Many types of algorithmic or automated trading activities have grown and evolved in response to the many changes that have taken place in the market landscape since the Internet became established in the late 1990s. In the last 10 years, algorithmic trading has been an established feature of the global financial markets. More recently, the advertising industry too is at an inflection point in their evolution – media buying is becoming increasingly automated. Programmatic ad buying can now be done at a large-scale and at very high speeds using computers to participate in real-time automated auctions for ad space across a large number of publisher sites.
There has been much discussion and debate around programmatic ad buying, so I will not repeat them here. Instead, I hope to offer a different perspective on programmatic ad buying – my perspective drawing influence and reference from the already mature algorithmic or automated trading systems that is in the financial industry. I offer my perspective in three areas – (1) what happens to the trading marketplace, (2) our increasing dependency and reliance on black box science, and (3) playing in a game of speed. Then I finish this post with some suggested guidelines we must consider as we build our programmatic trading systems.
I hope you enjoy!
The trading marketplace itself: Setting the scene.
I’ll start by giving you a simplified look into the financial trading process. Everyone can act as both a buyer and a seller for a commodity, say a stock. All prices of this stock is transparent; buyers and sellers both see what each other have paid or sold. Stock exchanges are the infrastructure that facilitates the trading of these stocks. Recent reports show that more than 60% of all global financial trading is now performed algorithmically or automated. All trading volume is then fed into the stock market which represents a pool of companies that list these shares for investors to buy and sell in the first place.
The reason why I think algorithmic trading has performed well for the financial industry is that, ultimately, the marketplace itself is the regulator. Based on the example above, the stock market determines the price of the stock through the amount of buyers and sellers at a given time.
I believe programmatic ad buying has the origins of algorithmic trading in financial markets where the principles are very similar. However, the players in this trading process are very different. What we have is a linear model of three distinctive player descriptions: Sellers – Publishers, who only sell, Agents – (Media)Agencies, who both buy and re-sell, and Buyers – Advertisers, who only buy. So while the advertising industry may have created ad exchanges, agency trading desks and demand-side platforms as infrastructures for bidding and trading to take place, it is seemingly clear that without a marketplace as a bedrock, a “stock market” per se, exchanges would have no reason to exist. The reasons for this are: Sellers do not have a ‘formal’ mechanism to list their ad inventory and buyers are unfairly determining the balance of demand and supply in the market at any point in time.
With many adland folk positioning programmatic ad buying to offer higher control for the buyer – that they can ‘sit in the driver’s seat’, make the buying decisions and pull the optimising levers, there is no doubt it offers an improved and democratised process versus more traditional buying techniques. Nonetheless, I have to wonder. You see, the dynamics of players in this process is rather unique when compared to algorithmic trading in the financial markets.
On one extreme end, we have the Sellers (publishers). The seller-side naturally wields tremendous power in this trading process because both Agents as well as Buyers want their ad space. With this in mind, Sellers are determined to set their own conditions – for example, pricing or ad viewability which leads in to the issue with data, or rather ‘information’. As we know, programmatic ad buying is highly reliant on information. Currently as it stands, the ownership of data is in the hands of a select few, big institutions, the likes of Google, Facebook and Apple to name a few. For all intents and purposes, any truly valuable data is safe-guarded and used as a competitive-advantage for the Sellers own benefit. New algorithms are regularly being developed by Sellers which take advantage of their proprietary knowledge. As such, there is very little information available about the types of programmatic advertising that publishers are currently deploying. All other data is therefore sold – such as location, platform, device, browser, and where available, other forms of behavioural data relating to specific but anonymous users. Thus, we have indirectly created a situation where Sellers have complete control over how their ad space is sold, to whom and with what data, and at what price.
On the other extreme end, we have the Buyers (advertisers). Direct Buyers may require to trade with limited information and have to accept the Sellers conditions at face value and Indirect Buyers who go through intermediaries, yes, their (media)agencies who act as Agents.
In the middle are Agents, (Media)agencies. This is where the process gets rather interesting. Unlike algorithmic trading in the financial markets, programmatic ad buy via agency trading desks are able to both buy and re-sell ad space. Predictive algorithms are used to target bids or execute an order in which information about a user is matched to a bidding rule in a minute fraction of a second – enabling for a bid to be made or a relevant ad to be served by the time the page being visited by the prospect fully loads. However as we know, algorithms can easily be adjusted to fit any criteria – and may or may not be in the interest of either the Seller or the Buyer. Agents operating on both sides are able to find ways to use information, not available to the general public, to generate tremendous profits for themselves and their organisations. Large orders can too be easily manipulated to exploit pricing and viewability discrepancies, fuelling a cowboy-like trading landscape where all parties are always in constant competition with one another.
In such circumstance, each party will find inventive ways to maximise profits at the unfortunate expense of transparency. If the advertising industry trades this way, the demand and supply chains of publishers, (media)agencies and advertisers will be badly disrupted.
Black box science: Indicative, not directive.
Algorithmic trading is very entrenched in the financial investment community because it has its advantages in certain (predictable) situations – faster execution, no emotions, macro trend identification, a high amount of data processing. However like any form of black box science, if an algorithm is not up to date or has been applied in the wrong way, in this situation, it’s possible to lose a lot of money very quickly. For example, algorithmic trading tends to correlate with zero understanding of causation, suggesting that strategies used in the market may not be as diverse as those used by human traders. After all, it is important to remember that algorithms are man-made instruments that use a defined set of rules which are to be followed in mathematical calculations by a computer.
