programmatic quality guide: dstillery

28
QUALITY GUIDE p r o g r am m a t i c how to improve the quality of Campaigns with

Upload: sean-lough

Post on 27-Jan-2017

104 views

Category:

Marketing


0 download

TRANSCRIPT

Page 1: Programmatic Quality Guide: Dstillery

Quality GUIDE

programmatic

how to improve the quality of Campaigns

with

Page 2: Programmatic Quality Guide: Dstillery

quality in advertising since 2008

Page 3: Programmatic Quality Guide: Dstillery

What does quality mean in marketing?

Good creative? Reaching the right audience? Viewability? Running campaigns in brand-safe environments?

All of these are inputs. We tend to focus on outputs. To us,...

Quality in marketing delivers real results that build brands and build businesses.

As a pioneer in programmatic marketing for brands, Dstillery has been delivering quality in advertising – “only the good stuff ” – since 2008. Last fall, we released the Inventory Quality Report, a revealing look at fraud in our industry and a guide to how Dstillery is leading the eff ort to thwart fraud in its many forms.

That report established some standards for quality in digital marketing.

Since then, quality has become a consistent theme in our industry:

• The ANA and White Ops released a study fi nding that 25% of video impressions and half of third-party sourced traffi c is fraudulent.

• A Google study found that fewer than 50% of ad impressions are viewable.

• Kraft said it rejects up to 85% of impressions because of poor quality.

But inventory quality is just one part of the equation.

At Dstillery, we’re committed to driving Quality ROI: the real results that build businesses and earn the trust of marketers. Our relentless focus on quality not only enables us to create and reach the right audiences, but also powers our predictions of who will be most likely to become your next customers, and when best to reach them.

In the pages that follow, you’ll learn how to improve the quality of your campaigns, and how a focus on quality can translate into increased marketing eff ectiveness. Let’s raise a toast to quality.

Cheers!

Louise Doorn, CMOOn behalf of all your friends at Dstillery

INTR

OD

UC

TIO

N

Page 4: Programmatic Quality Guide: Dstillery

TABLE of CONTENTS

PROGRAMMATIC QUALITY GUIDE

Page 5: Programmatic Quality Guide: Dstillery

1. 2. 3.AUDIENCEQUALITY

INVENTORY& DELIVERY

CROSSSCREEN

4.DRIVINGROI

Page 6: Programmatic Quality Guide: Dstillery

photo courtesy of IMDB

5 | AuDIeNce QuALIty • CROSS-SCREEN • INVENTORY & DELIVERY • DRIVING ROI

“ Anyone with an internet connection and an idea can develop an audience.”

- kevin spacey, Actor

Page 7: Programmatic Quality Guide: Dstillery

1. AuDIeNce QuALItyWhy Obsessing Over Audience Quality Drives ROI

Quality audience development – how the right audiences are defi ned, identifi ed, and targeted – is the backbone of any successful campaign.How do you ensure you’re reaching the right audience?

1. DEFINE YOUR GOALSThe fi rst step is specifying the population you want to replicate. The more specifi c the better, because this population serves as the seed data used in prospecting models to fi nd your next best customer. Prospecting expands your target audience and scales the reach of your message.

2. TAKE THE GARBAGE OUT: DATA HYGIENEBuilding quality audiences and prospect models requires quality data. This may sound obvious, but because of the complexity of data-driven marketing, it’s often overlooked.

We never assume that the data we use is correct, even when we partner with industry-leading data providers. Instead, we test and scrub data before it enters any of our systems. We look for three primary breaches of data hygiene:

2A. NON-HUMAN TRAFFICWe scrutinize online cookie data to ensure the cookie corresponds with a real human, not a bot. The foundation of all prospecting models is the data from these cookies, including site visits and web conversions. The most insidious and prevalent threat to cookie accuracy is web traffi c sourced from bots. We have established standards for realistic human activity, and applied them to all cookie behavior. We examine clickthrough rates, site visit patterns, and number of sites visited in a given period to identify cookies that look suspicious. We put the suspicious cookies in a “penalty box” – a time-out of sorts – until we have evidence that they’re human.

