation to the number of impressions served. However, some studies have demonstrated that CTR does not necessarily correlate to conversion rate.</p><h3>Conversions</h3><p>If a user installs an app, or makes a purchase, or otherwise realizes the advertisers goal this is considered a conversion.</p><p>Further reading on CTR and conversion rate: <a href="https://www.rocketclicks.com/client-education/ctr-vs-conversion-rate/">https://www.rocketclicks.com/client-education/ctr-vs-conversion-rate/</a></p><h3>Types Of Ads (Delivery Guarantees + Risk)</h3><p>Selecting the type of ads to run as an advertiser or to support as a publisher comes down to risk tolerance and risk ownership.</p><h3>CPM</h3><p>CPM is Cost Per Mille impressions — <a href="https://www.marketingterms.com/dictionary/cpm/">https://www.marketingterms.com/dictionary/cpm/</a></p><p>This is the model for impression based advertising. The way this works is it’s a guaranteed delivery of impressions and a guaranteed payment based on impressions delivered.</p><p>At the end of the day, literally all ad type models will ultimately run impressions, so think of CPM as the foundation for all ads.</p><p>Benefits:</p><ul><li>Great for brand awareness or informational/educational campaigns where the value is in reaching a wider audience.</li><li>Usually a straightforward contract where the payment is tied to the number of impressions served.</li><li>Usually the cheapest way for an advertiser to run ads.</li></ul><p>Risks:</p><ul><li>If the goal is some sort of user action, the payment structure for CPM is NOT directly coupled to the performance goal. The implication to this is that the risk is fully owned by the advertiser in this situation. The publisher has no motivation besides serving impressions to contract, while the advertiser is the one who cares about and assumes the risk for who is actually consuming the impressions, doing targeting optimization, and even figuring out how to indirectly measure the impact the ad campaign has on their ultimate goals.</li><li>The publisher risk in this model is just to deliver the impressions to the contract. If something unforeseen happens and the publisher is unable to serve the impressions promised then they are on the hook for that. The publisher is motivated to serve enough impressions to meet the contract, but they can optimize the profit by serving the minimum amount needed for each customer, keeping the inventory available for more customers. Fraud is the biggest risk here for the publishers (fake views).</li><li>Fraud: The risk here is of some automated robot that crawls pages in order to generate an impression instead of an actual user seeing the impression. Sometimes this can be intentional by bad actors — and in this type of campaign it would be the publisher who would benefit from inflated impressions, thus publisher reputation is important to advertisers in this model. However this can also happen organically without a bad actor because of all the search engines/spiders crawling websites.</li></ul><h3>CPC</h3><p>CPC is Cost Per Click — <a href="https://www.marketingterms.com/dictionary/cost_per_click/">https://www.marketingterms.com/dictionary/cost_per_click/</a></p><p>This is the first step away from guarantees because it depends on users interaction with the ad — outside the control of either advertiser or publisher. In this model the advertiser pays the publisher based on the number of clicks that their ads generated.</p><p>Benefits:</p><ul><li>Theoretically a stronger impact is left on the user if they paid enough attention and interest to the ad to click on it.</li><li>Easy to measure and to contract on.</li></ul><p>Risks:</p><ul><li>First, clicks are not necessarily correlated to conversion. So if an advertiser has an ultimate goal of a user to say install an app — there’s not a clear correlation that if a user clicks the ad that they’ll install the app, so this model is not driven by the ultimate goal.</li><li>This shifts more risk on the publisher. The publisher will be running impressions, but only getting paid for which ones are clicked on. Due to this shift, the publisher becomes more motivated to care about targeting. Publishers want more targeting options for advertisers and may even go so far as to offer targeting optimizations for their customers. Publishers may be motivated to build their own targeting segments based on their user base.</li><li>Fraud: Click stuffing is the fraud potential here. Particularly on mobile there have been a number of news stories about publishers including SDKs in their app which are generating clicks on ads as a background process while the user is in the app. I can’t think of any unintentional ways this could happen. However it’s important to note that “publisher” is the concept here, not necessarily the actual owner of the inventory but rather who’s making the money from the click happening. So an app owner with a fraudulent library/SDK integration may be unaware of the fraudster in the middle, but if the fraudster has enough knowledge of the click tracking URLs they can modify the URL to look like it’s coming from their source and they get the payout.</li></ul><h3>CPI &amp; Conversions (Performance Marketing)</h3><p>The next evolution is Cost Per (Conversion) — for example CPI (Cost Per Install). This has become more popular over the last 5 years. In this model the advertiser pays for precisely the user behavior they want, be that an install or a purchase or registering a new account, etc.</p><p>Benefits:</p><ul><li>The advertiser pays for exactly what they want and they don’t have any motivation to have to manage their own campaign optimizations.</li></ul><p>Risks:</p><ul><li>From the publisher perspective this shifts literally all the delivery and optimization risk on them. The advertiser pays for what they want and it’s on the publisher to figure out how to maximize that.