TripleLift: Preparing for a New Programmatic Ad-Tech World

23
Preparing for a New Programmatic Ad-Tech World

Transcript of TripleLift: Preparing for a New Programmatic Ad-Tech World

Preparing for a New

Programmatic Ad-Tech World

Agenda

• Speaker Introductions

• How has advertising changed?

• Who is TripleLift?

• Native Programmatic Advertising

• TripleLift Native Ad Exchange (TLX)

• The Importance of Data

• The Challenge

• Current Solution -> New Solution

• Why VoltDB

• Wrap Up

Speakers

• Michael Harroun, Head of Backend Engineering, TripleLift

• Jared Skolnick, Head of Publisher Marketing, TripleLift

• Joe Kelly, AdTech Solutions, VoltDB

• Dennis Duckworth, Senior Director of Product Marketing, VoltDB

4

Data is transforming business

How has advertising changed?

1996: The Google search engine opens the world to new ad forms and platforms.

2004: Facebook launches

2007: The iPhone brings the Internet to consumers’ pockets. The era of mobile advertising begins.

2009: NYT print advertising drops 27% (Pew Research); Google launches DoubleClick Ad Exchange

2010: Instagram launches

2017: Programmatic advertising captures 58% of ad spend

Three + screens. Real-time bidding. Consumers in control. Advertisers have

<200 MS to make an impression.

Native Programmatic Advertising

• Native Ads

• Match the style of the content

• Fit in with the layout

• Are relevant to the content

• Programmatic

• Automated and simple to

implement

• Digital ad exchange that facilitates

real-time auctions

TripleLift is an ad technology company and market leader in programmatic native

advertisements. By leveraging our real time ad exchange & computer vision we make

native advertising simple, scalable, and effective.

Who is TripleLift?

2013 2014 2015 2016

Native

212 & NY Times Ad Tech Startup of the Year

Native Programmatic

Market Leader in Native Programmatic First Native PMP Platform

Forbes: America’s Most Promising Companies

Crain’s Best Places to Work in NYC

Crain’s Best Places to Work in NYC

Ad Age Best Places to Work

Over 12,000 Advertisers More than 500

Publishers

Up to 1 Billion Ad Reqs/Day

US Native Digital Display Advertising

• The TLX was the first RTB native ad exchange

• FeaturesFully functional Native Ad Exchange

Supports both direct integrations and via 3rd Party systems

Internal Bidder for directly managed ad campaigns

Provides asset (image/video) auditing and validation

Tracks user interactions and conversions

Leverages current & historical data to optimize cost / value for our internal inventory

TripleLift Native Ad Exchange (TLX)

TLX Overview

Ad Placement

On webpage

TLX Overview

TripleLift

Exchange

Ad Placement

On webpage

TL & External

Bidders

Ad Request

Bids

TLX Overview

TripleLift

Exchange

TripleLift

Engagment Bus

Ad Placement

On webpage

TL & External

Bidders

Ad Request

BidsRender,

Click,

Conversion,

etc.

TLX Overview

TripleLift

Exchange

TripleLift

Engagment Bus

Ad Placement

On webpage

TL & External

Bidders

Ad Request

BidsRender,

Click,

Conversion,

etc.

EventBus

TripleLift

Data PipelineActionable

Data

Pre Render

Events

Post Render

Events

• Use Cases• Alerting & monitoring for:

• 3rd party / pipe integration issues

• Inappropriate or invalid ad assets

• Drops in traffic from external sources

• Significant changes in bid quantity or value

• Drop in supply based upon any metric (publisher, browser, region, etc)

• Drop in demand based upon any metric (advertiser, partner, ad type, etc)

• First party spend pacing & bidding insights

• Data science & machine learning

• Why speed matters...• The speed at which we can respond to changes or detect anomalies directly impacts revenue

The Importance of Data

• Large sets of data (13 billion events / day)

• Over 30 different events to track

• Each event is only one piece of the puzzle

• There is no order to the events

• Rebuilding 1 billion auctions from 13 billion events is complicated and

expensive

The Challenge

Publisher User Ad Slot Ad Asset Advertiser Bid Clicks Render Price

Auction

• Large sets of data (13 billion events / day)

• Over 30 different events to track

• Each event is only one piece of the puzzle

• There is no order to the events

• Rebuilding 1 billion auctions from 13 billion events is complicated and

expensive

The Challenge

Ad Request Event

Publisher User Ad Slot

Bid Event

Advertiser Bidder Ad Assets

Win Event

Bid IsWinner

Render Event

DidRender

Click Event

DidClick

Publisher User Ad Slot Ad Asset Advertiser Bid Clicks Render Price

Auction

The Challenge

Ad Request Event

Publisher User Ad Slot

Bid Event

Advertiser Bidder Ad Assets

Win Event

Bid IsWinner

Render Event

DidRender

Click Event

DidClick

Publisher User Ad Slot Ad Asset Advertiser Bid Clicks Render Price

Auction

• Large sets of data (13 billion events /

day)

• Over 30 different events to track

• Each event is only one piece of the

puzzle

• There is no order to the events

• Rebuilding 1 billion auctions from 13

billion events is complicated and

expensive

Current Solution: Spark + Storm

EventBus

Secor

Spark

Storm

Bid

Pacing

Real Time

Stats

BI

Tools

Druid

Reporting

• Real Time: Apache Storm • Simple real time data processing

• Powers bidder pacing

• Enables very basic real time stats

• Batch: Apache Spark• Processing accurate and highly granular data

• Supported by Secor for data persistence

• Processes the data for all other use cases

• Druid• Scalable to petabytes

• Supports real time data insertion & querying

• Main store for all reporting & business intelligence

Batch

Pipeline

Real Time

Pipeline

Current Issues & Drawbacks

• Apache Storm:• Data accuracy is not guaranteed

• Event chaining / merging is very complicated

• Apache Spark:• Computationally expensive

• Real-time processing limited

• Needs heavy support from other systems

• 3 hour delay to process events

• Druid• Requires data to be heavily pre-processed

EventBus

Secor

Spark

Storm

Bid

Pacing

Real Time

Stats

BI

Tools

Druid

Reporting

Batch

Pipeline

Real Time

Pipeline

New Solution: VoltDB For Real-Time

• VoltDB to ingest events in real time

• VoltDB decides on a per event basis

• If we have the other auction events we need

• Holds on to incomplete events

• Pushes out composite events to druid

• Now that Druid is real-time

• We can add real-time features to all systems

• Have all systems share a single data source

• Implement new / better systems that demand real-time data

• Limit Spark to backfills and auxiliary use cases

EventBus

Secor

Spark

VoltDB

Bid

Pacing

Real Time

Stats

BI

Tools

Druid

Reporting

Batch

Pipeline

Real Time

Pipeline

• Fast and highly scalable

• Relatively simple to understand and use

• Semi drop-in solution for our use cases

• Needs minimal support from external systems

• Has a lot of potential for many use cases within our stack

• Hardware / processing cost lower compared to our Spark solution

Why VoltDB?

The View from VoltDB

• Adtech, like other industries, benefits from accurate data and real real-time,

not near real-time

• Some new tech like NoSQL is great for quick and “dirty” data management

but more companies are rediscovering the value of true relational databases

http://www.infoworld.com/article/2617405/nosql/7-hard-truths-about-the-nosql-revolution.html

• Leaders in each industry, like TripleLift, are creating competitive advantage by

using their data to produce intelligent insights faster and more accurately

• Some cross-industry use cases are proving perfect for in-memory databases

like VoltDB, like Fraud Detection and Management (FinServ and Adtech,

among others)

Wrap Up