With the ever-increasing role of machine learning and constantly evolving landscape of user-behaviour, we went along to Google’s Smart Bidding event to find out what this means for the future of the day to day running of PPC campaigns.
It is now a recurring theme in PPC, every few years there is a hard push from Google for marketers to switch from manually managing the most time-consuming aspects of running an account to just letting Google deal with it for them. Surely, marketers want to spend less time in a spreadsheet, changing bids or re-writing ad copy and more time analysing and strategising – a point that is quickly brought up at this event. Everybody agrees – so why aren’t they doing it?
Despite the slick presentation and well thought out approach to Smart Bidding presented on the day, the attendees still took the Google representative’s products with a pinch of salt.
Everybody in the room had tried relinquishing control of a Campaign or Ad Group to a bid strategy. Occasionally, they had seen very good results, strong increases in the number of conversions, a CPA slashed in half. Google’s deck was chock full of incredible statistics.
Most of the time though, marketers will tell you that they saw their CPC inflate for very similar results against their KPIs and had to step in and take action to regain control.
The idea has a great deal of potential – everybody knows that the future of PPC is most likely automated, but that’s just it – it still feels like the future. It feels like the automation we need is not quite here yet.
I believe every person in the room would hand over control of their bids to Smart Bidding, if they fully trusted that they could expect consistently positive results. Who doesn’t want to free up the time it takes to optimise and spend it instead on strategy? Their problem with it was that it still remains untested – the benefit is not fully proven. Most of all, when it doesn’t work, they can’t explain why – which is a position nobody wants to be in when discussing results with a client.
The outcome of the day was never going to be everyone switching to Full Automation and rolling Smart Bidding out across all their campaigns immediately. What it did do, though, was get the ball rolling and give the attendees many ideas and avenues to test the water.
Why Smart Bidding
With the huge rise in mobile searches, Google has identified new patterns of behaviour. Users are more curious, want assistance in every moment and are ultimately more impatient.
They have noticed a stark increase in searches like “Where to buy_,” and “_ Near Me”. Users are giving Google a constant stream of signals, too many for any human to process. There is now literally no chance of a person or team of people being able to process and analyse these signals, and then use that information to set the most appropriate bid in each auction.
This is why Machine Learning is necessary and is growing so fast. Machine learning finds patterns in large sets of data and uses them to make predictions. It can make much faster and much more accurate decisions in (almost) real time. Smart Bidding, therefore, uses Machine Learning to spot these patterns in the signals given by users and adjusts your PPC bids accordingly, in theory.
Smart Bidding Strategies
Smart Bidding campaigns essentially break down into three different bid strategies, which will be familiar to most PPC practitioners:
Maximise Conversions – for when you have a limited budget and want to get as many conversions as possible within that amount.
Target CPA – for when you want more conversions but can’t spend more than a certain amount per conversion.
Target ROAS (Return on Advertising Spend) – for when you want to drive revenue and your conversions vary in value. (The mention of the name Target ROAS in the room drew an audible groan – it seemed many people had been stung by this one in the past.)
The biggest revelation was that smart bidding will now be query focused rather than keyword focused (another nail in the coffin for keywords?). The idea is that the Search Query carries more signals about intent, and can therefore give the bid strategies a better basis on which to make decisions. Search query data will be drawn from account level, meaning bid strategies can make decisions for small campaigns based on the queries in a larger one, or vice versa. In order to give bid strategies even more data to work with, campaigns that have the same targets can be bundled together in a Portfolio.
On the topic of ‘more data’, you should be using a Non Last-Click attribution model that includes as many touch points as possible – the data driven model being the optimum.
There is no need for bid adjusters (device, audience list etc) as Smart Bidding will take these signals into account.
Some of the main pitfalls to avoid when rolling out Smart Bidding include:
- Being too aggressive with your constraints which will limit performance
- Being too idealistic or unrealistic with your goals, e.g. setting a CVR of 80% for campaigns with a 2% CVR, then calling this a failure when it isn’t achieved.