</li><li>A conversion event may be a cross-platform behavior or tied to the purchase activity of the user. Because of this, tracking changes from just following a single user click and instead morphs into tracking a sequence of user actions and then on top of that determining the attribution of which publisher actually led to the conversion. Since the click is decoupled from the behavior tracking becomes much more complicated and entire attribution systems and providers have cropped up in this space.</li><li>Because there may be a number of ways to drive the conversion the advertiser wants, this has opened the door for many “people in the middle” to jump in and try to grab a chunk of the ad budget. Small companies and individuals and ad networks crop up in this area and will contract separately with the advertiser and with publishers, where they negotiate for higher payouts from advertisers and then either run the cheapest remnant CPM campaigns they can — hoping to pay out less for the CPM than they make from the advertisers for the conversions. Because there are so many parties motivated to jump in here this leads to fighting for attribution and much more competition…which naturally leads to more creative and complex fraud.</li><li>Lastly, because this model requires some connection between ad interaction and conversions, GDPR and other privacy initiatives and the pressure for device providers to allow users to reset their advertising id or even prevent ids and other custom ad tracking parameters from being passed through the click tracking systems can have a huge impact on this overall market. This model may be very rewarding to the people willing to take on the optimization risks, but it’s also a rather risky model to stake an entire business on given the ongoing privacy and user data laws and lawsuits.</li></ul><h3>Insertion Order</h3><p>The insertion order is the contract. It usually defines a date range, inventory (where the ad is shown, which IAB standard ad type(s) are supported, etc), and some set of targeting parameters (day part targeting, user segments available to be applied, delivery spread algorithms, etc).</p><h3>Creatives</h3><p>The content of an ad may include visual and/or text assets. These assets are usually considered the “creative”. Advertisers may provide the full creative or assets that can be used to build the creative (like logo files), or they may contract an outside company to build their creative for them. Some of the larger publishers provide simple tools for enabling advertisers to build their own creative in the platform.</p><h3>Targeting + Optimization</h3><p>Targeting is the core of any online advertising. Regardless the model chosen, someone somewhere will be motivated to try to narrow the ads down to the users they believe are most likely to click or convert. Optimization is the process of adjusting targeting throughout the campaign in an attempt to apply the ad budget to the most effective channels or audiences. This wikipedia page has a lot of really good information about types of targeting: <a href="https://en.wikipedia.org/wiki/Targeted_advertising">https://en.wikipedia.org/wiki/Targeted_advertising</a></p><h3>Audience Segment Targeting</h3><p>One of the most common types of targeting is to render the ads to a subset of the users of the publisher’s platform that may be more receptive or likely to click or convert. Targeting segments are built in a way that a user identifier (device id, email address, etc) can be hashed and that hash can be associated to a segment. When an impression is about to be served ads may be narrowed down based on the desired use segment for the ad and the segment(s) associated to the user who’s about to receive the impression. This is an ethics area where gender, race, and age come into play. How the user data is mapped to segments is also a key area of system design — this is where privacy systems and designs come into the picture.</p><h3>Keyword Targeting</h3><p>For any kind of search-based systems there is “keyword” targeting. Instead of needing to have awareness of a specific user (or in addition to that) search-based systems may just target ads based on what the user entered into the search bar. This is powerful because it can be effective while maximizing user privacy.</p><h3>Geo Targeting</h3><p>Advertisers may only be interested in reaching people in certain countries or cities. Imagine a local restaurant with only a single location — they may only want to run their ads for users who are in that city or county or region. This can generally be done anywhere from a country level down to zip code levels. HOWEVER, a word of caution here — at least in the United States a zip code may contain an imbalanced population — by race or socio-economic status. So the owner of a campaign management system needs to be very careful about how they allow this to be used, and some companies just don’t allow zip code targeting as a result. If a lawsuit were to be filed due to racial profiling/targeting, the campaign management system owner could be called out. Before building this option into a system it would be a good idea to review the risks with the legal team at your company and use their guidance for how much or how to offer this.</p><p>One reference for the zip code “redlining” is mentioned in this article:<a href="https://www.theverge.com/2018/8/21/17764480/facebook-ad-targeting-options-removal-housing-racial-discrimination"> https://www.theverge.com/2018/8/21/17764480/facebook-ad-targeting-options-removal-housing-racial-discrimination</a></p><h3>Daypart Targeting</h3><p>Imagine you have an app for a food delivery service. You may want to run your ads around common meal times. This is what daypart targeting solves for. Can be days of the week or hour ranges within a day.