Spend is Performance

Meta (Facebook) just unveiled at their latest media buyers conference that they are going all-in with machine learning (Advantage+) and merchants who stick to manually controlling ads will likely see lower performance versus those who get out of the way and allow Facebook to spend toward a goal.

That doesn’t mean you can relax. You still have a job. Advantage+ just means your job is a little different.

Have you ever owned a dog? They love to please you by doing their job. If they think you want them to fetch the ball, they will fetch until they collapse in a heap. AI in general, and Facebook/Meta’s Advantage+ is a lot like that. Be careful what you ask for; it’s a big world and if your ask is vague, your results will be messy and vague.

However, if your ecommerce business has a proven online sales process, quality content and ads, Advantage+ shopping campaigns are worth a try. Advantage+ is the purest realization yet of Meta’s push for ODAX – outcome-driven advertising experiences. If your goal is being met while Facebook is spending, then great, all you need to then do is review how and to whom they are serving ads for the sake of protecting your brand. In the simplest terms, Advantage+ shopping campaigns maximize performance and find new customers. Settings are streamlined, to say the least. You do get to choose in which country your ads will be seen, but that’s about it. Automatic placements and lowest cost bid strategy are baked in as part of this “best practices” approach. Frankly, calling them “settings” at all may be misleading. Essentially, your budget, your ads and your site (and/or shop) content are the settings for an Advantage+ campaign. The more focused, engaging, clear and market-specific your content and ads are, the better your results will be. 

Does that mean your targeting will be off? Will you get a bunch of junk results? Here is where the leap of faith comes into play. You need to be able to trust that Facebook is smart enough to recognize that you sell commercial kitchen equipment, for instance, not ice maker parts. You also need to be willing to let Facebook trial and error all over itself while it strums for the perfect chord. You won’t see (or pay for) the stumbles, however, and the AI doesn’t get upset by failure; it’s relentless and endlessly enthusiastic about its mission – again, kind of like that dog.

The core strength of Facebook’s Advantage+, the endless ad iterations and testing, can make the branding team a little nervous. When using Advantage+ placements, you might see some strange cropping or branding with your ads. It can add a banner (it picks one of your colors) and may insert a headline or other element – again based on your content.

This sets up a tough choice between traditional thoughts about the inviolability of branding vs maximizing results. Given that the demonstrated benefits of adherence to rigid branding are long term, we are curious and open to the idea that there could be unintended longer-term negative consequences (audience loss, diminishing recognizability?) when you allow Meta to mess with your look. The jury is out.

It would be easy to dismiss Advantage+ as merely an amateur’s tool. Indeed, a search for the term nets a lot of pundits downplaying Advantage+ because of the lack of granular control over demographics, creative, test elements and spend apportionment. Marketers have always based their careers on being at the helm – tweaking campaigns and closely monitoring results. Facebook’s Advantage+ doesn’t even want interference from diligent and informed human decision-making. Granted, this can (currently) only be true when we’re talking about millions or billions of data points, such as are at Meta’s disposal. 

Further, Advantage+ requires KPI history on creative and targeting, so you can’t run it on a brand-new ad account. Someone needs to do testing before deploying Advantage+ to ensure that it’s optimized for success. You can upload customer lists as demographic “seed beds” to help train the AI on proven successful profiles. You also control the existing customer budget cap – well, sort of. Some have found that Facebook always exceeds this setting (usually doubling it) – don’t set any higher than 10-15% if you want to end up with 70-80% new customers vs retargeting.

The bottom line is, Advantage+ might look simple, but we would advise marketers to not quit their day jobs!

If you find yourself choosing the manual experience, check your reasoning. Is it fear? Is it because the manual process offers more granular control?

It’s a bit like investing. Many people want to choose the stocks they buy, convinced they can outperform professional mutual fund corporations – or just enjoy the control. On average and over time, It’s an illusion. Just as, most of the time, and with much less effort, allowing Facebook’s Advantage+ to just do its job will net you greater gains over the long term.

Contact us to learn more about how to optimize your eCommerce site.

This conundrum can lead to paralysis – especially as we consider the effects artificial intelligence (AI) will have … and is already having …  on eCommerce development. Should you greenlight your IT team to use AI? Or should you block AI apps until they can be further vetted for risks and security? These are real questions merchants are wrestling with and we see both approaches (aggressive and conservative) currently being employed.

AI Developer Tools Today

AI Developer Tools Today

AI-Based Programming Assistants are already in wide use among software developers for code compilation, code debugging, and code-driven testing. Already, a few dozen tools like this exist and are being used by developers everywhere. Programming assistants can solve relatively simple but often time-consuming problems, making developers more productive and decreasing the fatigue that comes from executing repetitive tasks. 

