• -
robot1[3] 1000x500

Save your automated marketing project in 2016

Tags :

Category : Blogs

Automated Marketing (or the merger of AI technologies and digital marketing systems, like email, display and on-site targeting) is the new ‘buzzword’. It’s being fervently discussed and was regularly listed in the many “2016 predictions for data” articles, last year.

Now, normally, I dislike predictions (perhaps somewhat ironically considering my profession!). The fact is that most of these types of projections don’t turn out to be true, or are so specific that they are irrelevant for most people. The noise around Automated Marketing, however, has chimed with many of the conversations I have had with clients.

Big Data, Move Over

It’s one of the key buzzwords at the moment because it’s the next step on from “big data” (that was last year’s ‘next big thing’). AI is marketed by many tech companies as the merger of marketing communications and machine learning. This duo enables you to predict a consumer’s behavior and then to automate an exact message to reach that person through a specific channel, at a precise time.

This, according to the AI sales pitch, means you no longer need to worry about large teams having to build and manage segments to understand customer behavior. Sizeable departments don’t have to work tirelessly across email, display and/or any of the many channels through which you might want to target customers.  As one vendor put it: 

“a single staff member can execute complex and ongoing campaigns and can connect with many more customers than would be possible manually”.

When is a buzzword not a buzzword?

The idea of melding marketing with AI techniques is a great one – and it’s a vision we share. The benefits of better, more effective segmentation, and indeed, personalisation, are clear. But we are also aware that often the sales pitch offered by tech companies doesn’t quite chime with our clients’ experiences once they start putting pitch into practice (as it were).

This seems particularly acute with Automated Marketing in 2016. As an emerging ‘buzzword’, it can be difficult to separate the reality from the hype. Automated Marketing, as a subset of ‘Machine Learning’, is right at the peak of the downslope of the latest ‘Gartner Hype Cycle’. This is often called the ‘Peak of Inflated Expectations’ – and you can tell that from the sales pitches themselves.

This means that in 2016 Automated Marketing will start heading swiftly towards the ‘Trough of Disillusionment’. The key question is, then:


How can you make sure that your Automated Marketing project avoids hitting this trough in 2016?


The main reason why so many new technologies fall out of favour is that the people and process part of the activity is ignored. So how can you implement an Automated Marketing project that just might work?

Automated Marketing in Practice

Before you even start working with this kind of technology, you will already have some fundamental automation tools in place. Many of the existing channel tools (i.e. email sending tools, display and PPC buying programmes) already rely on some automated functions. For example, programmatic buying for display is now a well-established marketplace in its own right. OK, so:

Why bother adding another automated tool to your technical tool belt? 

The motivations are clear. The new tools available enable users to go beyond a single channel and target customers across many different ones. That’s significant – because we all know that customers use lots of different channels in different ways. Having a tool that works only in one channel will just not be good enough.

Full customer view

Targeting your consumers in this way first requires you to have a single customer view mechanism in place. To begin with, then, work with this tool to analyze how your multichannel customers behave. (Ideally, this will have formed part of your business case for the project in the first place – but if not, get this analysis done.)

Teach an old dog new tricks

Next you need to teach the tool how to perform. The challenge with automated modelling has always been that the tool will spot the most obvious correlation – unless it’s told otherwise.  Tasked with analysing Black Friday, for example, the untrained tool will come back with an absolutely certain correlation first time: people bought because it was Friday.  Likewise, if it were assessing financial savings products, it would say that the beginning of April (because it doesn’t know when the tax year ends) is the most important factor.  You will need, therefore, to tailor the tool to your own needs.

Build a good model

To do this you will need to build your own models and to understand your customer segments well. Building good models takes time. The best rule of thumb, for establishing whether or not you’ve constructed a good model, is your own reaction to it or its results. If “I could have told you that” is not your immediate response then you’ve created something that has gone beyond traditional human research capabilities. You’ve brought something to the table that surpasses anything that hiring another person or building a bigger team could have done.

In our experience it normally takes a couple of iterations to reach this point, so it’s really important to persevere. Have the courage of your convictions when building the model and you will get better results in the long run.

The second point to add is that it’s YOUR model. 

One sales pitches often voiced by tech companies is that because they have data scientists you don’t need to worry about it. It will be automated. It will just do it for you – sorted. Except, its never that simple.

Often these suppliers don’t provide you with any documentation on how your model works – “it’s a black box and it’s all done for you” – so you actually have no idea how you are targeting your customers.  That’s great for them – when it breaks, you have to use their engineers to fix it – but it’s not so good for you and will almost certainly cause problems down the line. Its not good for operational reasons, but also for data security and data auditing.

If, for instance, you need to prove that your customers have seen certain pieces of information before they buy (this is particularly important in regulated industries like financial services, or telecoms) using a black box to automate marketing communications, with no documentation, is a problem.

The best tools allow you to input your own algorithms so that you can remain in control of your own processes.

No rest for the wicked

Of course maintenance is essential and you will need to maintain both your tool and your model.  This is important for when it goes wrong – and it will go wrong at some point – or when you want to add a specific marketing campaign, by hand, as it were.

You will naturally be told that it’s not possible for the tool to go wrong. But, it will.

It will because it is still a process – even if it’s more hidden than usual. And righting those wrongs will be more difficult than usual: you can’t sit your model down and tell it off for not pulling its weight! You will need a data engineer who understands the model and can change it to help manage and maintain the tool.

At the same time you will need to have marketers and marketing analysts who can interpret and analyse what is going on across different channels and customer segments. They will use this analysis to create new campaigns and assets.

If you thought that you could reduce or redeploy your team as a result of marketing automation; think again. That’s one of the best ways to end up in the Trough of Disillusionment.

Most Importantly, People

Nonetheless you will be changing what your team does. The model and the AI engine will now be deciding what to show to which customers – so the role of your marketing team will change. You need to make sure they are on board with this and that they understand how their roles will evolve.

The last thing you want is a “saboteur” in your team, who will try and demonstrate why the new process isn’t working as well as it was before (when it was run by humans). The Luddites were very effective in disrupting and destroying automation technology in the 1810s – you don’t want that part of history to be repeated 200 years on!

Perhaps somehow ironically then, your people might be the most important part of any automated marketing project. Your technology can’t succeed without your team. Change can be challenging – and you don’t want anyone to feel as though they are being usurped. Your team needs to be prepared to work with the new technology – not against it.

If you want to succeed with AI technologies and safeguard your program from potential pitfalls – first then, ensure that you have the right processes in place and that your people are not only well equipped but eager to engage with this new way of working.

Sign up for our newsletter