On- and off-site A/B testing is one of the mainstays of UX analytics. As most of you will know, it involves comparing the reactions of users between two versions of a CTA, page, button, email, or any visual or text-based media. Further uses include testing business services, discounts, marketing methods and communications between users. For marketing departments, every step within the funnel can evolve through the guiding results of these comparative tests.

One of the downsides of A/B testing is the need for a control group - A. Without an existing reaction to compare to, gathered data is of little use. This makes it difficult for startups to fine-tune their digital presence without making expensive mistakes. Furthermore, numbers count. The reaction of five visitors may be completely different to the average reactions of a few thousand. Again, startups are unlikely to have extensive traffic - as is also the case with many established companies. 

Another A/B negative is time. As Neil Patel tells us, these run from a few days to a few weeks and are best carried out one at a time to get accurate results. Tests that end too soon or tests running simultaneously (multivariate tests) can skew results. Naturally, time is also required to analyze and implement every result.

Once a business is up and running and has had time to test the more profound aspects of its funnel, A/B testing can then run in the background, helping to prepare new campaigns or suggest adjustments attuned to changes in customer behaviour. Again, this is a long-term expense that, although essential, produces less and less cost-effective solutions as your online presence adapts to its targeted public. With the rise of artificial intelligence, perhaps machine learning is the solution?

Up until recently, SaaS AI solutions were thin pickings. In mid 2019, Forbes reported how AI-based marketing patents were the fastest-growing global category at that time and that 83% of machine-learning startups were still in the early stages of funding. Now these drives toward providing the  prototypes of AI marketing solutions are coming to fruition.

The difference between current A/B testing and AI solutions can be described in a single word - prediction. Machine learning looks at facts and figures to predict an outcome. It could, in theory, tell you which CTA to use according to target group, product, season and even time of day. This would be based upon its algorithm and the mass inputting of data over time. During later years of development, forward-thinking AI concerns have been testing their machine-driven software’s results. Whether they have compared marketer gut-instinct as group A and AI as group B and found AI to be the outright winner is, as yet, unknown.

We know that a computer - when first created - is as intelligent as the human that inputs its data. But emotion, instinct, social interaction and communication are extremely difficult to translate into computer language. Without this extra dimension, exactly how predictive can software be about human user experience?

Creative and performance marketers are already being offered marketing AI solutions such as Pattern89’s suite of predictive software. Such a service would be of extreme interest to startups with larger marketing budgets that don’t have the visitor numbers necessary to carry out intuitive A/B testing or don’t want to waste time setting up and analyzing these tests. Advanced AI platforms like Anodot - a UX, revenue and digital partner monitoring package that uses AI to forecast incidents but has not yet stretched to cover creative marketing - are also readily available. Kameleoon’s platform offers client-side multivariate and A/B testing tools based on their AI solution - its eCommerce sign-ups include Lidl, Coca-Cola Europe, Swarovski and Shiseido.

The above-mentioned brands hint at what is probably the biggest problem for small to medium businesses wanting to integrate the latest, highly-researched AI algorithms and, in doing so, completely replace less-intuitive software.

Pricing information concerning the higher-spec vendors is far from transparent. Companies that do provide prices seem to offer cheap plans that don’t match the lengthy processes necessary to develop an AI-driven solution. How can you offer a cheap plan for a new technology that has taken years to perfect?

Any programmer can set up an algorithm that learns from the data it collects. Exactly what this algorithm measures and what metrics are used to guide future decisions mean the broad range of resulting programmes are rarely on an equal footing. Computer science is, after all, a science that depends on preliminary research, case studies and theoretical and practical testing that must deal with millions of variables.

Current medicine is the result of thousands of years of trial and error. The first computer that could store information as memory was invented in 1941 by Atanasoff and Berry. We haven’t had much time to test them. Even though computers significantly speed up the learning curve, understanding how they learn is another topic altogether. And while marketers have existed in various forms for thousands of years, the combination of marketing career and future-proof computer-driven technology has not yet been perfected.

The closest we have come to a fail-safe machine that can be implemented in A/B testing at a price most of us can afford is IBM’s Watson. Yet at this stage, Watson is still only advertised as an augmentation of data science practices such as auto data shaping, hyper-parameter optimisation, feature engineering, guided A/B testing and end-to-end flow improvement.