Rules tend to be linked to a specific benchmark, price or time. So, algorithmic optimisation can only do so much – in fact, it can only do what it’s been programmed to do. As such, there is a risk that many of the algorithms would generate similar investment decisions as they all incorporate the same data set.
In the advertising industry, however, qualitative filters are still significantly important for the campaign execution. In the early 2000s when I was still optimising TV plans on an excel sheet using macros I had written (you heard me, there was no “optimising tool” during my time), there was a need to adjust the final plans to also take into account highly relevant ad environments – affinity buys, we used to call it. The reason for this was to ensure that the ad running was considered ‘brand safe‘ by the advertiser, not just driven by ratings or low CPMs. Therefore, content and context as we know it today, can be undervalued in a programmatic landscape which places a premium on audiences rather than its specific ad environment. Where Agents or Buyers are bidding based on reaching a specific cookie indicating a select audience segment, often using price as the determining factor, it can introduce an opportunity for mistakes in ad placements.
Therefore, possibly the best solution is to use a hybrid (human/computer) approach to make use of each entity’s best traits. The human factor has some additional features to offer that a computer can’t match – sensibility, intuition, experience, courage, creativity and most of all the ability to assess complex situations by filtering out the noise and to focus on the bare essentials. We should consider human judgement superior because it incorporates intangible attributes that no computer program can duplicate.
Trading speeds: A level playing field turns into a myth.
One of the said benefits of programmatic was an attempt to create a level playing field across the buying and selling landscape in which price will be more fluid and objectively determined. On the surface it does look like a level playing field but from what we can learn from the financial industry, algorithmic and automated trading can also lead into high frequency trading. It is essentially a program trading platform that uses powerful computers to transact a large number of orders at very fast speeds. To understand the magnitude, in today’s electronic financial markets, a single investor can execute more than 10,000 trades a second – that is more than 1,000 trades can happen in the blink of an eye. But when fast connections can connect to gather detailed information about the movement of the Sellers in microseconds, the difference is making a great trading decision over a mere good one.
This too can happen in programmatic ad buying, specifically favouring the Agents, and especially the large conglomerates. Why do I say so? A competitive advantage can be had by those who can afford the IT infrastructure, data talent, and high CAPEX to invest in sophisticated machines to run high frequency strategies which will win at the expense of all other ‘ordinary’ investors. Yes, you heard me.
It will be the battle of who will have the most resources to throw into their automation process who will win. Indirectly, this leads to large Sellers favouring large Agents with the deepest pockets, giving them an advantage over the rest in order to make strategic investment decisions which will ensure mutual commercial benefit. So what chances is programmatic ad buying actually giving to smaller investors and smaller buyers to participate in the open market? Not much, I think.
So, what now?
I hope so far you haven’t gotten me wrong, I still think the advantages of algorithmic trading for the financial industry or programmatic ad buying for the advertising industry far outweigh the disadvantages. However it is also important to recognise the current shortfalls to be able to take the right approach and put the necessary investments behind building a solid programmatic trading system. Therefore, if you or your company is still adamant to jump on the programmatic bandwagon, please refer to my checklist below as a suggested guide. As a member of the advertising industry myself, it is important that if we were to seriously consider this to be game-changing, then let us take the right approach. My last piece of advise to you, I continue to stress is, please don’t embark on this transformational journey if we’re not fully invested. Otherwise, we will end up with a haphazard process which will have detrimental impact overall.
#1. Be prepared to invest in a strong infrastructure and have the necessary controls in place.
For the most basic algorithmic trading system: database, decision engines and computing infrastructure is required up-front. Then, hire exceptional talent in IT specialities. The technical requirements of any algorithmic trading are high.
Appropriate controls must be in place even before any trading begins. This is a necessity when conducting any electronic-trading business. A ‘defense-in-depth’ strategy should be considered – a concept regularly used in the information security field, in which multiple layers of security controls are placed throughout the system at multiple points in the process. This means, companies using programmatic ad buying need to have controls that cover all aspects of the trading process – from order generation, order handling to order execution.
With the myriad number of companies entering the programmatic ad buying landscape, it may not be all that practical to ensure that every company has all the risk controls in place.
#2. Test the algorithms to death.
It is necessary to conduct as many simulations and non-live testing within a trading environment. Every test helps to ensure that algorithms pass the risk management controls. Once ready to be live, the algorithms must be rolled-out in a controlled, cautious fashion. It must be the responsibility of the company to self-impose risks and limits, compensation controls and limit the number of instruments where this algorithm will be deployed. Then keep testing, ideally keeping it up to date annually. It is important to ensure that this system, these controls can withstand any significant market volumes and any external events that could potentially exert stress. When we face a situation where its algorithms acting on other algorithms, all it takes is an error to create a domino effect.
#3. Get as many internal and external stakeholders involved.
To guarantee any success, all stakeholders must have a voice at the table from the onset in determining the right balance between risk and controls. This effort should ensure transparency and credibility. Control functions and developers both need to understand the inherent risks to ensure that the proper controls are in place. The minute we have detached ourselves from the physical currency and move to a virtual state of money exchange, any human touch is lost. This is possibly the most dangerous aspect because it triggers for riskier investment decisions. Senior management and boards must too be aware of risks being taken at the desk level of trading. Therefore, everyone should know what they are getting themselves into.
Through the lens 