PROGRAMMATIC QUALITY REPORT

Page 8: Programmatic Quality Guide: Dstillery

7 | AuDIeNce QuALIty • CROSS-SCREEN • INVENTORY & DELIVERY • DRIVING ROI

50% of location signal data can be innaccurate

This process ensures we’re delivering real ads to real people, and that our data inputs are accurate. White Ops recently found that we delivered more than 97% of ads to humans, a figure that surpasses not only all of the other programmatic vendors but also the average direct publisher buy.

2B. STALE SIGNALSCookie data needs to be fresh. What a consumer did this morning or yesterday tends to be far more relevant for marketers than what she did last month or last year. We score cookies every day for every brand. We look for signals that a user is no longer in-market for a given product or service, and we take them out of the target pool. Prefab segments are the standard in the industry, but they are by definition stale, and they consistently underperform our up-to-the-minute models.

2C. ERRANT LOCATION DATAScrubbing location data is as important and nuanced as cleaning up the cookie pool. Mobile companies, app companies and data providers all share location data on consumers via their smartphones and tablets.

Much of this data is fabricated or erroneous. In fact, we’ve found that 50% of the location signal data we

receive is inaccurate. While other companies play along – attaching a location signal to a consumer increases the value of that consumer to a marketer – we are ruthless about scrubbing out the errant data. What comprises that 50% inaccuracy? A wide range of deception and errors. For example, we receive raw data that indicates that thousands of people are in a location where they simply can’t be. The geographic center of the continental United States is a case in point. It’s a field in Nebraska, in the middle of nowhere. The data we’re given indicates that tens of thousands of people are gathering in this field every day. This erroneous data can be generated when people decline to share their data, so the cell phone defaults to “U.S.,” which is all the device knows about that user. Or it can happen if the device can’t locate itself with enough accuracy – its location data will default to whatever level of precision it can provide – city, state or in this case country.“Third-party location data can be dramatically imprecise, depending on the methods used to determine that location’s lat/lon,” explains Daniel

Yi, Dstillery Data Analyst. “All of our location data is verified by human beings. We validate that each reported location actually exists where it says it is on a map, and if necessary we adjust the lat/long to rooftop accuracy of any physical buildings.”

“ There is no such thing as an average customer. As our tools for learning about consumers improve, we find that the customer population is more diverse than conventional wisdom would suggest.” – Brian d’Alessandro, VP of Data Science, Dstillery

All of our location data is verified by human beings

Page 9: Programmatic Quality Guide: Dstillery

– 1 –

– 2 –

PROGRAMMATIC QUALITY REPORT

checkLIst Audience Quality: Use fresh customer intelligence (online KPIs and offline

behaviors) to inform your au dience building models

Build audiences using verified human cookies and accurate location data (tested and scrubbed)

Employ real-time scoring of audiences vs. stale segments to reach your best prospects at the right moment

Beyond removing large swaths of erroneous data, we do a lot of work to contextualize and verify the location data we receive. In fact, our Location Still – the data repository where we store all of our location and place data – goes through rigorous human QA, making sure the location data we use matches the actual places where we find your audiences.Our human and technological checkpoints are constantly on the lookout for data errors, so your campaigns reach the right target audience, using models built from accurate consumer signals – both online and in the real world.

3. BUILD YOUR MODELSOnly with quality data can you build quality data models. The models allow us to identify new consumers for prospecting by first building a unique signal for your brand. Your brand signal is developed using billions of data points, from expressed behaviors important to your brand (e.g., retail store visits) to actions derived from first-party digital data (e.g., website visits, social activity, online conversions, desktop and mobile activity) that differentiate your top prospects from the rest of the world.

