How can you attribute post visit calls?

By Andy Crellin, Head of Data and Decision Sciences at ResponseTap

Call Based Marketing Automation (also known as visitor-level call tracking) allows you to follow a customer’s journey through your website and then offline, onto the phone call. This is incredibly powerful, allowing you to tie together the online and offline experiences of that customer and optimise them both.

This all works phenomenally well when a visitor makes a phone call while they’re looking at a web page (they will call the number on that page that has been uniquely assigned to them and you can join visitor and caller together).

However what happens when they write that number down, leave the website and call at some point in the future? They are then a “post-visit” caller.

 This is a huge challenge for the industry, but there are solutions. Usually these solutions take the form of simplistically assigning the caller to the most recent visit that was shown that number. This is almost certainly going to be wrong. At my company we have developed a statistical approach to matching these callers back to the most likely visit that belongs to them.

 This approach not only gives a more accurate view of which caller matches to which visit (and therefore visitor), but it also fully exposes the expected accuracy of making this match – something which is critical if you want to make business decisions based on this data.

I wrote the following piece aimed at businesses looking to use my company for call based marketing automation. However it contains detail on the general principals involved in accurately attributing post-visit calls – something that anyone in the Marketing Automation industry might find interesting… 

What is a Post-visit Call? 

A post-visit call is a call that happens once a person has recorded the unique tracking phone number given to them on your website (either by writing it down or entering it into their phone) and then calling the number at some point after they have left your website (this tracking number is therefore not assigned to them anymore). This could be minutes, hours, days and even months after they have left your website.

But I Don’t Want Any Post-visit calls:

No Post-visit call can be assigned with certainty to a visitor (as some call tracking companies will claim). However, ResponseTap will work directly with clients to match up to 100% of post-visit calls to the most likely visitor who made that call, using proprietary algorithms developed by our in-house Data Science team. A 100% match is not possible with other solutions.

What is visitor level call tracking?

Each concurrent visitor to your website is shown a different phone number from a number pool, and when a call comes in on that number (while it is being shown) the visitor who made the call can be identified.

What are Number Pools?

Number Pools are sets of phone numbers that are available to be shown on the website for a visitor. Websites with large numbers of concurrent visitors need large pools of numbers to service these visitors. Number pool sizes are kept to a minimum by using number recycling.

What is number recycling?

Because there are a finite number of telephone numbers available (and because there is a cost associated with managing phone numbers) not every visitor can get their own completely unique phone number to call. Instead, a pool of numbers is used, and each of these numbers can be presented to multiple visitors over time, but never at the same time. How quickly these numbers are reused depends on many factors, including the number of concurrent visitors, size of number pool, the time of day, the length of the visits, and so on.

The diagram shows an example of number recycling with a pool of 3 numbers (0800 900001, 0800 900002 and 0800 900003) and up to 3 concurrent visitors.

As time elapses and visitors arrive at and leave the website, each number is recycled and shown many times. By the time we get to time T, we can see that 0800 900001 and 0800 900002 have each been seen by three visitors and 0800 900003 has been seen by two.

In the example we can see that when visitors D and E arrive, we are close to requiring 4 numbers to service all of the concurrent visitors. At ResponseTap, we automatically and dynamically provision additional numbers as and when they are required (so long as it is within the fair use policy).

Identifying the website visitor who made a call

If a visitor is looking at a web page with a phone number on it and they then call that number, we know with near certainty (much better than 99%) which visitor made the call because we know that the number they called was only visible to that specific visitor at that specific time. As on average only 2 in every 100 website visits results in a phone call, we can be highly confident that if a number is currently visible to a visitor and a call comes in to that number, then the call is being made by that visitor. This is how all call tracking solutions work – there is a small degree of potential error when linking phone calls to online visitors, but this error is usually substantially lower than 1% for calls made while online.

But what about when a visitor makes a call after their visit has ended?

“Post-Visit Calls” – calls made by visitors after they have completed their browsing session and closed their browser – are notoriously difficult to match back to the correct visitor. This is because there is absolutely no empirical connection between the caller and their visit any more. All attempts to match a post-visit call to a visitor have inbuilt error – and usually this error is very large and misunderstood. It is a misdirection to say that any post-visit call can be assigned to a visitor without also explaining the level of error involved in this process. Many call tracking providers who say they can reduce the number of calls classified as “Post-Visit” not only ignore this error, but they also fail to make their clients aware that it exists – essentially pulling the wool over their eyes in the hope that they don’t realise and they usually don’t!