</p><h3>Delivery Spread Algorithms</h3><p>When running impression based ads, if you have a 10 day campaign and a 100 dollar budget, you could conceivably spend your entire budget and serve all of your impressions within the first minute of ad serving. Usually that’s not desirable, so part of campaign optimization is selecting a delivery spread algorithm. The simplest being the even spread, so if you have a 10 day campaign, and a 100 dollar budget you want to divvy it up so you spend maybe 10 dollars a day (however many impressions that divides out to based on cost). By spreading out the serving of impressions this way it allows the advertiser or campaign manager time to optimize the targeting while the campaign is running.</p><h3>Frequency Capping</h3><p>Impressions are user agnostic. So you could serve 1000 impressions to 1000 different users or you could serve them all to a single user. Users may find it annoying to be presented with the same ad over and over again. Usually advertisers are hoping to reach many people, so they may want to set a frequency cap for how many impressions of this ad a given user may see during a time frame (like 2 impressions per day or 1 per hour, etc). This technique is called frequency capping.</p><h3>Max Bid</h3><p>For any particular ad spot — it may be worth different amounts at different parts of the year or even within a day. So the cost of an impression can vary at the baseline level but also by demand (if a bunch of other ads are competing for that spot at the same time). When there’s more demand than supply a bidding model works nicely to resolve a winner and maximize profit for the publisher. This is achieved by real time bidding systems inside the ad server. The advertiser or campaign manager may set a range from one penny up to a maximum bid amount per impression. If there’s competition the price goes up, but this way the advertiser can ensure they don’t spend more per impression than they want.</p><h3>Anti-Targeting</h3><p>Anti-targeting is used by both advertisers and publishers. An advertiser may not want to show their ads on the same page as a competitor’s ads. So if you’re a publisher with more than one ad on a page/search result, you’ll need to provide a way for advertisers to control who they don’t want to appear with. And as a publisher you may want to control your own brand such that you don’t show ads for your own competitor, or you don’t want to show adult content ads, etc. Because this is a feature desired by both publishers and advertisers it’s common in campaign management feature requests.</p><h3>Ad Serving</h3><h3>Takeovers</h3><p>Takeovers are usually expensive and essentially is a guarantee of all impressions for the inventory — not a guarantee for a specified quantity. No targeting is needed for takeovers — the ad server only needs to know how long and which ad to apply to the takeover. In general takeovers are very easy for ad serving but extremely costly to the advertiser. You can also think of influencers writing a blog post or posting a video review of a product as a takeover.</p><h3>Real-time Bidding</h3><p>When you have more than one ad competing for an impression the tie breaking is usually done via real-time bidding systems (RTB). This is how the market drives the value of your inventory.</p><h3>Readability</h3><p>Ad servers can be very complex because there’s a lot going on at runtime and it has to be ultra-fast. Ad servers are some of the most cutting edge technology because they have a high request rate while trying to do heavy processing. Because you’ve got targeting, anti-targeting, real-time bidding, delivery spread, etc — it’s crucial to make this code readable. If something breaks here it’s got to be fixed fast — you will be under contracts with advertising partners for delivery, and will be under pressure to serve the ads fast so you don’t negatively impact users or user behavior. For these reasons, readability should be a top priority — you can’t afford a long time troubleshooting. Deploying to the ad server should also be done after rigorous testing — so this should never be rushed out the door or hacked in.</p><h3>Latency</h3><p>Latency is the most important feature of an ad serving system. All targeting should be pre-processed — you do <strong><em>NOT</em></strong> want to have to perform database lookups and analyze the data to find the most relevant ad. To address this the industry uses targeting segments. Targeting segments are pre-processed collections of users (generally hashed PII so it can’t be reversed), such that you pass in a user identifier to tell if that user is part of that segment. Be extremely cautious if you decide to accept targeting segments from customers or any other 3rd party — would recommend working with legal before even considering that.</p><h3>Reporting, Billing, &amp; Reconciliation</h3><h3>Reporting</h3><p>Advertising partners are going to want as direct and clear as possible metrics and reports to understand their ROI/ROAS (return on investment/return on ad spend).</p><p>When working on reporting solutions, it’s important to understand who owns what data. Even though the ad serving systems may collect the data, generally the ad servers don’t “own” the data. So it’s important to work with legal to get a clear understanding of who owns what parts of the data as it passes through your systems as that will be key to what kinds of reports you can construct, how they have to be saved (encrypted at rest or can they even be saved at all), and who has access to view what reports.</p><h3>Fraud detection</h3><p>Fraud happens in many ways and for many different reasons. One of the common reasons for fraud is for publishers to attempt to maximize their revenue by inflating their impressions and/or clicks, thus getting paid for delivery or last-touch conversion attributions even though their user may not have even seen an ad. Fraud detection is such a huge area there are entire companies dedicated solely to offering products and services to analyze ad traffic and detect fraud. Below are just a couple of articles that introduce some fraud concepts.