Page 10: Programmatic Quality Guide: Dstillery

You need some level of diff erentiation when building a target audience. Otherwise your target market is everybody, which would be a complete waste of time and money. So, we look for ways to narrow the target.Thankfully, today we know much more about people. The technology we’ve developed allows us to scale the old idea of segments out to who is truly interested in your product, at a much more granular level than geography or demographics.In fact, there is so much data that no human could possibly sift through all of it. We need computers to analyze the data. We use your existing set of customers as a seed for building new audiences. We use technology not just to fi nd out who they are, but more important to fi nd out how they are diff erent or special from the average consumer.We let computers sift through billions of atomic behaviors – people’s actions and choices. We look for behaviors that diff erentiate your product. Once the model fi nds

9 | AuDIeNce QuALIty • CROSS-SCREEN • INVENTORY & DELIVERY • DRIVING ROI

the secret sAuce: How to Build Incremental Audience

By Claudia Perlich, Chief Data Scientist, DstilleryOne benefi t of programmatic marketing is the ability to deliver incremental audience. But how do we go about building expanded audiences? While the math and data science is quite complicated, the concepts behind audience expansion are actually quite intuitive.When you think about reaching an incremental audience, the fi rst thing to recognize is who that audience is not. It’s clearly not the brand’s existing customers; they wouldn’t be incremental. It’s also not people who have absolutely no interest in the product. For example, there are people who love reading books and aren’t interested in sports. If I’m selling sports products, those people are probably not the audience I want to reach. When we think about building incremental audience, there has to be a basic interest in the product we are trying to sell. So who to target? In the past, marketers would consider who the product is for – a certain demographic group of a certain age, perhaps, or a certain segment of consumers with an interest, like sports or books. Marketers would do segmentation analysis to understand that, say, their customers tended to be females in their 20s, or men over 50. But that actually doesn’t tell us very much – even through a lot of your customers may be women, there are also a lot of women who don’t buy your product!

Page 11: Programmatic Quality Guide: Dstillery

out, “Oh, your people visit this store or these websites at greater rates than other consumers,” those become relevant signals that we feed back into the system to generate your unique Brand Signal.Then we rank everyone in the market based on those diff erentiating behaviors. We leave out the top group of consumers – your existing customers – and target the next level, those consumers who are likely ready to purchase your product.The advantage in this approach to incremental audiences is that you have much more precision in defi ning who they are, and it is much easier to get them to try your product than lower scoring consumers.

In the end, every programmatic platform has access to the same 300 million people in the U.S. But access isn’t a diff erentiator. The diff erentiator is knowing which one million or 10 million people to target out of the 300 million.That’s why the data that feeds the models is also a diff erentiator. At Dstillery, we have data partnerships with providers that give us a very diff erent view of what’s happening than what you can see on the exchanges alone. Ultimately, quality depends on having good data – and data scientists who know what to do with it.

PROGRAMMATIC QUALITY REPORT

\ soop r-man ’fekt \ noun :When one mobile device zooms across the map, seeming to travel dozens of miles in mere milliseconds. this occurs when location data reported by an app is based solely on an iP address or a user registration, or is hard-coded by a developer, instead of recorded through GPS or Wi-fi geolocation methods. (For more on The Superman Eff ect, check out our blog.)

superMAN effect

THE BOTTOM LINE: BETTER PRECISION MEANS FEWER WASTED IMPRESSIONS.

The differentiator is knowing which one million people to target out of the 300 million. - Claudia PerLICH

Page 12: Programmatic Quality Guide: Dstillery

Device Matching: IP is just the beginningFreshness + Frequency = Better Accuracy

11 | AuDIeNce QuALIty • CROSS-SCREEN • INVENTORY & DELIVERY • DRIVING ROI

Page 13: Programmatic Quality Guide: Dstillery

Think about all the laptops, tablets, and smartphones hanging out at your local Starbucks. Technically, these devices are related, since they are all using the same IP, but the web browsing activity of one user has nothing to do with the location activity of another. The resulting profi le would be neither useful nor valid to marketers – except, maybe, to Starbucks. And what about when customers leave Starbucks? What if you want to take a mobile audience and serve ads to them at home on their laptops, or to continue the conversation with desktop users on their mobile devices? Eff ective marketing reaches the right consumers wherever they are. This necessitates

connecting devices – including desktop, smartphones, tablets, and connected TVs – to create a single consumer profi le.

