A packaging view on Artificial Intelligence

* 7 min read

Artificial Intelligence is, whether you realise it or not, already likely to be part of your daily life.

If you’re a Facebook user, AI is attempting to remove fake news from your feed after criticism of some of the (human-posted!) content during the recent US election. If you’ve got an Amazon Echo playing music in your kitchen, it’s doing a tremendous amount of AI-driven processing to understand your request for the latest Taylor Swift album (ok, maybe that’s just us!), and if you’re using Siri on your iPhone or Cortana on your Windows device then it’s AI that’s translating your spoken queries into requests and interrogating your diary to suggesting reminders. If you’re an insurance actuary in Japan, you’re probably very much more aware of what AI might mean for you – Fukoku Mutual Life is planning to replace 34 humans with an AI-driven payment calculator which is (allegedly) more accurate and faster than the people it’s replacing.

Untitled design 41

In our industry, AI is likely to have a somewhat different impact – we’re a long way from being able to use AI to do any of the creative tasks a designer would do, or to be able to automate the complex series of checks that goes into an artwork or pack copy approval (though we can certainly use it to speed some of them up and provide a quick check first). In our industry, there’s a few likely wins from AI – some of which are still pretty much in the realm of sci-fi, and some of which we’re already working on delivering to our customers.

We’ve mentioned AI in other blogs, but here’s a more detailed look at my top three predictions on how AI will impact the retail and FMCG world of packaging launch:

1. Project Planning will become far more proactive than it’s possible to achieve today

This is one we’ve mentioned before, but I’m unapologetic about mentioning it again because it’s such a huge potential benefit to project managers struggling to get a portfolio of products on shelf on time. AI (and its cousin Machine Learning – which is basically AI that ‘teaches itself’ based on a set of data) allows us to mine tens of millions of data points across tens of thousands of projects (in many cases, across multiple retailers and CPGs) and build up a model which we can use for predicting future events.

The human equivalent of this is a project manager getting to know that “supplier X is always late” or that “printer Y is a good choice for fast track jobs” – but with AI we can refine that down to the most minute combination; perhaps Supplier X is in fact late on 99% of lines, but early when that line is Irish Beef instead of UK? Or maybe Printer Y is great for blister pack jobs, but not so good for gravure runs – or perhaps a job that goes to Supplier X and Printer Y will be early, but if you change either of them, it’s likely to be late.

That kind of modelling means we’re able to predict on day one of a project whether or not it’ll be on time, and to allow you to make smart choices about where to send work which give you the highest chance of getting it on time. It means for the most part, jobs will be on time all the time – and for the exceptions that do go off track, we can make some amazingly accurate predictions about whether a particular rescue strategy is likely to get you back to where you should be. It means shortening your normal project timelines without resorting to fast tracking everything – and it means that when you do need a fast track, you can get the absolute minimum timings at the outset.

2. Drug interaction prediction will shorten regulated product launch timings

For the Pharma industry, interactions are a big deal; good old paracetamol (acetaminophen to our American cousins) is a staple of most people’s bathroom cabinets and is generally considered to be a pretty safe medication as long as you don’t overdose. The mechanisms by which paracetamol becomes toxic are quite complex unless you’ve studied biochemistry, but in essence paracetamol is normally broken down in the body by two routes – one harmless, and one which produces a toxic ‘metabolite’, or by-product (which is called N-acetyl-p-benzoquinoneimine or NAPQI, for those of you who like pub quizzes…). NAPQI is produced even if you only take a tiny amount of paracetamol, but it’s produced in minute quantities and – most importantly – it’s then broken down into a non-toxic chemical very quickly. If you overdose on paracetamol (or if you’ve got a rare genetic predisposition to paracetamol overdose) the “safe” route becomes saturated and the second, NAPQI-producing route, takes over – which in turn produces more NAPQI, which in turn saturates the mechanism for breaking down NAPQI – and then bad things happen to your liver.

If that all sounds complex, it is – and that’s one drug which we’ve been using since 1877, which we’ve studied extensively, which we now understand pretty well, and which is usually administered on its own. Imagine trying to work out all that metabolic action not just for paracetamol but for a combination of drugs (paracetamol and ibuprofen, perhaps?) which might react together or might create metabolites which react to each other, and so on. As an aside, not all interactions are bad – in fact, some chemotherapy drugs only have their intended effect because of the interactions with the metabolites of other drugs which are also ineffective alone!

Companies like Leeds-based Lhasa (www.lhasalimited.org) are using AI to predict those incredibly complicated interaction pathways and to help predict the effects (both positive and negative) of drugs in the human body – that should mean shorter testing and regulatory approval times for medicines and vitamins, which will translate to shortened launch timings and longer sale windows while patents are in force. Longer-term, combining individual genetic information with those models could lead to individually formulated drugs and a fundamental change to the way we sell and launch medications – personalised packaging move over and make way for personalised contents!

3. Bots will make “one-to-one” conversations with millions of consumers possible

Bots are the hot topic in lots of tech arenas; Microsoft has its own bot framework and a variety of AI voice, text, image and face recognition libraries released as part of its Project Oxford efforts, and Amazon recently released their own equivalents. In their simplest forms, bots are little more than those annoying “intelligent voice recognition” (IVR) systems that ask you to say one for customer services, two for sales and so on – in their most advanced incarnations, they’re increasingly hard to tell apart from real people. The key thing in effective use of bots is finding a particular niche and making a bot that is capable of servicing that requirement well – we still lack the generally-available computing power to make a bot that’s good at everything.

Pernod-Ricard has already jumped on the bandwagon, with a bot that allows their customers to ask for suggestions for party drinks – in essence, an AI “bartender” you can ask for advice and guidance on what to buy for your get-together; it will chat with you and work out whether you’re entertaining a knitting group or a rock band and suggest an appropriate mix of bottles to buy to stock your back shelf (presumably with an aniseed theme..) as well as recipes you can pre-mix and so on. It’s a pretty interesting novelty idea, and I’m sure it’s helped get Pernod on the menu at more parties than it was before – but it’s really just the tip of the iceberg.

[24]7 – a company that makes its money out of selling bots, so they’re perhaps a little biased – recently released the results of a survey showing that 40% of US retail customers would be comfortable talking to a bot and that almost a third would actually prefer to talk to a bot online than a human face-to-face or over the phone. If those figures are representative when scaled up to the overall worldwide retail market, that could represent a very significant shift in the way people pick their products – and an equally significant shift in the way they interact with the brand owner (be it retail or FMCG) after the purchase. It clearly isn’t feasible to staff a call centre inviting every customer who buys your pre-packed steaks product to call and ask for suggestions on how to cook them or what to serve it with; with a bot answering the questions it would be comparatively easy to field those questions and generate a premium post-sale experience. Top-tier brands are growing – Kantar’s post-Christmas results show a growth from 5.7% last year to 6.3% this year – and the kinds of “experience enhancing” opportunities that bots offer gives further incentives to customers to buy from those lucrative lines.