- Analysing at the wrong time – this was a big one, Google made it very clear that you should not analyse campaigns when the bid strategy is in its ‘Learning’ phase, and you should leave enough time after the end of the test for the conversion lag to kick in and fully credit your campaigns.
- Looking at the wrong metrics – this meant that CTR, CPC and Impression Share should not be used to measure the success of a bid strategy based on Conversion metrics. Some sceptics might see this as an attempt to divert attention from inflating CPCs.
- Making constant changes to campaigns – not letting the Smart Bidding work its magic. Manually stepping in and making changes to campaigns could affect performance and skew your view of the bid strategy’s performance. Drastic changes will of course send the bid strategy back into ‘Learning’ and re-set the timer on that phase.
Automating Ourselves Out of Work?
The reps touched on the second biggest fear marketers have about giving up control of their campaigns (second to tanking performance and having to explain it on the phone to a client). We are automating ourselves out of work.
Google argues that we are not in fact doing that, but we are just evolving. They use the example of the introduction of the ATM & the subsequent fear felt by Bank Tellers. Bank Tellers’ main job was to dispense cash, so the introduction of the ATM appeared to be the death knell for that particular profession, however in the end it freed them up to take on more high quality, customer facing roles and the number of bank tellers eventually doubled.
In terms of PPC they estimated that our evolution would look something like:
OLD: Bid Management, Keyword mining, Ad Copy writing
NEW: Data Specialists, Customer Relationships, Competitor Analysis, Strategy
Everybody was wondering what the structure of a fully automated account would look like. It seems a major part of our job has been ‘splitting’ – split out your match types, split out your themes, split out anything you can and then write more granular ad copy.
Google’s suggestion was that ad groups should now actually be consolidated – throw all your match types back in with each other, consolidate themes, the more data the better. This feels like it goes against everything we are used to.
An audience member asked, naturally, if Quality Score would be a factor, to which the answer was no, however another member rebutted with the question “if I consolidate all of my keywords, I can’t write granular ad copy, and so my Quality Score will go down, so it is kind of a factor, no?” Google’s answer was that the ad group should be using RSA’s, which will allow fluidity. It was a very quick answer and showed that they are going all-in with the Smart Bidding, the answer to every ‘what about my structure?’ question was ‘let our automated feature deal with it.’
This was exemplified again when a user asked about seasonal fluctuations – I believe it began as a question about the Education Sector and Results Day, but it quickly moved on to Black Friday, which would affect the accounts ran by 99% of the people in the room. Google’s answer in a nutshell was to do nothing, more data is better for the bid strategy.
Google gave two scenarios for performance changes based on a seasonal fluctuation, and what you should do about it:
Scenario 1 – a large increase in traffic to the website.
Action – Nothing. Provided the conversion rate and CPA metrics are not going to vary too much, no action is needed (apart from uncapping your budget)
Scenario 2 – a conversion rate change
Action – for longer trends, make slight ongoing adjustments to the target CPA or ROAS based on the changes that you see. For short spikes (like Black Friday) adjust your targets beforehand – if you know your ROAS drops by half, cut your target by half.
In answer to the question, ‘are we automating ourselves out of work?’ it appears that we are evolving, as Google said, but we just don’t know fully where it is leading at the moment. For a while our role looks to be a hybrid one where we apply and test automated strategies and step in to take back the reins when they don’t work as we had hoped.
All in all the event was very enjoyable, the venue was great, all the speakers were very knowledgeable and engaging. There was a variety of activities and plenty of Coffee breaks. The most enjoyable aspect for me was the verbal jousting during the question sections, one audience member even bluntly asked “isn’t there a conflict of interest here as our clients want Sales for the best ROI, but you just want to take as much money as possible?” which got a chuckle from the crowd. We really got into some of the nitty-gritty during these questions, Google had good answers for everything and we all left with plenty of food for thought.