The solution? At Dstillery, we’ve developed proprietary CrossWalk technology to match devices in the most intelligent and accurate way possible.

DEVICE MATCHING: IP IS JUST THE BEGINNINGDevice matching starts with an IP connection — as soon as two devices are seen on the same IP, they are connected. But that connection might be weak or strong. And as you can imagine, simple IP matching creates a lot of false connections.The popularity of the IP address is only one of the criteria evaluated when making a device match. If too many devices or too many cookies share the same IP, the are ignored — no accurate connections can be built out of that noise.We take extensive measures to ensure your location and cookie-based audiences are built using accurate, verifi ed data.

... desktop, smartphones, tablets, and connected TVs... create a single consumer profile

2. CROSS-SCREEN QUALITYCrossing Screens with Confidence

PROGRAMMATIC QUALITY REPORT

Page 14: Programmatic Quality Guide: Dstillery

13 | AuDIeNce QuALIty • CROSS-SCREEN • INVENTORY & DELIVERY • DRIVING ROI

FRESHNESS + FREQUENCY = BETTER ACCURACY Once we determine an IP is useful, with a reasonable amount of devices shown on it we evaluate device activity on the IP – namely, how recently devices have been on it, and how frequently they are seen on the IP. If we see a low-populated IP with the same laptop and smartphone connected to it every day for the past few weeks, it’s a pretty safe assumption that the devices belong to the same household, and the right user profi le can be constructed from this match.

GO CROSS-SCREEN WITH CONFIDENCEOnce a user profi le is constructed, the fun part begins. With our CrossWalk device-matching technology, you can target a desktop user accurately on her smartphone, continue the conversation with a mobile user on her desktop back at home, or deliver sequential messaging across screens. You can even

build profi les of multiple users traveling to attend the same concert, then continue to message them across their devices after the show is over.Quality device matching enables quality cross-screen campaigns.

Figure 1

Figure 2

Quality device matching enables quality cross-screen campaigns.

CROSS-SCREEN QUALITY IN ACTIONAdobe / Goodby Silverstein & Partners

adobe wanted to connect with senior-level decision-makers attending the dreamforce conference in real-time across screens.

“ We wanted to get in front of our audiences while they were at dreamforce the second we knew they were attendees. the fact that dstillery was able to accomplish this in real-time was remarkable.” - Victoria Barbatelli, Communications Strategist, Goodby Silverstein & Partners

Dstillery’s location intelligence helped deliver more than one million highly qualifi ed impressions and hundreds of site visits during the weeklong campaign.

Page 15: Programmatic Quality Guide: Dstillery

us totAL MeDIA AD speND By chANNeL shAre (By Media 2014 - 2018)

source: eMArketer, juNe 2014

As usage of new media channels grow, advertising dollars follow.

2014 2015 2016 2017 2018

% of total

37.3%

26.4%

TV DIGITAL MoBILe prINt rADIo outDoor

PROGRAMMATIC QUALITY REPORT

Page 16: Programmatic Quality Guide: Dstillery

15 | AuDIeNce QuALIty • CROSS-SCREEN • INVENTORY & DELIVERY • DRIVING ROI

WE’RE BIG IN ANtArctIcA: Location Data Foibles and Fixes

Dstillery Geospatial Analyst Peter Lenz recently spoke about location-based targeting, latitudes, longitudes, and Somali pirates.

Dstillery: How has location data changed the nature of advertising?Peter: Location data provides a whole new dimension to ad delivery. We used to be in a two-dimensional world: the page where the ad is delivered and the user who sees it. Now, with location data, we’ve moved into three dimensions: the inventory, the user, and the context of where that user is located when he or she sees the ad. It’s a relatively recent shift, and it gives us a whole new way of looking at the world. It’s really exciting.

