How machine learning is helping marketers get the edge

Reinforcement learning can help you continuously improve your customer knowledge

AI systems are getting smarter every day thanks to rapid advances in machine learning. The wrinkle, however, is that most algorithms still require careful human supervision, accurate data and constant fine-tuning to function effectively.

Reinforcement learning (RL) attempts to minimise the human input and trains machines to ‘learn’ themselves without being give explicit instructions. Over time, the algorithm keeps self-optimising its performance until it masters the game. AlphaGo Zero, an RL algorithm created by Google’s DeepMind team, was trained this way and recently became the world’s strongest Go player.

Microsoft recently made available a new Custom Decision Service on Azure, powered by reinforcement learning. Among the described use cases are:

• Text and video content personalisation for news/media portals
• Optimised ad placements on web pages based on user behaviour
• Advanced product recommendations for e-commerce websites

Unlike most ML (machine learning) algorithms, such systems do not need to be constantly programmed to recognise the right course of action. Instead, they automatically gather and distinguish unbiased and relevant data that can be used for decision-making and self-tune their performance as new data becomes available. This way your system continuously learns about your customers, their shifting preferences and behaviours, and prompts you to take the best possible action in real time.

Beyond personalisation, RL algorithms are now being tested for cross-channel marketing solutions and real-time bidding in display advertising.

Get prioritised marketing insights

Marketing analytics has become granular, tracking each and every insight imaginable – from social media sentiment expressed by some micro group on Twitter to your website's technical performance, benchmarked against fifty other competitors. This is good news as we can precisely measure, estimate and optimise our actions up to the point when enough data becomes too much data. The volume of accumulated data stops being productive and ultimately leads to analysis paralysis – too much data that makes too little sense.

Hitting this plateau means that you are doing great with data analytics, but you may need to bring in machine learning tools to do the analysis part. To give you an example from the search marketing industry, our ML-based system Apollo Insights can perform billions of audits automatically every day and transform the accumulated data into prioritised insights, indicating where the most crucial impact can be made.

By using machine learning tools that have already located the data and done all the heavy lifting , you can then just focus on performing meaningful actions and making an impact.

Introduce dynamic pricing to maximise profits and customer satisfaction

Discounts, loyalty bonuses and promos – customers love these perks. Businesses, not so much as consistent ‘low bailing’ can negatively impact the bottom lines and dilute the brand value.

Strategising the optimal price for a set of nearly identical products (e.g. an autumn boots collection) is relatively easy. But how should you price custom-made products if you are an on-demand shoe manufacturer or strive to offer highly-personalised services to each customer?

Machine learning comes to the fore again. Regression techniques in machine learning allow you to estimate how every possible price configuration will impact your revenue and sales forecasts. After churning available data, the algorithm will choose to display the optimal price based on different variables.

For instance, Airbnb’s dynamic pricing system relies on three components.

The particular advantage here is that unlike an A/B testing approach or segment pricing, regression models allow you to constantly adjust your prices in real-time depending on individual user behaviour. For instance, you can recalculate prices for international customers that will account for FX rate fluctuations; deploy instant price increases for goods that are in high demand in a certain area (e.g. shovels during heavy snowfalls) or pitch special deals to reluctant customers, who keep visiting your sales pages without committing to a sale.

With the abundance of meaningful data that you can source and leverage in your marketing, it no longer makes sense to rely on manual analysis only (aka. the beloved spreadsheets and pie charts). ML-based algorithms can obtain better results than any human analyst, at a fraction of the time and cost. As such systems will only get smarter from now on, it's no longer a question of whether you should be using them, it's a question of when.

Chris Pitt, head of marketing, Vertical Leap