How can we predict that a customer will buy soon (or not)?


Datainsightout helps crunching your numbers... discover how

Datainsightout is for you... for a fair price!

61 days full access for only $75

Discover our product

Churn prediction and likelihood to purchase are key to success to grow a business. But how to predict when a customer will purchase for the next time at your shop? How can one foresee which customer will buy in the next few days or not? This is part of our ultimate guide regarding the customer lifecycle.

Before reading this post, why don't you have a look at our product page - we have a module allowing our customers to predict likelihood to purchase.

Why predicting future sales is important for your business

Engaged customers drive value

Acquiring and activating a customer costs a lot of money. You need to target them, invest a lot of money on online and offline advertising, build your brand, send consistent communication messages accross multiple channels and keep repeating these actions, before finally a customer decides to go ahead and make his first purchase at your shop. Before this first sales to happen, you had to invest quite a lot of money. Keeping alive a customer and engaging him is a key component to actually grow your business and actually capitalize on past marketing efforts.

Repeat purchases will increase your ROI. When the customer started to buy at your shop, it's a first win, since the customer starts to engage with you and to trust the ability of your company to fullfill their needs. Subsequent purchases might be mechanically cheaper for you as you already have the customer's email address and contact details to reach out directly. You will spare search engine acquisition costs, retargeting money and avoid doing cold calls.

Put into financial equation, the breakeven point to get amortize the cost of acquisition of a customer will come after repeated purchases:

Number of purchases needed to break-even Cost of Acquisition of one customer  = Total Cost of Acquisition of the customer / Gross Margin per Customer per purchase


Datainsightout helps small and medium companies crunching their data. What about yours?

Discover our product

Easy to understand: the higher the gross margin per customer, the faster you'll break even your customer's cost of acquisition - which means that your need to sell your high margin products or cross-sell your products to increase the size basket. Otherwise,  you would want to maximize repeat purchase.

Predict customer churn and boost your ROI

Boost your ROI (return on investment) by predicting how likely a customer is to purchase. If you understand which customer will be likely to purchase over a certain timespan, it will allow you to push the right offer to the right customer with the right incentive and to re-engage with a customer who was about to forget about you and your company. By understanding purchasing patterns; you basically keep your customers alive. You might not re-engage them at the end of the day, as not all marketing strategies are successful, but let them know you still exist and probably convince them to make another purchase.

Marketing spendings are made more wisely and triggers cost savings. Predicting the probability of each customer to purchase next will allow to segment your customer database and address your customers with the most adapted marketing set of tools. Price reduction, couponing are indeed not ideal if the customer was already very likely to purchase at your shop. Segmenting your customers by purchase probability can prevent sales cannibalization. Marketing actions can also be undertaken to nurture other segment of customers who might need an incentive to make a repeat purchase.

Purchase and churn predictions are very actionable. As you understand, being able to predict the likelihood of a customer to purchase has immediate impact on your activity. This is not a pure "good to know" metric, but a metric you could use to help you drive your activity and guide your marketing spendings. You will get a flag in front on each customer with their likelihood to do a purchase and will be able to target them with sales and marketing actions to push your sales up.

How to calculate for a customer its probability to purchase over the next 5, 30 or 90 days

Find the pattern!

Purchase prediction is related but different to overall customer segmentation. While RFM segmentation might good when it comes to long term customer database management by sending regular newsletters, communications and engage differently depending on the customer' overall level of engagement, sometimes you might need to have a more precise measurement of the likelihood for a customer to purchase. The main difference is that purchase prediction can trigger short term actions while customer segmentation is more long term.

You would create specific marketing actions to push a customer to do re-buy at your shop: this is very actionable and can lead to immediate results. Typically, for tactical sales boost actions. However a marketing segmentation is generally not only based on a RFM segmentation, but has also demographics components, information about the activation channel of the customer, information about geographics and place of residence and so on. 

How can one predict the likelihood to purchase?

Calculating sales prediction: a simple concept. Assumption running behind sales prediction is that a customer who purchased once per month, every month since 12 months, and also purchased last month, will be very likely to purchase next month at your shop. On the contrary, a customer who purchased every week during Jan to March but hasn't purchased since 6 months, will be less likely to purchase over the next 30 days, by comparison. As you can see, theory behind customer sales prediction is fairly easy to understand.

How sales purchase prediction works ?

So let's say you need to boost your sales to meet end's month financial expectation, accross your most engaged segments.You might want to focus on a segment of customer to push a promotion. Not the whole segment, just a few of them to cherry pick them, for instance if you run an ecommerce website and want to target customers who only purchased with coupons. Here comes our module using the Lifetime calculation algorithm. You can know more about the model used here. We use the lifetime statistical probability model to predict for each customer, the probability to purchase at your shop over the next 5, 30 or 90 days window timeframe. 

Under the hood, the Lifetime algorithm used is leveraging the BG/NBD model, which follows a Poisson law distribution pattern. 

3 steps to predict customer churn

First step, the model will transform transaction historical data for each customer into an "Recency Frequency-like" scoring table. This will make it possible to use understand the consumption patterns of your activity. Transaction date, recurrence over time and consistency will be analyzed at a customer level, but also relatively, comparing the customers between them to rank the customers. This sales prediction requires as you can imagine to have a clean dataset... while triggering many questions: Are you going to factor in the order returns? or only consider only net sales (no return). Do you want to run this analysis for all your customers at once, or break your customer database into multiple subsets to then run multiple batches; E.g. a subset for each market you sell your products, assuming that all the markets where you operate don't have the same level of engagement of your customers.

Second step, the model will, using statistics, score your customers and assess their likelihood to purchase over a certain timeframe - something you can choose as an input while uploading your data. This timeframe selection is a key component of the Poisson law. How to to choose the right timespan? It really depends purely on your activity. If your business is about delivering water and bevarages within 1h30 to local customers, you would bet that your customers will pass an order at your shop every week. Same if you're a supermarkt on line. However, if you sell luxury shoes online, 2 or 3 pair of shoes at year would be considered as a great achievement. Same if you sell pieces of furniture online - you might not sell a sofa twice a year. Lifecycle analysis is key to understand which parameter will be suited in the sales prediction analysis.

Third and last step, you will download a CSV file which has for each customer their likelihood to purchase score, computed by the model and, why not, upload the customer IDs into your marketing tool to trigger a campaign.  It's just a scoring model, hence, has usual, should be checked with your business accumen knowledge, but it's a guide you could use to perform further sales and marketing activities.

We encourage you to read more about our product and how our solution can help you dealing with sales prediction. You can have a look at our product page.

This post was rated: 0 / 5 by 0 readers
You rated: Not rated yet... but please be kind!

Datainsightout helps crunching your numbers... discover how

Datainsightout is for you

61 days full access for only $75

Discover our product