For example, Codex is an AI tool (in Beta) from the creators of the Chat GPT system (which is itself frequently used for coding tasks) that translates natural language to code. Codex powers another similar tool, GitHub Copilot. Developers must be skilled in asking either of these AI assistants well-defined questions in order to produce helpful results. Left unchecked, AI assistants are also known to leave vulnerabilities or even create new ones. So choosing a developer whose expertise allows prudent and well-monitored use of the tools is of prime importance. 

A newer tool called AlphaCode shows amazing promise in solving more complex programming tasks, but still requires extensive problem definition, testing guidelines and outcome descriptions.

Another class of tools resembles the “auto-complete” function you frequently see in text programs and search fields. As a developer types a line of code, the AI program (Tabnine, as an example) will suggest completed code lines. The more these tools are used, the better they will become at anticipating what the developer is doing.

Wherever you have a defined outcome or set of outcomes, plus inputs for the AI to track against pattern recognition, then you have a feedback loop. AI thrives on feedback loops and is already making great use of them to provide new features and functionality for on-site search, product recommendations for personalized shopping experiences and personalization (see https://www.adobe.com/sensei.html). 

Looking Ahead

In the next 1-3 years, we believe we will see significant portions of simple development tasks nearly overtaken by AI. For example:

• The development of more “code-less” themes. These are themes for platforms like Shopify, Adobe Commerce (aka Magento) and BigCommerce that permit speedy modification and customization of the content, design and layout while no longer allowing users to edit the code. Instead, all CSS/design code changes are managed by AI. This should allow for cleaner, less bloated and more stable code because the AI will be able to quickly handle debugging and testing and won’t be relying on patching or shoehorn approaches that human developers often use to effect a change.
• Development of easier-to-use and faster design tools. Drag and drop has been around for a while, but imagine that the software will be able to anticipate and implement consistent-looking changes across the theme elements. Then, you can clean up any changes that AI didn’t get quite right on its first pass.

The Likely Path 

Software development, like many other skilled trades and professions, consists of a high percentage of tasks that follow a pattern or rely on standardized solutions. The skill and experience come into play on the details. AI adoption will most likely line up with those truths; without a doubt, AI use will be (and already is) widespread within eCommerce development but will not replace skilled programmers. That may sound naive, so we’ll put a finer point on it. Programming as a profession will definitely change as AI assistance becomes more powerful and reliable, but as long as the system needs to know exactly what problem to solve, there will be a need for highly-skilled developers and solution architects.

In the near term, AI will have much less effect on the more complex elements of website development like software architecture, database management, new feature development and advanced security.

If You Want a Job Done…

In general, like most technology-driven changes, AI integration into eCommerce programming will follow the Pareto Principle (aka 80/20 rule). Developers will get most of the benefit available from AI by using it for limited tasks. Other tasks involving more subjective inputs: (opinions, experience, preferences) would be very difficult to ask the AI to accomplish while fairly easy for a skilled human programmer to grasp. Like the use of robots in industry – AI will be used where it has a clear benefit and where it is economically feasible based on the scale of the activity.

Human developers use their native intelligence to translate expressed specifications (sometimes vague – and always contextual) into iteration attempts that must be checked by other humans (the project managers, the clients). This can be an unpredictable and sometimes arbitrary process when it comes to the details. 

What should eCommerce managers do in order to use AI to their advantage?

It is important to be on an eCommerce platform that is aggressively working to take advantage of the trending AI technologies. For instance, developers are already experimenting with codeless themes for Shopify (ie. nyla.app) that rely on AI for fast and simple design changes. Still, in an atmosphere of disruption there is potential for bad decision-making when it comes to choosing a platform. Software that relies too heavily on AI at this point may carry a load of unintended consequences – levels of maintenance and setup that would currently be highly specialized (and thus cost prohibitive) in the medium term.

More than ever, excellent and experienced guidance are your best ally. Pick an agency partner you can trust, with a track record of objectivity when it comes to platforms and tools. InteractOne doesn’t claim to be unaffiliated, but we’ve also never been shy about the strong and weak points of Magento, Adobe, Shopify, BigCommerce and all of the underpinning technology. Our role is to provide unvarnished advice, based on what we see happening in the ecosystem.

Don’t just take our word for it. Stay informed, by following industry leaders, experimenting with AI yourself, attending trade shows and seminars, etc.

Carefully review and vet new AI features before choosing to implement them on your site… don’t just use it because it has an “AI” label on it.  Do your homework to ensure the AI feature will provide a positive impact on your website.

Contact us to learn more about how to optimize your eCommerce site.