</p><p><a href="https://www.adjust.com/blog/a-common-sense-approach-to-mobile-ad-fraud/">https://www.adjust.com/blog/a-common-sense-approach-to-mobile-ad-fraud/</a></p><p><a href="https://www.linkedin.com/pulse/how-avoid-fraud-mobile-app-marketing-click-injection-stuffing-oded">https://www.linkedin.com/pulse/how-avoid-fraud-mobile-app-marketing-click-injection-stuffing-oded</a></p><h3>Reconciliation</h3><p>Reconciliation is the process whereby contract fulfillment is assessed. When the contract is very simple, as in performance marketing, then these systems may be very simple. However these systems can get more interesting when there are any impression- or click-based aspects of the contract. With those at play the time targeting and delivery spread must be evaluated and fraud detection must be factored into the delivery (ie, fraudulent activity is to be removed and any delivery outside of the contract must also be removed as it’s not billable). After the extra data is removed then there is an evaluation of if the contract was fully served, and if not then there becomes a reporting and negotiation on how to handle the parts not met. That may come in the form of discounts on the current bill or future bills, make goods to offer alternate services or campaign extensions, etc. This is a people-process but relies on heavy data analysis and reporting.</p><h3>Payments</h3><p>Payments may be brokered through third parties, be direct between the individuals involved, and they may be made at time of a conversion or made in aggregations monthly or quarterly. The form and time of payments is part of the contract negotiation along with reconciliation and contract fulfillment. From a technical perspective this includes invoicing systems and actual money movement systems. In very large publisher systems there may also be custom credit and/or prepayment arrangements.</p><h3>Finding Your Niche</h3><p>The ecosystem provides interesting challenges for all types of software engineers.</p><p>What motivates you as an engineer?</p><p>What kinds of problems do you like to solve? People problems or technical problems or process problems?</p><p>Campaign management systems offer some complex front-end work, enables engineers to get a feel for the overall business and customers and understand their motivations. These systems usually have less traffic themselves than the ad-serving, but do get to work with big data in the end where reporting, fraud detection, and reconciliation come into play.</p><p>Ad server systems provide some rich technical challenges because you have big data with very strict latency requirements. The ad server is where the incoming data (user-based or keyword-based) comes in and an ad needs to be selected, and then rendered, all in the matter of milliseconds. If you’re the type of engineer that doesn’t care about the “what” so much as the “how”, or you nerd out over algorithms and performance optimizations the ad server will be fun for you.</p><p>Targeting segment building systems provide a challenging puzzle, rich in data and the opportunities for engineers in this space include creative ways to target faster (how much can be pre-processed? how can targeting be architected to be fastest at runtime?). Largely this is also an interesting area in terms of data science and machine learning opportunities. Engineers in this space will have to find the right balance between building segments that customers want vs. applying ethics in limiting what they can offer or how they offer it.</p><h3>Ethics</h3><p>You’re going to be presented with all kinds of fun challenges in this space and you will be rewarded for optimizing systems for advertisers and/or publishers. The most crucial takeaway here is that you should ALWAYS remember that you are an end user and ALWAYS remember that the end user is the most valuable actor in the system but has the least power (be it by lack of choice or lack of education of the psychology behind the industry).</p><p>It’s really easy to get excited over the variety of puzzles/challenges in this space, so it’s extra important to stop yourself periodically and reflect on the ethics of what you’re trying to solve. Just because we have the ability to do certain things and just because the problems may be fun — does not mean that we should do them from an ethical standpoint. Keep in mind that when we are running experiments or attempting to improve targeting — we are actually conducting psychological studies of human beings and attempting to then leverage that learning in order to modify the behavior of large groups of people. With great power comes great responsibility.</p><p>A couple of user privacy gotchas:</p><p><a href="https://www.forbes.com/sites/kashmirhill/2012/02/16/how-target-figured-out-a-teen-girl-was-pregnant-before-her-father-did/#52e6fdbc6668">https://www.forbes.com/sites/kashmirhill/2012/02/16/how-target-figured-out-a-teen-girl-was-pregnant-before-her-father-did/#52e6fdbc6668</a></p><p><a href="https://www.adweek.com/agencies/a-brief-rundown-of-racism-within-advertising-and-why-its-still-happening-today/">https://www.adweek.com/agencies/a-brief-rundown-of-racism-within-advertising-and-why-its-still-happening-today/</a></p><h3>Further Resources</h3><h3>Fraud</h3><ul><li><a href="https://www.adpushup.com/blog/types-of-ad-fraud/">https://www.adpushup.com/blog/types-of-ad-fraud/</a></li><li><a href="https://www.tune.com/blog/types-of-advertising-fraud/">https://www.tune.com/blog/types-of-advertising-fraud/</a></li></ul><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=cf7980b8fdee" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>