Automated Bidding Strategies

   

Automated Bidding Strategies

Machine learning, smart bids, and the risk of being sacked for a machine...

Having attended the Automation Expert event at Google HQ a couple of months ago, to discuss artificial intelligence, machine learning, and bid strategies, I thought it would be fitting to write a piece for those who don't know a lot about bid strategies or how Google's machine learning works. We'll be answering the questions:

  • Why machine learning?
  • Why smart bid?
  • Am I going to lose my job (to a machine)?

Why machine learning?

According to Google, we are now moving in to the 4th evolution of computing. First, there were computers, then came the web, then mobile and now, machine learning. 

At this stage it is important not to confuse Artificial Intelligence (A.I) with Machine Learning.

  • A.I is the science of making things smart.
  • Machine learning is a technique used to develop A.I.

Smart bidding and Machine Learning make use of the ever-growing number of signals and identifiers available from users. We are searching more and using more and more devices, I think the following stat puts this in to context:

There are currently 10 billion devices in the world which connect to the internet, Google predict this will be 34 billion in 10 years' time (Automation Expert event).

An example of a few user signals:

There are, of course, many more signals to make use of through Advertising platforms. We have seen signals grow massively with the emergence of mobile. 

Why smart bid?

Smart bidding has many advantages, but the key ones are:

  • Smart bidding can rapidly evaluate sets of data signals.
  • Smart bidding can help you work smarter by automating routine tasks & freeing up time for strategic thinking. 

Bidding levels

The levels of bidding through the AdWords platform vary:

  1. Manual bidding
  2. Rules based bidding 
  3. Intraday bidding
  4. Smart bidding

1. Manual bidding

This involves manual keyword level changes, which can be based on certain criteria of performance. Time constraints may lead to a subset of keywords being optimised such as top performers or poor performers. 

2. Rules based bidding

Conditional based. The system automatically adjusts bids when keyword performance changes. This type of bidding takes the rule into account only and does not look at other signals.

3. Intraday bidding

Machine learning algorithms learn from historic and on-going keyword performance to optimise keyword bids several times a day. Often referred to as 'real-time' bidding.

4. Smart bidding

True auction time bidding, not just a few times a day - bidding is flexed during every single auction. Smart bidding identifies the right conversion opportunity for every single auction. Machine learning algorithms rely on robust conversion data to build accurate bidding models that perform at different levels. Smart bidding is a proactive optimisation approach. 

With smart bidding there is no need for bid adjustments on device, user lists, locations and more. This is because the algorithm is taking all these signals into consideration before determining the appropriate bid.

Bidding algorithms typically take 1-2 weeks to calibrate to a newly created strategy, but this is dependent on data. Google still recommend 30 conversions in the past 30-days as a threshold, however this is no longer a requirement. 

Am I going to lose my job!?

All of these strategies make use of the user signals AdWords and analyse this data within a split second of an auction happening. It's far quicker and more agile than any PPC guru can ever be. But are PPC managers going to be obsolete as a result? Are we all going to be replaced by machines?

The short answer is no. Of course, our roles will change, and we will have to adapt in line with more and more smart features being released, but that's the industry we work in. When has paid search ever stood still? We can blend our skills with the machines and work in perfect harmony, and there are still things we can do that machines cannot.

PPC managers still need to apply context to machine learning results and performance, we have a duty to harness and control which automated systems work best for us and are more suitable for our clients. 

With human context and creativity and machine learnings' speed and AI, we can combine these traits to produce quicker results and test more and more for our clients, which in turn will see us drive better results.

Through the use of bid strategies and machine learning, we can test more and unearth new areas which work well for and drive incremental volume. Machine learning algorithms can inspire us to test and branch out to more tactical keywords. If a strategy is working well, why not test this on keywords that have historically been poor performers? 

Be sure to test...

Careful consideration is encouraged when moving in to the world of smart bidding, the strategy needs to be right for your business and objective. Once this has been determined it is absolutely crucial you give the strategy time to learn and optimise. It's also important not to set unrealistic targets or have too many constraints on the strategy. For example, if you employ a target CPA strategy, do not set the target CPA significantly lower than what the account has achieved previously and also ensure that the campaign has ample budget.

A bid strategy needs to be tested for at least 4 weeks, to allow the algorithm to calibrate, learn and optimise. During this time, changes should be minimised to reduce the number of variables and ensure it is as fair a test as possible. Also, be sure to choose the right metrics to measure the success of the bid strategy, and this should be linked to the objective of the test.

To summarise what is required for testing:

  • Minimum of 4 weeks
  • Minimise changes
  • Realistic targets and objectives

If all is working as intended, a phased roll out across key campaigns should be carried out but be sure to keep a close eye on budgets and daily spend. AdWords has a 200% overspend threshold built in to campaign daily budgets in order to capture relevant traffic, so care should be taken.

To conclude, smart bidding and machine learning is nothing to fear and you are already using it more than you know, it's just running in the background. The key is to test and run smart bidding on a subset of your account, then use your findings to aid future implementation. 

If you're interested in how we can help you optimise your AdWords account, contact us to secure your free PPC audit.

Book Audit

 

   

Contact Us

Do you have a challenge for us to solve?

get in touch