“Today, I had the kind of conversation I absolutely love.”

Personalisation has finally evolved in the last few years. We’re no longer talking about the Starbucks approach to personalisation (putting your name on it), nor are we trying to “personalise” large segments of people (e.g. show all people who’ve come from a campaign a single thing). Instead, today, I had the kind of conversation I absolutely love.

Talking to a large, well-known retailer who understood that personalisation was important – not only to improving the experience of customers who are finding good products, but also to maximising their ability to merchandise and sell better. There were a few key themes and pieces of functionality that we discussed.

Data collection and manipulation

Webtrends has been a company dealing with analysing data for 25+ years – that’s the pedigree we’ve come from. So, it should be no surprise that we’re looking to collect everything that we can about visitors using the site – what actions they take, in relation to what products, including the attributes we know about those products, etc. One of the fun challenges we’ve looked to address in the last couple of weeks is augmentation – working now with Google Shopping feeds, which retailers are so familiar with, to understand things like profit margins and stock levels as well as traditional attributes.

Once we have this, we have the confidence to aggregate it, store it effectively, and query it quickly to fulfil use-cases that affect the loading of web pages where speed is a crucial factor to success. This lets us build whatever we want, freely.

Custom UIs

We’re the only company I’ve come across that looks to build genuinely custom bits of UI for clients, so that they have an easier time doing things that might otherwise seem manual.

Whether we’re talking Banner Messaging for Travel customers to manage service disruptions, Product Badging for retailers, or something completely unique to solve a completely unique use-case – we’ll build it. It’s as simple as that.

We feel successful when clients can easily go about their day, and if that means completely bespoke parts of our UI, it’s completely worth our while.

The problem with Machine Learning

This feels like a topic worth talking at length about, but for now I’ll keep it concise.

Let’s say you’ve got a banner, and you want to know who to show it to. Machine Learning is great at finding groups of people that respond to it well.

What you’d struggle to do with it, however, is run 1000 banners to 1000 groups of people, each of whom would convert extremely well if they saw it. The reason is quite simple – you’re not going to have the time to build 1000 banners, and it’s often a bad idea to just “let the machine do it”.

“Don’t be fooled”

So, where “AI”, “Deep Learning”, “Neural Networks” and “Machine Learning” are all terms that people throw around today, please don’t be fooled – there’s no such thing as a silver bullet for marketing – if you want something that’s extremely effective for CRO, you will still need to put the effort in.

Cookie Cutters in the Product Recommendations/Social Proof Space

Let’s take Social Proof messaging as the example. If you were to go to one of the “leading” 5-6 companies that specialise in this, what are you likely to see as examples?

  • On the PDP, the product image is has a giant message or two over it, including things like “Purchased X times this week”, “X people have viewed this product in the last day”, etc.
  • On the PLP, they’ll show “X people have viewed/purchased this product in the last day/week”, just underneath the product image.

It’s always the same few messages, in the same positions, with the same style, from every tool – they’re producing Social Proof messaging using cookie cutters.

Let’s compare that to how we do things, keeping in mind that we are a Testing provider first and foremost:

  • Impartially test whether their current messaging actually gives the advertised lift – you’d be surprised.
  • Run our own (fractional factorial) MVT, looking at things like colour, copy/tone of voice, duration period, minimum value thresholds, location, etc.
  • Identify what works overall, and also what works based on as many contexts as possible
  • Serve the right slice of Social Proof messaging to the right people – i.e. Personalise it.

Product recommendations are no different – quite often taking just a product image, name and (maybe) price because it’s easy, whereas we’ll make sure we’re patching into Quick View popups, hooking into onsite Add to Basket functionality, lifting stock levels and size/colour matrices from other pages, etc.

The clear difference

We don’t just dangle a platform in front of people and repeat “AI-Machine Learning-Blockchain-Neural Net” until contracts get signed. Everyone’s different, and we’ll literally build something brand new, from scratch, if the situation calls for it. And, we’ll actually give thought to what we’re doing. Building new/bespoke functionality is fun, it’s exciting, it often lets me tell some fun stories, and most importantly it’s the reason I enjoy being a developer after 15+ years at it!