Build and Use an E-mail Response Curve Model
A response curve model can predict how an e-mail campaign will perform within hours after the send. Four steps to building and using one.
A response curve model can predict how an e-mail campaign will perform within hours after the send. Four steps to building and using one.
One benefit of e-mail is a quick payoff. With direct mail, it can be four to six weeks before you get all the responses and know whether a campaign is a success or failure. With e-mail, it’s more like two to four weeks. But if you’re impatient (like I am!), there’s a way to project, just hours after a send, how an e-mail campaign will perform in the end. It’s called a response curve model.
A response curve model uses past results to predict future performance. It’s inexpensive and easy to build your own response curve model. All it takes is some data from past campaigns. Here’s how to get started.
Determine Key Success Metrics
Your key success metric is the piece of data that directly corresponds to the success or failure of your campaign. Open and click-through rates are rarely, but sometimes, key success metrics. Perhaps the measure of your success is the number of opt-ins or registrations, in which case you can use that. Data such as sales, revenue, or leads generated are the best success metrics, as they directly relate to your business’ bottom line. If you’re tracking one of these in real time, use it.
Pull Data From Past Campaigns
Once you’ve decided on the success metric, start pulling data. If you have data on past campaigns by time (some e-mail service providers (ESPs) provide this), you can use those. If not, you can begin collecting them when you next send and build the model based on the performance of your next couple campaigns.
You want to pull data from specific periods after the send. You can use as many or as few periods as you wish. If your metric is leads, the data might look like this:
E-mail Campaign 1 | E-mail Campaign 2 | E-mail Campaign 3 | Average | ||||
Hours Since Send | Total Leads to Date | Percent to Final (%) | Total Leads to Date | Percent to Final (%) | Total Leads to Date | Percent to Final (%) | Percent to Final (%) |
2 | 25 | 5.6 | 10 | 3.8 | 5 | 5.0 | 4.8 |
4 | 56 | 12.5 | 24 | 9.1 | 14 | 13.9 | 11.8 |
6 | 82 | 18.3 | 39 | 14.7 | 21 | 20.8 | 17.9 |
8 | 106 | 23.6 | 68 | 25.7 | 23 | 22.8 | 24.0 |
10 | 118 | 26.3 | 73 | 27.5 | 26 | 25.7 | 26.5 |
12 | 124 | 27.6 | 91 | 34.3 | 32 | 31.7 | 31.2 |
24 | 180 | 40.1 | 108 | 40.8 | 41 | 40.6 | 40.5 |
36 | 275 | 61.2 | 159 | 60.0 | 59 | 58.4 | 59.9 |
48 | 340 | 75.7 | 209 | 78.9 | 75 | 74.3 | 76.3 |
60 | 375 | 83.5 | 230 | 86.8 | 86 | 85.1 | 85.2 |
72 | 401 | 89.3 | 240 | 90.6 | 90 | 89.1 | 89.7 |
Final (30 days) | 449 | 100.0 | 265 | 100.0 | 101 | 100.0 | 100.0 |
Develop a Model
In addition to pulling the leads generated by hour, I have a final number of leads (after 30 days) from each campaign. This allows me to calculate what percent of the final leads were in hand at each interval:
Averaging the “percent to final” number across all campaigns (the column at the far right in the above table) provides the figures I can use in my model to predict where future sends will end up.
Use Your Model to Project Results
To use your model, just collect the same data at the same intervals after the next send:
Average | Current Send (12 Hours) | ||
Hours Since Send | Percent to Final (%) | Total Leads to Date | Projected Final Leads |
2 | 4.8 | 9 | 188 |
4 | 11.8 | 24 | 203 |
6 | 17.9 | 35 | 195 |
8 | 24.0 | 48 | 200 |
10 | 26.5 | 52 | 196 |
12 | 31.2 | 61 | 195 |
24 | 40.5 | ||
36 | 59.9 | ||
48 | 76.3 | ||
60 | 85.2 | ||
72 | 89.7 | ||
Final (30 days) | 100.0 |
Use this data, along with the average “percent to final” figures in your model, to project the final results of a campaign. The calculation is simple:
In the example above, the calculation after 2 hours would be:
Using the 12 hour, figures you’ll get:
The more hours of data you have in the “total leads to date” column, the more accurate the projection will be.
The timing of sends also affects the model. An e-mail mailed midday is likely to see more response in the first few hours than one mailed in the middle of the night. If there are dramatic (more than 3 hours) fluctuations in send times, try to gather a full 24 hours’ worth of data before projecting to get a more accurate figure:
Average | Current Send (24 Hours) | ||
Hours Since Send | Percent to Final (%) | Total Leads to Date | Projected Final Leads |
2 | 4.8 | 9 | 188 |
4 | 11.8 | 24 | 203 |
6 | 17.9 | 35 | 195 |
8 | 24.0 | 48 | 200 |
10 | 26.5 | 52 | 196 |
12 | 31.2 | 61 | 195 |
24 | 40.5 | 79 | 195 |
36 | 59.9 | ||
48 | 76.3 | ||
60 | 85.2 | ||
72 | 89.7 | ||
Final (30 days) | 100.0 |
In our example, the projected final leads figure is 195 after 24 hours. As you get further from the send, you should see the projected final leads figures converge like this.
In most cases, you’ll have 80 to 90 percent of campaign response after 72 hours. And you’ll have a good idea of where the campaign will end up:
Average | Current Send (24 Hours) | ||
Hours Since Send | Percent to Final (%) | Total Leads to Date | Projected Final Leads |
2 | 4.8 | 9 | 188 |
4 | 11.8 | 24 | 203 |
6 | 17.9 | 35 | 195 |
8 | 24.0 | 48 | 200 |
10 | 26.5 | 52 | 196 |
12 | 31.2 | 61 | 195 |
24 | 40.5 | 79 | 195 |
36 | 59.9 | 118 | 197 |
48 | 76.3 | 151 | 198 |
60 | 85.2 | 172 | 202 |
72 | 89.7 | 179 | 200 |
Final (30 days) | 100.0 |
The numbers in this model are just for example, they’re not from an actual campaign. Always develop your own response curve model based on your own campaigns. A model based on someone else’s data won’t necessarily provide accurate results for your own sends.
Continually check the accuracy of your model against actual results. In the above example, you’d want to see how many leads you actually end up with after 30 days. If it’s around 200, your model is working well. But if it’s only 180, the model is overstating projections. If you end up with 230 leads, the model understates results.
If the model isn’t quite accurate, adding in data from additional sends should help get it on track. Go back to step one and add in the information from your latest send. Do the average percent to final calculation again and adjust your model accordingly.
Develop your own response curve model, then use it to project future campaigns’ performance. Let me know how it goes!
Until next time,
Jeanne
Want more e-mail marketing information? ClickZ E-Mail Reference is an archive of all our e-mail columns, organized by topic.