However, there are techniques that we can use at ResponseTap to match a post-visit call to the most likely visitor who generated that call – and we make sure that our clients are fully aware of the pros and cons of this approach. You can only ever say that a post-visit call is likely to have been made by a particular visitor. Only by being fully aware of this and understanding the pitfalls can you use post-visit call attribution to make powerful business decisions.

Looking back at the diagram showing number recycling, we can see that if a call comes in to 0800 900003 at time T, then no visitor is actually being shown that number at that time. If we want to attribute the call to a visitor we have a choice to make – should it be visitor C or E? Both have seen the number on the website and both may have written it down for future use. Some call tracking companies will simply assign the call to the last visitor to see the number (by pretending that the browsing sessions last longer than they actually do). This is almost certainly wrong in most cases.

We typically see that post-visit calls account for approximately 20% of all calls (this usually ranges between 15-25% depending on the industry), meaning that 80% of your data is rich and highly accurate (much greater than 99% accuracy). The remaining 20% of calls are then reported as post-visit and do not have visitors associated with them. One way that other call tracking companies have claimed to reduce post-visit calls is by increasing the length of time the tracking number is assigned to a visitor (e.g. for 30 minutes after their visit has ended). In other words they are arbitrarily extending the call to visit attribution window by 30 minutes.

However, consider the common situation where post-visit callers are delaying their call because they are in the office, or they want privacy, or they need more information from other sources. There are a multitude of common reasons to delay the call by several hours. Let’s consider a best case scenario where all post-visit callers are expected to call within 48 hours of their visit (our data tells us that this period is actually much, much longer) and the most common period between visiting the website and making a post-visit call is 4 hours (browsing while in the office, then calling when they get home). In this scenario only 0.7% of post-visit callers are expected to call within 30 minutes of their session ending (contact us if you want to talk about the maths!). This means that if you reserve a number for a visitor for an extra 30 minutes after their visit ends and a post-visit call comes in on that number, 99.3% of the time it will be attributed to the WRONG visitor. The visitor who is more likely to be calling is one who saw the number on the website earlier in the day.

So, if this tactic resulted in a “5% reduction in post-visit calls” you might assume that 85% of your data was rich and highly accurate. However, this is not the case – that extra 5% would come with an error rate of 99.3%! Worse than that, you are now poisoning your once accurate attribution of in-session calls with these inaccurate ones, dropping your overall accuracy of your in-session data from well over 99% to around 95%.A comparison chart showing no extended attribution window, with a 30 minute extended attribution window

You have gone from being able to use 100% of your data to drive marketing decisions with very high confidence (80% of them where you know exactly what their journey has been, 20% where you know with certainty that you don’t know their journey), to a situation where you no longer have that confidence in your marketing decisions. What’s more concerning is if you are using this data to drive better customer engagement through CRM and call centre integrations you will be arming your sales teams with inaccurate data 5 out of every 100 calls, which will reduce the confidence your team has in any of the data provided.

At ResponseTap we do things differently. As usual, we ensure that you continue to understand the journey of in-session callers with substantially better than 99% accuracy. Then, in addition to this, we can not only identify the most likely visitor to attribute a post-visit call to, but we can also tell you the probability that this attribution will be correct. We will not simply say that “we can reduce post-visit calls by x%”, because this is disingenuous. What we will say is that we can attribute 100% of your post-visit calls to the most likely visitor to have made them, with a clearly understood, and mathematically rigorous level of accuracy. Most importantly we will identify which individual calls have a higher degree of uncertainty so you can decide the level of potential error you are willing to accept when making important decisions based on this small subset of calls.


On the face of it extending the window that phone numbers are assigned to visitors (in the example about it was 30 minutes) to reduce the number of post-visit calls seems like a good solution, but in reality almost all of the additional data is actually inaccurate. In fact in the example above almost the entire 5% of the additionally reported in-session data provided by this option is proven to be inaccurate. Therefore adopting such an approach only serves to pollute your accurate data by almost the same level of magnitude as the reduction in post-visit calls you are trying to achieve.