According to Webster’s Dictionary, the word “programmatic” was first used in the late 19th century. Despite its long tenure in our lexicon, the word was an obscure one until recently. If you aren’t familiar with it yet, if it hasn’t permeated your corner of the business universe, just wait. Programmatic thinking might soon join the pantheon of 21st century buzz words, alongside big data and cloud.
The current industry being transformed by programmatic thinking is the advertising industry. A few years ago, software entrepreneurs began to realize that as advertising started to go digital, there was an opportunity to apply algorithms to media buying decisions. Instead of having a 27 year old neophyte designing your media plan over a three martini lunch, have the world’s most powerful machines do it for you “auto-magically”, leveraging all your best data – and streams of other’s best data – to inform the decisions. And the best part? The machines learn how to make better and better decisions with every purchase.
The speed with which programmatic advertising has taken over the industry has been breath-taking. From nowhere a few years ago, $12 billion of advertising was purchased programmatically in 2013 and the forecast for 2017 is $33 billion (Magna Global report). 86% of advertising executives and 76% of brand marketers are using programmatic techniques to buy ads and 90% of them indicate they intend to increase their usage by half in the next 6 months (AOL survey). Companies like AppNexus, DataXu (a Flybridge portfolio company), MediaMath, RocketFuel and Turn are among the leaders in the field.
The next industry to be transformed by programmatic thinking is financial services. Decisions to underwrite loans have historically been based on a few simple data points such as the lender’s zip code, credit score and job history. With the application of big data techniques and sophisticated machine learning algorithms, underwriting decisions are becoming programmatic. For example, Flybridge portfolio company ZestFinance evaluates thousands of data points in credit applications (even trivial ones, such as whether the applicant uses capitalization properly) to make loan underwriting decisions programmatically. Like other programmatic-based businesses, ZestFinance sees a powerful network effect: the more data they inhale and the more decisions they make, the smarter their decisioning algorithms become.
What other industries might see programmatic thinking ripple through? Once I put the programmatic lenses on, I can see dozens of industries being affected. Just think about all the decisions consumers and businesses make, and whether programmatic thinking could automate and enhance those decisions. For example:
- Navigation decisions: my navigation behavior follows clear patterns, as does that of millions of others. Navigation software in cars and phones will soon become more programmatic in anticipating where I might be going and the best routes to get there based on real-time data and experience.
- Hiring decisions: evaluate thousands of data points to evaluate the best candidates and then watch their performance and make better decisions next time.
- Security decisions: evaluate thousands of possible threats and patterns, watch the outcomes, and design algorithms that learn from these experiences to reduce acts of fraud and terrorism.
- Investment decisions: One of our portfolio companies, MatterMark, evaluates thousands of data points to determine private company performance, and then seeks to tune those algorithms for more and more accurate predictive investment decisions. Today, their service is being used by hundreds of investment firms.
Some might object that all this automation and machine learning designed to replace human judgment is going to be bad for society – making humans less relevant and eliminating jobs. But in fact, many researchers believe the advent of machine learning will generate new kinds of jobs – where a hybrid of automation and common sense is applied. MIT's David Autor presented a paper a few weeks ago that argued:
Many of the middle-skill jobs that persist in the future will combine routine technical tasks with the set of non-routine tasks in which workers hold comparative advantage — interpersonal interaction, flexibility, adaptability and problem-solving.”
So don't be afraid to put those programmtic glasses on. I think they're pretty rose-colored.
Sorry–also pricing decisions (especially in travel, ex: Duetto) could be hugely impacted. Wouldn’t be surprised if Hotel Tonight began using programmatic techniques as well.
And in fin-tech, leveraging data might be tough. Oliver Wyman did a cool study on how social media data often biases people racially, which is border-line illegal depending on which court precedences a judge wants to consider relevant.
I think “augmenting” decision making and “making decisions” are two different things. Even in the ad industry, programmatic buying supplements some of the biggest purchases (like many custom homepage ads, etc.).
Mattermark, I’m sure, supplements investing decision rather than makes them.
I think AI has an awesome ability to get rid of rote functions accomplished by junior employees, but high-level decision making probably won’t be overcome by tech in a while–in large part because most NLP algorithms haven’t improved in a while, and there would need to be a fundamental shift in available technology to do so.
I might look at this comment in 5 years, though, and think I was an idiot… who knows.
There must be a risk of refining these things so much that it does more harm than good.
With the hiring decisions – what if the first few successful hires are white males (for example) – logically the machine would then discriminate women and minorities. Diversity is usually good for a team, but the program would continue to optimize for the perfect white male each time – (if true) surely this would be bad.
I guess the programs will solve this as they get more advanced but we are not there yet.
No doubt, programmatic tasks will become automated with the automation improving with each iteration. Certainly this case is true for tasks with a linear decision tree such as credit scoring – there is one “client” and a sole AI reviewing that single application. Perhaps in time, less true for tasks like programmatic advertising where the decision tree is far more complex. Here multiple AIs impact the market simultaneously and therefore the subsequent interventions by each original programmatic algorithm. Each iteration generating a response in the competing AI. Do you remember WarGames when the AI, Joshua, played tick-tac-toe? The only way to win is not to play? Of course, not playing in these markets is not an option, but, like High Frequency Trading, the secret sauce will ultimately be the algorithm created by the algorithm rather than the original source code. And then who actually owns code created by code?
Other authors see it differently, starting with http://raceagainstthemachine.c… or http://www.pewinternet.org/201…. You have to remember the example by Ray Kurzweil with the chessboard: with #AI, or with computers in general, we are now entering the second part of the chessboard, where our linear experience will now greatly differ from the exponentially working effects. It seems possible that using ever more AI will raise inequality at the least by condemning us all to ever more stupid jobs, aided by AI – and those stupid jobs will get paid less.
On the other hand, I see this will not be stopped just because some of us point out it might be dangerous. But maybe we all could benefit by a discussion of the risks involved – and then trying to minimize or mitigate those risks?
Nice round up. This is going to be the century of AI and it’s not going to take long before we start rolling our own and turning to dedicated platforms providing intelligence as a service, that is companies that can build sophisticated neural networks from large datasets. That company may be the next Google.