D: So when marketers talk about location data, do they mostly mean latitude/longitude?P: Yes. Cell phones work on latitude/longitude, because that’s what people are familiar with. We use “lat/long” for two things mostly: location targeting – to reach users who are currently in or have gone to certain locations – and analytics, so we can analyze how and where ads are delivered.

D: So where does the lat/long data come from?P: What we’re seeing is data sent to us by data partners and ad exchanges. In the case of the exchanges, they either get the lat/long data from the cell phones themselves, or from third-party vendors, which take an IP address and then fi gure out the lat/long of a person to a certain level of accuracy. It’s often very inaccurate.

D: Really? How can you tell?P: According to the incoming data from exchanges, there is a perpetual gathering of hundreds of thousands of users standing in the exact same spot.

D: Seems unlikely. But why does this happen?P: What’s going on is people are declining to share their data, so the cell phone is defaulting to “U.S.,” which is all you could know about that user. Or the device can’t locate itself with enough accuracy, so it defaults to whatever level of precision it can provide – city, state, or in this case, country. We get lots of “centeroids” of states. There’s one point in Brisbane, Australia, which is actually capturing many users all over Queensland. The devices don’t know exactly where, so each one makes a best guess and says, “I’m in Brisbane!”

D: What about in populated areas? Do you see inaccuracies there, too?P: Yes. Very often it’s false precision – you’ll see data for

Page 17: Programmatic Quality Guide: Dstillery

users who are often in the same building, maybe using the same IP address. So they’re roughly within a couple feet or hundreds of feet from each other, but the device will say it’s ‘here’ at a specifi c point, when it’s actually over there.

D: Any other common errors you see?P: Well, latitude and longitude divide the Earth into four quadrants, so the order the latitude and longitude are reported in, along with whether they are positive or negative, makes a diff erence. And the data often comes to us in the wrong order. We see a lot of data where it’s as if the Earth were turned upside down.

D: Upside down?P: It happens a lot with app developers. They mix up the way they report lat/long inside their code, so you end up with this upside down ‘Shadow America’ that lies in Antarctica and the Indian Ocean. Oddly, with the same type of error, there’s also a ‘Shadow Europe’ in the Indian Ocean, right off of the Horn of Africa.

What we are seeing is app developers switching latitude and longitude. There probably aren’t 10,000 people with smartphones sitting on the Antarctic ice cap or fl oating in the Indian Ocean. Maybe the Somali pirates have cell phones, but they’re probably not playing Candy Crush!

D: How is the rest of the industry approaching this?P: To be honest, location data is so new, it’s still the wild west in many ways. I’m not sure if other companies are paying attention to these issues the way we are, but we think it’s important, so we’re focused on it and investing in it.

PROGRAMMATIC QUALITY REPORT

“ There probably aren’t 10,000 people with smartphones sitting on the Antartic icecap or floating in the Indian ocean.” - Peter lenz

Arrrr... Candy CrUsh!

D: How do Marketers go about fi xing it?P: The fi rst step is just being on the lookout for these types of errors. We have a banned lat/lon list and, depending on the product, we also apply multiple fi lters to the underlying data to ensure accuracy.

Such as:

• Eliminating coordinates with suspicious volumes (in many cases, “hard-coded” IP locations in the center of zips, states, etc.)

• Removing physically impossible patterns (like a device traveling across the country in milliseconds)

• Ignoring apps with unlikely readings (like 80%+ of all users being located at only a few distinct locations)

