Month: November 2018

The End of the Beginning

I have followed Benedict Evans for many years and he has an amazing ability to put context around large datasets.

There are some great insights into a number of markets that we currently invest in such as Grocery, TV and Machine Learning, so take 23 minutes out and watch this

Database Marketing 2 of 2: Tactics

In my previous post we covered the database strategy. This time I wanted to dig into tactics, mainly looking at email which is currently the core database marketing channel for most e-commerce businesses (my strong suspicion is that won’t be true soon).

Anyway, given the strategic objective of maximising the value of the database, let’s look at some tactical considerations:

1. Are we simply bringing forward sales that would have happened anyway?
2. Are we offering a discount to a customer who would pay full price?
3. Are we mailing at the right frequency?
4. Is the content right?
5. What product personalisation should we be using?
6. How strong an offer do we need?
7. Who could be a Member-Get-Member advocate?
8. Who is at risk of leaving us?

On reflection, it becomes clear that there just isn’t one answer to these questions. They are fundamentally questions about segmentation.

The first question should really be “Who is going to buy anyway and who needs to be prompted?”. The other questions could be re-written in a similar way.

These questions are inherently much more complex than our strategic question which could be answered with one simple formula to measure the value of the database. There’s a lot of crossover between the questions: some people probably do need a strong offer at a high frequency for example.

These sort of complex, multivariable questions are very well suited to machine learning where we can just put a bunch of data into an API and action what comes out.

This is what we are currently working on at our new startup Machine Labs. If you are interested in joining us as a beta customer then please get in touch.

Database Marketing 1 of 2: Strategy

At their heart, nearly all e-commerce businesses are database businesses. Oddly enough it’s a side of the business that tends to be ignored. There is a virtual obsession with customer recruitment and conversion rates but almost no discussion about database marketing.

This is ironic because the database is where all of the profit comes from. In an efficient market, which the e-commerce world tends to be, it will be impossible to recruit profitably. So logically profit has to come from the database.

Currently most e-commerce businesses market to their database mainly be email so I will focus on that for now.

So the strategy should be to maximise the value of the database. I think that is quite an uncontroversial statement. I would however suggest that the vast majority of marketers don’t do this.

The first thing is to think of what the value of the database currently is. In investment there’s a very simple Net Present Value (NPV) calculation:

NPV Formula

This is just a sum of the cash flows from the database with the future cash flows discounted back to its present value. NPV is very widely used for valuing assets.

The tricky bit though is working out what the future cash flows from the database are likely to be. I would divide this into two parts:

1. What is the value of leads?
2. What is the value of customers?

Valuing leads should be based on i) the likelihood of them turning into customers and ii) when that is likely to happen. This is generally a fairly simply calculation providing the database has about 6 months of history. (Leads tend to go off faster than fish.) The more metadata you have the more accurate this calculation is likely to be.

Valuing customers is about estimating lifetime value. There are several ways of doing this. One method would be to add up historical lifetime values and use survival analytics (we’re Python developers at Move Fresh so we like Python Life Lines for this) to estimate future lifetime values of current customers.

Another approach would simply be to dump the dataset into a machine learning API and see what comes back, although you would need a good few years for this to work. This has the advantage that many more variables could be taken into account.

Once you have your estimate of cash flows you can then calculate the NPV of your database at will. This gives you a very different way of thinking about your business.

For recruitment campaigns, instead of thinking of Cost per Customer Acquired you can think of cost of the recruitment campaign v. increase in NPV of the database.

It also helps you understand how to email the database. The value of sales of an email campaign must be greater than the decrease in NPV of the database as a result of sending the campaign out. Thus you should be well on the way to maximising the value of the database without killing the golden goose by over mailing.

Finally, it will also help you to understand how much capital is sensible to allocate to marketing. If a marketing campaign results in £3 of NPV for every £1 of spend then that would be very powerful.

In my next post I’ll cover tactics.

© 2019 Move Fresh

Theme by Anders NorenUp ↑