In short, it is still too early for small to medium businesses to make the switch to predictive AI; those that do must either pay a price only the big brands can afford or settle for unproven algorithms that may or may not improve UX and increase ROI.

Until the well-researched solutions come down in price or the lower-priced solutions prove their accuracy in a broad range of scenarios, marketer gut instinct in combination with non-predictive, analytic tools will remain relevant. And A/B testing, although limited in its lifespan, should still be the driving force for strategic change and amendment.

All of which brings us back to the vicious circle that is A/B testing for startups. If predictive software is beyond our budget and daily visitors to existing web pages and social media profiles a mere trickle, how can initial data be reliable? How do you retrieve qualitative data if you can’t access quantitative data?

One method is to use the services of an analytics company that has paid for predictive software; however, as we have seen, predictive software is not always as advanced as it could be. Even so, renting high-spec AI services via a third party would be the ideal solution. Thanks to the high number of patents and the developers’ careful guarding of prime technology, these solutions are less prolific than you might think. Certainly there is the option to seek out a third-party distributor of mid- to mid-high range AI solutions.

Another tip is to buy website traffic - not from the disreputable sellers of bot visitors, but from the providers professional marketers turn to for en-masse, human-generated results. Try the better-known and respected Max Visits and Ultimate Web Traffic. By ordering tens to hundreds of thousands of human visitors from niches that apply to your new (or existing) business, quantitative data is no longer an issue.

As for qualitative data, even if only 1% of these masses engage with your chosen domain, this is still more than would be generated by an unknown brand campaign, and much, much faster than any SEO strategy. Paid web traffic can immediately solve the data-generation issues of lesser-known brands.

Similarly, for smaller companies new and old without an internal marketing and analytics department, how do you know what to test and how long do you test it for? Most newcomers are surprised at the length of time it takes to collect accurate results for a single test. A basic A/B test calculator - there are many free versions available online - will give you an idea. By stopping a test mid-run, you risk error; the success of A/B testing depends purely on the accuracy of the result. If you are thinking of taking a quick analytics course so you can test your users’ experience without help, think again. A/B testing is a science. Leave it to the scientists.

When starting from scratch, how do you know who your company’s ideal customer is? To run any A/B test, you need to input your Ideal Customer Profile. While a new dog grooming startup can be almost 100% sure its customers are dog owners, other factors apply. What about friends of dog owners who want to buy them a less run-of-the-mill gift?

The truth is, very few new businesses can build up an accurate Ideal Customer Profile without casting out a significantly wider net. This is yet another reason why buying visitors in the form of paid website traffic is recommended. By selecting different niches and measuring reactions, even if the majority don’t stay on the page for more than a few seconds - a 2% engagement rate from a group of 50,000 visitors means 1,000 registered reactions - you get extremely useful insights into who your ideal customers are. 

If, by way of an example, you order 55,000 visitors from another reliable and trusted provider - Web Traffic Geeks - you could ensure all of these are pet enthusiasts by selecting pets from the list of available niches. However, by selecting other niches such as recreation and leisure (dog-walking) and gifts, your dog grooming business might discover other interested parties. Their input could change potential A/B results and their accuracy. Even established businesses with a well-researched ICP use paid traffic to cast out the net for unusual interested targets and so increase the accuracy of the changes advised by their analytics department.

Unfortunately, it seems that budget is the main driver of top AI predictive software, at least for the immediate future. Accurate, predictive software is priced out of the market for all but the biggest brands. To jump on the current bandwagon you need to dig deep into your business pockets. 

As we have seen with the solar energy industry (solar panels), those with shallow pockets can rest in the knowledge that time will increase competition and lower prices. The innovative enterprises that paid for patents will be forced to take the next steps in machine learning to provide an even higher level of technology at prices they can justify. This only means one thing - eventually, AI will take over the running of the entire analytics and marketing sector. But not in this decade. And after it has taken over, who is to say that consumers will not start to hanker for nostalgia - sales in its old, face-to-face form. And how will a machine overcome that?

For now, intuitive, affordable and dependable AI solutions can only take the place of less robust analytics. This isn’t necessarily a bad thing. It gives us plenty of time to A/B test human (A) and machine (B) marketing skills - because only when these results are in can we make the right decisions.