Part 3: Ai Marketer – Profusion’s response to AutoML

Hi there, I’m back with part 3 of Ai Marketer (AiM) – our response to AutoML – predictive segmentation models designed to super charge your marketing strategy.   

AiM integrates directly into your preferred environment to enhance segmentation and, ultimately, increase retention and sales. 

In case you missed out, you can recap on part 1 and part 2 here.

AiM will enhance what you learn about your customers and enable you to make best-next-action predictions about whom to target, when to target them. This edition of the series looks specifically at three core ‘sales generator’ models.  

First up is Churn Prediction –  

  1. Customers ‘At Risk’ of Churn 

Predict customers at risk of churn and/or churned, test and then automate ‘win back’ campaigns

It costs 5 x more to attract new customers than to retain the ones you have, while existing customers spend 31% more when compared to new customers.  Furthermore, you may well have increased your database during Covid-19 times so it’s crucially important we keep them coming back to you on a regular basis as we slowly come out of the crisis. 

So, we need to make sure you know when they might stop shopping with you, catch them before they stop shopping, and win them back. 

Call this churn prevention, an AI prediction layer in addition to the lapsed efforts you currently run. The power of prediction helps you avoid churn before it happens. 

A picture containing shirt, player

Description automatically generated

To do this, we need to understand the natural rate of return for each individual customer rather than using a single ‘at risk’ and ‘churn’ duration for all customers. We use historic data to categorise customers to understand each person’s natural rate of return and then determine whether they are currently ‘neutral’, ‘active’, ‘at risk’ or ‘churned’.  

Each customer has their own unique ‘customer life-cycle curve’ and therefore their own unique timeframe to churn. Based on the personalised life-cycle curve of each customer we obtain variables to help us understand similarities between customers who have a similar purchase lifecycle, which then helps us understand how each of these groups are likely to behave in the future.  

From here, it’s time to build a win-back campaign including a range of targeted messages and offers to drive the customers into store, or online, and an appropriate control group to measure impact. A testing approach is key, enabling you to then automate the optimal win back approach. 

It’s worth noting that this model identifies customers in their ‘active’ period, giving you the opportunity to test offers to increase AOV. 

Here’s out it worked for TradePoint (the trader arm of B&Q) 

Tradepoint increased revenue by £13m after 6 months using this model. Importantly, 53% of churned customers were shifted back to their normal buying pattern and were still operating in it 6 months later with no further nudge activity. 

  1. Customer Lifetime Value (CLV) 

It goes without saying that you want to grow your loyal customer base 

To do this, it is critical to recognise what the signals are for someone who is going to be a loyal customer, and ensure you treat them as such from day one.  

The model would review your existing customer segments to understand your tiers and the factors which tip them to ‘loyal’ status. The predictive element will help you identify the future of new ‘best’ customers from day one and understand the potential from your existing customers to tip them over into the ‘loyal’ status.  

A picture containing table, room, food

Description automatically generated

Early identification of these individuals means you can enhance their customer journeys, use offers and incentives accordingly and ensure a loyalty treatment happens when they contact you, either online, instore or via the call centre.  

About the model.  Traditional CLV estimation relies on the use of Recency Frequency Monetary (RFM)  segmentation, survival analysis, and ‘buy till you die’ models. More sophisticated methods require the use of heavy computational deep learning algorithms.  

OUR model, implemented into AiM, takes your customer transactional data as input for an advanced Neural network and determines accurately the value of each customer in the next 12 months. We achieve this by training our model to minimise the procedure for estimating unobserved quantities. Without going into the super technical details, the result confirms our methodology outperforms in predicting power and implementation time of traditional techniques mentioned above. 

A popular tiling retailer has benefited from this model: 

We are currently doing this for a popular tiling retailer, helping them introduce a Gold Trader loyalty scheme to retain and grow revenue from their high-value customers, as well as predicting newer/lower-spending customers who show signs of becoming ‘best’ customers in the future. The test campaign (which ran for only 1-week pre Covid-19) generated £32,000 in incremental sales (over £500,000 in total sales). We’re expecting the programme to drive an extra £1,000,000 in incremental sales as this now rolls out. 

  1. Propensity to Buy 

Predicting when they have the highest propensity to buy: You want to know when your customers have the highest chance of buying, and AiM can predict that.  

A clock that is on a pole

Description automatically generated

AiM ingests as much as you know about your customers (transactions, demographics, profile, trade, voucher redemptions, browsing history etc.) enabling the Propensity to Buy model to determine their likely buying window, meaning you can nudge them with the right message or offer, at the right moment, and not hassle them if there’s no chance they will buy.  

This model naturally moves you towards 121 messaging and a chance to test against blanket BAU campaigns which might currently consist of a bit of segmentation and first name personalisation.  

Here’s how a luxury car brand benefits from this model: 

The propensity model for one of the most luxurious cars on the planet was tested in five dealerships and made £10m in year one. It is now integrated into their Salesforce Einstein system and rolling out globally across dealerships. 

I hope this gives you a taste of the power of AI (AiM) as a fundamental part of your marketing segmentation.    

Next week we will look at a further 4 models that support your email health and sender reputation, which only naturally increases reach and revenue.  

Any questions, please let me know:  emmac@profusion.com 

[mr_rating_form]
Comments
(0)

View more recent postings

+ Add comment

Leave a Reply

Your email address will not be published. Required fields are marked *

Loading...