Page 18: Programmatic Quality Guide: Dstillery

17 | AuDIeNce QuALIty • CROSS-SCREEN • INVENTORY & DELIVERY • DRIVING ROI

pLAceABILIty: The Key to Measuring Mobile Signal QualityBy Lauren Moores, VP Analytics, DstilleryThe use of location data as a signal for building and targeting audiences continues to grow. At the same time, as we have seen with other emerging data streams, with more signal comes more noise. Without taking certain precautions to ensure signal quality, the likelihood of using inaccurate or even fraudulent location data is signifi cant – and increasing.Location data, whether it’s from mobile phones, tablets or wearables, is only as good as its origin and classifi cation. Don’t ever forget the importance of placeability, which is the accuracy of the location data we’re receiving.Think about your own experience using applications that rely on location to provide content or functionality. On a recent road trip, for example, I found that as often as not at least one of my map apps could not accurately determine my current location or reverted back to a

location from hours before. Similarly, how many times have you used Uber or Hailo and had to manually adjust the pin to ensure a pickup within a few feet and not blocks away?Advertisers running geotargeted campaigns demand accuracy, and should be similarly demanding when they build audiences using location intelligence. ThinkNear estimated that 19% of location-based ad inventory available through mobile ad calls is more than six miles off target. Ouch. Given the increasing value placed on location information by publishers and marketers, there is a growing sub-industry in spoofi ng locations. We’ve even seen spoof location apps crop up. Less sinister but equally disruptive, some apps hardcode randomized locations so that they qualify for a premium data category. In some cases, more than half of mobile bid requests can be considered suspicious.With the growth of location as a means to provide services and advertising to the right person at the right time, we need to be aware of placeability. It is essential that we examine the data we use for audience and targeting so that we can avoid wasted ad impressions.

An extended version of this piece fi rst appeared in AdExchanger

19% of location-based ad inventory available... is more than six miles off target

Page 19: Programmatic Quality Guide: Dstillery

Ask your vendors which data types they’re using, and what safeguards they use confirm accuracy!

LoCATIon DATA: A PrIMEr SourceS of location data, ranked by accuracy

Courtesy of Lauren Moores, VP of Analytics, Dstillery.

PROGRAMMATIC QUALITY REPORT

1. GPS tracking opt-in through an app; it is likely the most accurate but limited if the device is not turned on or “sight lines” are obstructed, which more likely happens indoors.

2. Wi-Fi sensors Incorporates the use of floor plans and measurement of how a device moves through an indoor area.

3. Cell tower triangulation relies on the density of cell phone towers and the response of at least three tower pings to be able to pinpoint a device.

4. registration data uses the address that a user has given, most likely a zip code, which is then translated to a location centroid.

5. reverse IP geocoding takes an iP address and chooses a nearby latitude and longitude coordinate. This type of identifier is fraught with error.

6. iBeaconopt-in by the user; built into the app by developers; broadcasts where the device is approximately once per second and is not considered as accurate as other sensors.

Page 20: Programmatic Quality Guide: Dstillery

“ Bots, botnets, it’s all so abstract, isn’t it? But what we’re really talking about is advertisers paying for all of our computers to get broken into.

It feels good to strike back. As more and more of the good guys join the fight, the difference between the world we’re creating and the one we’re leaving behind becomes ever more stark.”

- Michael Tiffany, CEo, White ops

19 | AuDIeNce QuALIty • CROSS-SCREEN • INVENTORY & DELIVERY • DRIVING ROI

Page 21: Programmatic Quality Guide: Dstillery

by Alec Greenberg, VP of Media Operations, Dstillery

You can have the best performing data, models and audience targeting, but campaigns won’t truly succeed unless they are delivered to human beings on curated inventory.

According to the recent study released by White Ops and the ANA in December, bots were responsible for between 5% and 50% of monetized traffi c on even

the most premium publisher sites. In total, advertisers will lose $6.3 billion globally to ad fraud in 2015. As online ad fraud continues to proliferate and the majority of the ecosystem plays catch-up, quality inventory and ad delivery is far from a given.

You’re probably asking yourself, “Isn’t losing a certain percentage of my budget to bots the cost of doing business?” The answer is absolutely not. Every marketer and every agency should care about wasting advertising dollars, even a small percentage of them, delivering ads to machines that will never make a real purchase or be infl uenced by a brand’s message. At Dstillery, we have used our three-year head start in fi ghting fraud to prove, beyond a shadow of a doubt, that you can run an honest business and serve ads to humans rather than bots. We’ve developed an eight-step inventory hygiene process to handle every inbound bid request. These steps use two fraud-fi ghting patents and two partnerships with fraud and brand safety vendors, all to ensure that our inventory is the highest quality in the industry.

