Advertising platforms are offering more automation tools. However, it is important to look at the bigger picture and not always view channels in isolation. There are several key examples when automation may not be the best solution for retailers using paid search.

Over the past few years there is no denying the huge investment Google has made in machine learning and their ever-growing offering of automation. As consumers we increasingly expect more from brands as we navigate the web, and with so many user signals to process, improvements in personalisation, bidding and creative testing can help retailers drive growth.

However, there is an argument to say that this shift doesn’t always have a positive impact, and in some cases automation is costing businesses money. So how do you know what is right for you? And when should you use it? In this article we’ll explore the pros and cons of automation and address some of the changes retailers may need to make in order for their Google PPC campaigns to work best for them.

The automated products available through Google predominately centre on bidding, but also an increasing shift towards more dynamic and responsive ad messaging and ad placements.

When it comes to bidding and automation the best approach for you will vary depending on your objectives. If driving traffic is your goal you may want to maximise clicks, whereas if you are aiming to drive conversions you’ll want to focus on smart bidding techniques such as targeting ROAS (return on ad spend), maximising CPA (cost per acquisition) and not forgetting the more recent Smart Shopping, which “combines Standard Shopping and display remarketing campaigns, and uses automated bidding and ad placement to promote your products and business across networks.”

When may automation not be your best route?

While the shifts towards automation are positive, there’s also a need to consider when and why this may not be your best route for you. You may feel you are falling behind industry best practices by not rolling new features out, but it’s important to remember Google doesn’t know your business as well as you, and there are several key examples of when sticking to the good old fashioned manual approach or looking at developing your own proprietary tools and scripts may be best.

For omni-channel retailers

Several of our clients at Vervaunt have multiple physical store locations, and for others factors such as telephone sales are still a key part of their business. In these instances, it’s important not to view the channels in isolation, but to understand what impact online may be having on offline and vice versa.

A key example here is a music retailer we work with, who has stores across the UK and a catalogue with thousands of products. For each store they have bestsellers which may underperform in terms of ROI, but ensuring we have strong online coverage is critical to drive the footfall and stay ahead of competitors, which automation wouldn’t yet optimise towards. In the same vein, they may see products selling well through other online channels and telephone sales but have limited coverage on Google, and we can take these learnings and adjust bids accordingly.

There will also be products which generally resonate with the brand and are important to keep coverage on. Some consumers may only visit a store if a certain collection or brand is available, you therefore want to make sure you keep coverage on these without risking a drop in bids or impressions due to automation. These may include loss leaders which you know deliver a positive customer lifetime value whereas Google currently looks only at the individual purchase ROI. 

For stock and inventory nuances

Working with online retailers, stock and inventory is a key consideration for us at Vervaunt when optimising accounts, particularly across our clients in the fashion industry. Each of our clients have their own individual nuances, and understanding this can help drive stronger results compared to when we have tested automated bidding. For example, a luxury sportswear brand we work with knows that a large proportion of their sales come through sizes small and extra small. When these sizes are low in stock or sold out it becomes unprofitable for us to only bid on remaining sizes, as we know that these will see a much lower conversion rate. We developed an in-house tool that flags when a medium or above has spent over a certain threshold and when the small and extra small are out of stock. This notifies us to then go into the account and either pull back or exclude the other products across shopping until stock is replenished.

The way in which another client of ours manages their stock within the product feeds means that products will never be shown as out of stock even when they are. Therefore, automated strategies would continue to bid towards these products but for us that would be wasted spend. 

Where profit margin varies

This is another area that will vary quite vastly between advertisers and there may be areas that while results look good online, from a broader business point of view become unprofitable to bid on. Further to this there may be categories not worth including at all or in the contrary, certain brands you want to push further if you are a reseller.

While you could simply exclude certain products from feeds, if you are using Smart Shopping automation for example profit margin won’t be a consideration. For one of our clients we pull in dynamic cost price from an external source, and our 3rd party feed solution then updates the current profit margin into the feed, which then helps us optimise activity accordingly, leading to more informed bidding, and a good example of where machine learning doesn’t consider all of the most relevant data points or some advertisers.

