Many businesses recognise the value of using customer decisioning models. Most also recognise that the fresher the model, the better the decision. But how important is it to keep the models fresh, and is it worth investing in automation?

Many clients struggle with building that business case. Below are some calculations that can help justify that decision (and they can be tailored to your business).

Placing a value on model ‘freshness’

How much additional revenue can be achieved from having models that are significantly fresher? For example:

• A fresh model is likely to give a lift of 1.6 (where lift is the ratio of improvement over a random sample)
• Let’s say the model degrades over a year, and the lift drops to 1.52 (which would not be unusual).
• This reduction in predictiveness would result in 5 percent fewer responses being captured than if the model was kept up to date.

This is a tangible benefit of keeping models up to date. You can calculate an estimate on the core business KPIs of revenue or cost. For example:

• Revenue: Essentially for the same marketing budget, you will achieve 5 percent more responses. So, multiplying the number of decisions by the average response rate, then by the average revenue per response and finally by the 5 percent difference in predictiveness will give a ball park estimate.
• Cost: To achieve the same number of responders, and hence revenue, you will need to spend more marketing budget – probably more than 5 percent, too. This equates to a 5 percent delta to your direct marketing budget as a ballpark estimate.

So, let’s say a business has:

Five million customers and will contact 40 percent of them six times each year. The response rate of the targeted group is 1 percent. The revenue is $50 per response and the cost of contact is $0.25.

The potential revenue from fresher models is:

5 million customers * 40% * 6 contacts/year * 5% model improvement *$50 * 1% average response = $300k for one model.

Or, the additional cost that would be needed is:

5 million customers * 40% * 6 contacts * $0.25 = $300,000

Realistically, a model that could generate as much value as this is unlikely to be left to degrade over the course of a year. However, given that automation of the modelling process could reduce time to market from more than three months to as little as a month, then in essence, the business could save two months’ worth of degradation – which in the above ballpark calculation is $50,000 for just one model. Clearly this adds up as more models are considered.

In case you are wondering how much value is being generated from the model itself (rather than the just the incremental value of keeping the model fresh), then back solving from the 1-percent average response and the 1.6 lift, you would have a base response of 0.6 percent. This means the incremental revenue value of the model itself is:

5 million customers * 40% * 6 contacts/year * (1- 0.6)% response lift from model *$50 = $5.6 million

Benefits beyond response rates

Results vary from business to business and from client to client, depending on how much of the decisioning is driven by models, and how fast the current update process is.

There are also cost considerations in terms of automating the model update processes. However, there are also the significant tangible organizational benefits such as:

• Improved governance.
• More robust processes.
• Greater knowledge sharing.
• Less reliance on scarce and expensive resources.
• More time spent on more innovative analytics rather than the grind of updating models.

Now of course, the business case here is deliberately simple, but it is a good tool to assess other business cases – and perhaps compare and prioritise.

For example, when a new data source is added, such as real-time data or some external data, how do you assess its value? I would argue that there is no value in this data unless it is used to improve decisions. The most tangible example is decisions based on models. If the lift in performance can be estimated, then a business case like the one above can be made in an analogous way. Many SAS customers use SAS Model Manager to help automate the process for updating models, and you can find out more here.