...advertisers will lose $6.3B globally to ad fraud in 2015

3. INVENTORY & DELIVERY QUALITYSigned, Sealed & Delivered

PROGRAMMATIC QUALITY REPORT

Page 22: Programmatic Quality Guide: Dstillery

21 | AuDIeNce QuALIty • CROSS-SCREEN • INVENTORY & DELIVERY • DRIVING ROI

8 STEPS TO QUALITY INVENTORY (In 40 Milliseconds)

1. BID REQUESTThe BRQ is where we see inventory specifications, including the impression URL. We look at only bid requests that use Dstillery cookies and originate from the U.S., Canada and the UK.

2. INITIAL APPROVAL: INTEGRAL AD SCIENCEIAS helps weed out 1.5 billion suspicious or inappropriate bid requests daily.

3. DSTILLERY APPROVALSWe check URLs against blacklists and for high user overlap, weeding out an additional 30 million impressions.

4. CHILD ONLINE PRIVACY PROTECTION ACT (COPPA) COMPLIANCEIn compliance with COPPA, we do not target or collect data on anyone under the age of 13.

5. URL CHECKPOINTWe assure that the bid URL matches the URL where ads are served, and that CTRs are within a normal range.

6. WIN AUCTION!YYour bid wins the auction – but we still keep working to ensure quality delivery.

7. FINAL CHECKPOINTOnce we win the auction, we have actual inventory data to confirm:A) The final landing page URL matches the initial BRQ URL.B) The final URL clears global blacklists.C) The URL still originates from the U.S., Canada or the UK.

8. SERVING THE IMPRESSION Only if inventory passes these final checkpoints will the ad be served. White Ops and IAS then verify the impression was served to a real person.If inventory doesn’t pass the final check, Dstillery serves a blank ad at no cost to the advertiser.

Page 23: Programmatic Quality Guide: Dstillery

INTEGRAL AD SCIENCE: Ranks Dstillery Leader In Fraud Prevention in Q4 ’14*

checkLIst Questions Marketers Should Ask Their Media Partners: What monitoring tools and practices do you have in place?

What quality-related criteria have your publisher partners adopted?

Do you work with third-party providers to ensure brand safety and compliance?

How do you determine which impressions are exposed to real humans?

How do you ensure that ads are served as reported, and that URLs are visible to the advertiser?

whIte ops: Ranks Dstillery 97%+ Human Delivery in Q4 ’14+

Integral Ad Science, one of the industry leaders in inventory quality and fraud prevention, continues to show Dstillery having the lowest percentage of ad fraud – over 4X lower than industry averages.IAS also scored Dstillery as the lowest in terms of brand safety risk.

White Ops is a pioneer in the detection of and systematic defense against bot and malware fraud. White Ops confi rms Dstillery’s patented technology works, scoring us among the lowest fraud rates in the industry across all inventory and campaigns:

over 97% of our ads are served to real consumers, consistent with our Q3’14 performance.

% of ad fraud

14.5%

3.4%

DstILLery INDustryBeNchMArk

brand Saftey: Moderate to Very HiGH riSk

17.5%

1.7%

DstILLery INDustryBeNchMArk

*SOURCE: INTEGRAL AD SCIENCE SEPT 2014, +SOURCE: ANA-WHITE OPS DEC 2014 REPORT

PROGRAMMATIC QUALITY REPORT

Page 24: Programmatic Quality Guide: Dstillery

23 | AuDIeNce QuALIty • CROSS-SCREEN • INVENTORY & DELIVERY • DRIVING ROI

“ Acquisition is always our goal. We’re helping the brand reach people who are considered high lifetime value customers.