In future, it is likely that we will see bidding strategies allow such profit margin considerations to be included in your targets (some of the 3rd party bidding solutions already offer this). As a retailer in the short term, you might want to segment out higher and lower margin brands/ categories and then have multiple bid strategies targeting different KPIs based on the margin.

For ecommerce bestsellers

We have also found when testing automation that often top selling products are not being aggressively pushed by Google. They will often find a number of SKUs that achieve the targets set, and then simply allocate the spend here, whereas you may have stock of a new product for example that you know will sell well that you want to push, or top selling products across the business as a whole that have low impression share across Google. You want to make sure automation is not limiting coverage across new products and general top sellers.

Pros and cons of automated bidding

Pros;

  • Pretty straightforward to set up and saves time to focus your attention on other areas.
  • Google’s machine learning will use a multitude of signals to help optimise your campaigns, which will be unattainable through manual bidding.
  • Bidding machine learning has great sophistication in terms of portfolio based bidding and decision making, 24/7 bid changes and constant tweaks to find optimal bids.
  • You can set goals at different levels, e.g. for ad groups or campaigns, offering control.

Cons;

  • Whilst automated bidding does take into account signals such as device, location and time of day, it doesn’t have the ability to take into consideration factors such as flash sales, media coverage, weather, competitor changes or sports results, to name a few factors that may be impacting search volumes.
  • The current Covid-19 climate is a prime example of where external factors will be impacting performance and where you would need human context to support decisions. For example, you would want to begin upweighting categories and products you expect to perform well, such as home and garden, but machine learning would likely take longer to see this opportunity.
  • Google lacks your discretion on what keywords or products are most valuable and doesn’t take into account factors such as bidding due to stock or margin.
  • Again with the likes of Smart Shopping, your ads can appear across more Google inventory, which may not be aligned with your brand. You also have no visibility on placement-specific performance.
  • You need to ensure you have accurate conversion tracking in place, and reach a certain threshold in order for Google to ‘learn’.

How to decide what bidding to use?

If you are looking to increase the adoption of automation across your account it’s important to introduce this steadily so you can begin to measure the impact and get a good understanding of how it is performing compared to your manual campaigns. A/B testing through drafts and experiments is a great place to put this to the test. These can easily be set up for your search campaigns, but you want to ensure you are isolating your bidding tests and aren’t testing other elements such as new ad copy, which may be impacting performance of the account at the same time.

Through drafts and experiments you can determine the proportion of traffic that goes through automated or manual bidding, dictate your test timeframe and view the comparative performance as you progress. You do need to ensure that the splits provide Google with at least 30 conversions in the past 30 days. It’s best practice to start with your historical average CPA or ROAS as your target. Google bid simulator forecasting tools can also be used to help you find an optimal target.

The key things to remember is keep your test simple then monitor and measure the impact. There will be a period of learning for a week or so where Google tests different approaches from the algorithm, but following that it should be working at full capacity, and you can monitor performance over the following weeks before making a decision about progressing further with automated bidding or not.

What about Smart Shopping campaigns?

Smart shopping sees most control being handed over to Google, while the ‘time saving’ element may seem attractive, this can’t be tested like the above, and there are many aspects to consider when monitoring the performance and understanding if it’s right for you as a retailer.

While Google suggests you should put your whole product catalogue into one campaign I’d suggest starting small and testing specific product sets. Smart Shopping will always trump standard so make sure to exclude your ‘test’ products from existing shopping activity. When testing Smart Shopping we’ve always found the best approach is to group products you possibly have high stock of or are struggling to shift but avoiding your bestsellers and the ones for which you really want to monitor performance.

If this is the path you choose to go down, there definitely needs to be a shift in the level of control and accountability you are willing to have for your own results and performance. Two key things to bear in mind here, you will have no control over negative keywords and no way to attribute performance back to search terms and there is no ability to upweight certain products should this be needed.

Conclusions

Advertising platforms are offering more automation tools. However, it is important to look at the bigger picture and not always view channels in isolation. As discussed, there are several key examples when automation may not be the best solution, and this will very much depend on the complexities of your business. There is no real right or wrong answer here. The key is to test and to understand what works best, and remember, no one knows your business as well as you!

For more on this topic, download Econsultancy’s PPC Best Practice Guide.