top of page
Search
Writer's pictureBill Kantor

The Magic of Parsimonious Models

Updated: Aug 4

“Garbage in, garbage out. How can you make great forecasts with bad data?”


You can't. But you can make forecasts less susceptible to the bad data problem.


Funnelcast models extract as much insight as possible from as little data as possible. Data that is automatically recorded by CRM systems. This minimizes exposure to messy, noisy data. It may seem like magic, but it's just statistics.


This is the Principle of Parsimony or Occam’s Razor, which recommends searching for explanations using the smallest possible set of explanatory variables (fields) possible.

The core Funnelcast model is based on empirical win rates, computed as a function of time from opportunity creation (or entering to a stage) to won. This approach needs just two fields per opportunity and sales stage: start date (when an opportunity began or transitioned to a sales stage) and won date (for opportunities that were won).

These data are readily available from most CRM systems. Salesforce, automatically captures this data in a history file. So, if you are on Salesforce, there is no need to collect data for several sales cycles before producing meaningful insights. Similarly, there is no need to integrate with third-party systems. You can see results immediately.

Since our data requirements are so light, and the required data are such important indicators of deal status, our data are less likely to be wrong or missing than other, possibly messy and incomplete, CRM data.

Salespeople are sometimes sloppy in recording even these basic fields. But given enough data, errors average out. While individual deals scores may be inaccurate, the aggregate forecast for the portfolio can still be quite good.

Funnelcast produces win rate curves like the ones shown. These curves are very powerful signals about the timing and likelihood that a deal will close.


Win rates segmented by business type
Win rates segmented by sales stage

We also use other fields! Segmenting analyses can produce even more powerful predictions. New business for example, has considerably lower win rates than expansion. Other fields become increasingly relevant near the end of sales cycles, and Funnelcast uses them when it makes sense.


Summarizing:

  1. Funnelcast core models rely on only two fields.

  2. Those fields are automatically recorded by Salesforce.

  3. They are less likely to be wrong than other fields.

  4. Funnelcast models use more fields when it makes sense.


You get very powerful models that produce great forecasts and insights—with no third-party data integration or delay. Moreover, because the forecasts are built on the simple concept of win rate (as a function of time), the models are produce intuitive insights. In contrast, "AI" models are black boxes that don't give you any insights about why the forecast is what it is—or what to do about it.


Immediate, understandable, intuitive, and actionable insights. Oh, and great forecasts.



Commentaires


bottom of page