- Maggie Summers, Media Supervisor, iCrossing

Page 25: Programmatic Quality Guide: Dstillery

by Gilad Barash, Data Scientist, DstilleryFrom awareness to DR campaigns, all marketing tries to spur some change in consumer behavior, and ultimately aims to drive a signifi cant and positive return on investment (ROI).

Right now, quality doesn’t play a huge role in evaluation of ROI – but it should. Much of measurement relies on a “dollars in, dollars out” approach. Without a focus on quality, however, we’re missing out on a major measurement aspect of that metric – how real those results are. For example, when running an awareness campaign in which reach and target market are the main success benchmarks, we need to ask whether we are truly hitting the number of in-market consumers as claimed? If the data used to calculate reach isn’t verifi ed, marketers run a real risk of not reaching the consumers they intended to, or not reaching individual unique consumers (or even humans) at all. In short, the numbers on your weekly status reports are only as good as the real results they add to your bottom line. Here at Dstillery, we’re committed to providing real results to our clients, and that extends beyond sourcing

audiences or delivering your ads to real humans on brand safe inventory. Our entire platform rests on the quality of data: in order to build accurate models for prospecting and to measure ROI, we need to ensure that the data we work with is verifi ed, and of the highest possible quality. Historically, our modeled audiences have performed well for all clients across campaigns and verticals – a feat that is achieved only when the initial data we use to build our model is clear and correct.What does that mean for you? Less wasted marketing dollars, and greater certainty that the results you’re receiving are real. White Ops/ANA projected over $6 billion in wasted marketing spend this year, mainly because many marketing channels do nothing to ensure that the audience delivered matches the intended target audience.“Quality” isn’t just an abstract term that data scientists use to describe the accuracy of their input data. Quality is the backbone of our entire marketing ecosystem. It’s a fundamental concept we should all place greater focus on to ensure that marketing campaign results are real and trustworthy, and truly contribute to your business’s bottom line.

4. DRIVING ROIYour Bottom Line

PROGRAMMATIC QUALITY REPORT

Page 26: Programmatic Quality Guide: Dstillery

1. 2. AUDIENCE QUALITYQuality in, quality out. Make sure the data sources (cookie and location) are clean, accurate, recent and verified before building models.

CROSSSCREENKnow how your partners build cross-screen profiles. Do they go beyond mere IP matching, or is it guesswork?

PR

OG

RA

MM

ATI

C Q

UA

LITY

GU

IDE show your work:

Key Takeaways on Quality For Marketers

Page 27: Programmatic Quality Guide: Dstillery

3.INVENTORY& DELIVERYMake sure your partners have the appropriate checks in place get your ads seen by humans in in brand-safe environments. The best-laid campaign plans will fall flat if your ads aren’t delivered as intended – or if they reach bots instead of humans.

4.DRIVING ROIHow real are your results? Find out how your partners verify reporting so you can trust your campaign will move your bottom line.

Page 28: Programmatic Quality Guide: Dstillery

Dstillery Quality GUIDE Contributors:

Gilad barash, Data Scientist

Brian D’Alessandro, VP of Data Science, @delbrains

louise doorn, CMO, @doorn

alec Greenberg, VP of Media Operations, @AlecGreenberg

Peter lenz, Geospatial Analyst

Sean Lough, Director of Marketing Communications & PR, @Sean_Lough

catherine Mietek, Director of Product Marketing, @catherinemietek

lauren Moores, VP of Analytics, @lolomoo

claudia Perlich, Chief Scientist, @claudia_perlich

tom Phillips, CEO, @tomdstillery

ori Stitelman, Senior Data Scientist

daniel yi, Data Analyst

FOLLOW US @DSTILLERY CONNECT WITH US

Dstillery is a pioneer in marketing technology. We capture the promise of digital media by applying data science and massive intersecting data sets to predict brand affinity and build audiences.

FOR MORE INFORMATON: [email protected]

470 Park Ave South, 6th Floor, New York, NY 10016 | P: 646-278-4929 | dstillery.com