I often get asked about IBM‘s position on Accelerated Mobile Pages (AMPs). AMPs are mobile-first web pages that you submit to Google, so that it caches them for quicker load times. Those who follow the SEO field have the impression that you can’t really do SEO in the age of mobile without them. Not only is page load time a huge ranking factor, but there is widespread speculation that AMP is a ranking factor in itself, despite Google’s explicit denial that it is a ranking factor.
People are often shocked when I say that we don’t do AMPs. But there are three good reasons not to use them. They all boil down to user experience (UX). If you run a commerce site for marketing or sales, there is really no way to build a suitable user experience in AMP. Why? Because Google’s specs for AMP pages are designed with one goal in mind: speed. That means they are flat HTML with no dynamic, interactive, or tracking mechanisms at all. I’ll get into what that means for a modern commerce website. In short, AMP pages are a non-starter for our marketing site.
We use various methods to track visitors to our experience, broadly placed into two categories: cookies and pixels. Cookies are bits of code that we place on a user’s browser when they land on our site. Pixels are bits of code that we place on our pages. In the simplest possible terms, tracking users through their information journeys is about matching their cookies to your pixels.
In practice, it’s a great deal more complicated than this. For starters, users come to our experiences from a variety of devices and browsers, so sorting their cookies and associating them all with the correct unique user ID is a big challenge. We have a system that does this using AI. As in most AI, its accuracy improves with more tracked interactions with users. If a user clears their browser cache of its cookies, we’re effectively blind of who they are when they land again with that browser and device. But because we aggregate all devices and browsers for users into one user ID, we can associate existing user IDs with new cookies if they do enough on our site. But it’s not the end of the world if we have multiple anonymous IDs for the same physical person. You can synchronize them after they register, either online or offline.
Secondly, we run a lot of services on our pages, each with its own pixel provider and coding system. For example, we run A/B testing on all our pages continuously. This is a service that tracks the different versions through third-party pixels. Need to know which version a particular user ID (cookie profile) landed on from a particular referring site (search, social, etc.)? The A/B testing service sorts this out, and associates each visit with user behavior once they land.
The main reason to do all this tracking is measurement. In the past, we could only know about users if they opted into our services by filling out a form to download an offer. But only a small percentage of our traffic and engagement results in this kind of conversion. Most user interaction with our site is earlier in their buyer journeys. We need to measure this behavior so that we can do a better job of helping users through their journeys.
Needless to say, none of this tracking can happen on AMPs. All you can measure from AMPs is traffic, clicks, and bounces from anonymous users, who might be bots or, as The New Yorker cartoon implied, dogs. Combined with ranking and referral data, this might be enough for you. But modern commerce sites need more for two primary reasons: attribution and nurture.
Attribution is the process of giving credit where credit is due. In the old days, everyone used last-touch attribution, giving all the credit for the last piece of content a user consumed. This might be okay for some B2C contexts, where people make impulse buys all the time. But in B2B, a prospect might touch 20 pieces of content on her way to purchase, and 10 of them might not even be on your site.
Giving all the credit for all of these interactions to the last thing she did is woefully inadequate. As an SEO, this has been the bane of my existence. I’m constantly trying to prove that the traffic I helped generate had positive business results. If I can’t track user interactions from search referrals to purchase, I can’t do it.
One of the reasons we never have trouble getting investment in paid search is you can attribute it to purchase, because tracking is built in. I have spent my career trying to get similar attribution set up for organic search. Now that we finally have it, giving it up for the sake of AMP would be a Homeric tragedy.
The promise of a modern marketing site is automated self-service marketing. In a perfect world, you build all the content the user needs to go through the buy cycle, and you string the content together into client experiences that end in purchase, or advocacy, or renewal, or whatever your end goal is.
In practice, every user has unique needs. Study after study show that users take non-linear paths through your carefully designed experiences. And that’s okay. But it means you need to develop ways of dynamically tuning the experience for individual users to what they are most likely to need, and adjusting experiences to actual user journeys.
Think of the Netflix example I used in my last blog post. Each movie I watch changes the items Netflix recommends the next time I come to my home page. The more movies I watch from my profile, the better Netflix gets at serving me relevant movies. Similarly, the more interactions a unique user ID has with our site, the better you can be at predicting what it is likely to need in its next interactions.
If you track anonymous user IDs, you can nurture prospects prior to them opting into your more explicit nurture campaigns. Once they opt in, you can then associate all the interactions they had with your content prior to opting in. Now you have a rich user profile that can help you serve relevant content: through chatbots, recommended offers, or even in-app messages within your free trials. The more you know about a user, the better you can serve them.
The paradox of observer effects
Several years ago, before taking a hiatus to write our book, I wrote a blog post on this site about the problem with observer effects in marketing measurement. At the time, IBM had a very strict interpretation of what constitutes personally identifiable information (PII). This prevented us from storing cookies of any kind. So we could never measure repeat visitors. This was one reason why we could never attribute business results to organic search traffic.
We now anonymize the cookies, satisfying legal requirements, even the strict GDPR rules coming into effect in March in the EU. Any modern commerce site must do this step to survive.
The other kind of observer effect is not so easy to deal with. That is, our very measurement systems reduce the amount of interactions we measure. For example, each of the pixels we place on our pages has a small performance load. The slower a page, the less traffic you can generate from it. So there is a law of diminishing returns on the pixels we place on our pages. We might, for example, choose not to A/B test pages if we feel they’re optimized well enough. That would enhance their visibility to search users by speeding up load times.
AMP epitomizes the choice to reduce your measurement load in order to have more traffic to measure. You can’t know who is visiting your pages or what they have consumed before. So you keep serving them the most relevant content for their search queries and hope that it serves their needs. If all you want to do is help them learn about what you offer, this might be enough. But if you want to help them do business with your company, it probably isn’t.
Can you have AMP learning experiences that drive users to your commerce experiences? Yes, but you could never attribute that traffic to the business it drives.
In the final analysis, it’s probably better to serve fewer users with more relevant experiences, if it means you can attribute leads and sales to those experiences.