Win-rates: what can we learn about sales forecasting from epidemiologists
Updated: May 12, 2020
Sales win-rates are a simple concept but are rarely computed properly. Turns out that we can learn a lot about this from COVID-19 pandemic reporting.
Coronavirus has turned us all into epidemiologists. One big question we ask is “what is the risk of death for a COVID-19 diagnosed patient?”
And here’s the rub (source: The Lancet https://www.thelancet.com/journals/laninf/article/PIIS1473-3099(20)30243-7/fulltext)
“During a growing epidemic, the final clinical outcome of most of the reported cases is typically unknown. Simply dividing the cumulative reported number of deaths by the cumulative number of reported cases will therefore underestimate the true case fatality ratio early in an epidemic.”
We’ll come back to that shortly. But first let’s acknowledge that there are other factors as well. E.g., there is a bias towards detecting clinically severe cases, so the denominator is probably smaller than true infections or true cases. But there are also deaths that were not diagnosed as COVID-19, so the numerator is also depressed. In due course, we will have better estimates.
We were struck by this diagram of the spectrum of COVID-19 cases. Note that the fatality rates we are discussing are for the symptomatic cases because we don’t know (yet) how many cases are asymptomatic. If we were to measure the rate for asymptomatic cases we would get a lower figure. So this brings us to the first important point:
The death rate depends on where you start counting cases from. Here we have three stages: infections, symptomatic cases, severe cases. For this reason, epidemiologists distinguish between the case fatality rate (for those who present with symptoms) and the infection fatality rate (for all people infected).
What can we learn from this? Turn that figure upside down and it looks a lot like a sales funnel. We could relabel those spectra with more common sales-stage names.
Asymptomatic (mild) cases ==> First call. Symptomatic cases ==> Proposal.
Severe cases ==> Negotiate. Deaths ==> Won.
To be clear, we are not saying that a death is in any way equivalent to a sales win. It's just that from a modeling standpoint, they are analogous. They are both called "hazards" in stat-speak. Admittedly the name is more apropos in epidemiology.
Epidemiologists are (usually) interested in modeling death—a negative event. But in sales, we are interested in modeling successful sales—a positive event. Epidemiologists start out with 100% of the population surviving and go down from there. In sales we start out at 0% of the deals won and go up from there. Other than that, the methods are very similar. So we can rewrite our epidemiology lesson for sales forecasting:
Win-rate is function of sales stage.
Put another way, opportunities at the top of the funnel have a lower expected win-rate than those lower down. That's why we call it a funnel.
Now let’s go back to the earlier point from the Lancet article:
“the final clinical outcome of most of the reported cases is typically unknown.”
The same can be said of B2B sales. The final outcome of a sales campaign is not known until… Well, how long do we have to wait? This of course depends on your sales cycle; and how patient you are. If you are measuring the B2C response rate to a banner ad (clicks) then maybe a few seconds is all you need. But for most B2B sales campaigns we only know when know when we win (which is typically measured in months).
Unless it was a competitive shootout and you lost, if we don’t win a deal when we expected to, salespeople tend to hold on to opportunities and are slow to call the opportunity as lost. It’s justifiable. Customers’ priorities shift. Reorgs happen. Prospects may go through several restarts of a buying process before actually making a decision. And yes, sometimes it’s just a slow fade. As a result, in B2B sales, its common to find that open opportunities (ones for which we don’t know the outcome) are 90% of your sales funnel.
Picking up again from the Lancet article and applying that to sales win-rates:
(Bold) below means delete. Red means insert.
“Simply dividing the cumulative reported number of (deaths) wins by the cumulative number of (reported cases) opportunities will therefore underestimate the true (case fatality ratio) win-rate (early in an epidemic).”
Note that we deleted that last phrase “early in an epidemic.” If we have a steady state (no growth) then we don’t care about time because all the changes will have settled out. But in sales, we are constantly striving to increase the number of opportunities (i.e., cases). We are forever living in the growth phase and we can never rely on this simple computation. This brings us to our second point:
Win-rate is a function of time.
Put another way, it makes no sense to say that you win some percent of your opportunities without also saying how long it takes.
A corollary to this issue is that we can’t treat new opportunities (newly diagnosed patients if you will) as equivalent to older ones (patients who were diagnosed long ago). There are three ways that people deal with this issue. (1) Ignore time and include all opportunities. This will understate the true win-rate. (2) Ignore time and exclude all open (unresolved) opportunities. This is the most common approach and it will overstate the true win-rate. These shortcomings lead us to our next key point:
Win-rates need to reflect all opportunity dispositions.
The third approach to dealing with this aging of opportunities is to select a cohort of opportunities from the past, and wait "long enough" for them all to resolve. This can work nicely if you have short sales cycles and a lot of data. But for most of us, it suffers from data limitations and model deterioration. You don’t want the aging of opps in the cohort to come into play before you start measuring, so you need to narrow the window for your cohort selection. Typically, there is just not enough data available to do this and still have a meaningful sample.
When we say “wait long enough for them to resolve”, we mean typically at least two to three sales cycles, more if there are still some open opps. Let’s say you have a nine-month sales process. This means that we would be computing win rates today based on opportunities that were started over a year and a half ago. Do you really think that what happened that long ago is indicative of what will happen in the future?
For these reasons, all three approaches fall short. Which brings us to our next key point:
Win-rates need to reflect the most recent data available.
Epidemiologists have developed ways to deal with all these problems and Funnelcast uses those same techniques to model sales win-rates as a function of time.
There is one more important point about sales campaigns, and we have kind of already covered it when we discussed how salespeople tend to hang on to opportunities that are not going to close. The best way to think about this is to relate sales opportunities to product failures. In product reliability studies, they talk about the “bathtub” curve.
Early in a sales process, opportunities may be binned out quickly. Later in a sales process—if the opportunity has not advanced—the opportunity may likely “wear out” or exit the stage. Actually, this analysis is a simplification because for product failures there is only one possible state that the part can advance to: broken. For sales, there are three states that the opportunity may advance to: open (another stage), won, lost. And each possible outcome has its own curve. Which brings us to point five:
Summarizing the five principles.
1. Win-rate is function of sales stage.
2. Win-rate is as a function of time.
3. Win-rates need to reflect all opportunity dispositions.
4. Win-rates need to reflect the most recent data available.
5. Staleness matters.
Win-rates must be based on all of these principles. Funnelcast builds win-rates that are a function of time, paramterized by sales stage, staleness in stage, and optionally any user defined business segment. The win-rates are based on all the data availalble, and reflect all opportunity dispositions.
Here’s an example: a win-rate chart from Funnelcast showing a simple four-stage sales process.
Each line represents a sales-stage, win-rates are a function of time, they are computed based on all data up to the reporting date, and staleness in stage is an input parameter (not shown but annotated on the chart title). Let’s examine the blue line for the “Contracts” stage. The vertical line tells us (empirically) that 90 days after entering the Contracts stage, 92% of opportunities in the Contracts stage have closed. At 30 days, that probability is considerably lower (68% in this example).
So, if we apply that to current opportunities, we can say that if you start the quarter with an opportunity entering the Contracts stage, you have a pretty good shot (92%) of closing it before the quarter end. For deals in the Proposal stage that figure is 62%. Qualified: 49%. Targeting: 9%.
Let’s take a look at how things are different for stale opportunities. Here we are looking at the vertical bar after 90 days. So, this would represent the probability of successfully closing a deal before the quarter end for opportunities that were 45 days stale in their stages at the start of the quarter. The example shows that staleness has caused the probability of closing a Contracts-stage opportunity to drop from 92% to 83% and for a Proposal stage opportunity to drop from 62% to 48%. We can see similar substantial drops in the Qualified and Targeting stages. Staleness matters.
So what does all this tell us about sales planning? Well if you compute your win-rates properly (the way epidemiologists would compute mortality rates) and you know the stages and ages of opportunities in your funnel, you can use these to forecast what you will sell. That's part of what Funnelcast does for you. The question of course is whether the past is indicative of the future when we experience the economic shock of a global pandemic? Of course not. But the forecast is a departure point—what would have been. We have managers for a reason. Until things settle back to normal, we would exclude every current open opportunity until it is re-qualified by your sales team; and eliminate any opportunities from the forecast until they advance to a subsequent sales stage.
Of course, all this depends on having enough data. Want to find out if you have what it takes? Check out the Funnelcast minimum data-readiness guidlines. Want to learn more or see Funnelcast work on your data? Contact us about our beta program.