What kind of forecasting models have proven to be most effective for you?

I think about forecasting and making it effective in three large areas: (1) process, (2) tools/models, and (3) data. Without getting each of those right, no "perfect model" will make forecasting great. My experience forecasting at Intercom is different from what we did at LinkedIn so it's important to think about your specific context and needs before replicating any one forecasting approach.

1. Process: At LinkedIn and Intercom, we run a forecasting cadence that met the needs of the business. This typically started on Tuesday with a sales team level forecasting meeting. I believe it's important to have a team meeting so reps can discuss their commit/outlook, identify large deals, and discuss this as a team. It's a great way to learn from each other and bring newer reps up to speed. On Wednesday, the Managers would gather with their 2nd line leader for a meeting (again best to meet in person if possible). Then Thursday the 2nd line leaders would come together for a session (or two depending on global time zones) that leads to the Sales VP/CRO weekly call. That last meeting was a chance for the leaders to pressure test the forecasts, understand any emerging trends, risks, or opportunities. From there, things differ considerably between Intercom and LinkedIn. At LinkedIn, we produced a lengthy write-up of the bi-weekly forecast with qualitative and quantitative commentary and submitted an official forecast through the Anaplan tool. That was taken into a global review meeting for the CEO and CFO across multiple business lines to ultimately provide information to Microsoft. At Intercom, we don't have the same needs so the process doesn't involve a formal write-up (we do a summary for ourselves and Finance partners). Once a month, we have a executive level business review and the forecast helps prepare us for that but is not the primary input. So the generalized "rule" is to really think through the end-use of your forecast and build a process that supports that.

2. Tools/Models: We use Clari at Intercom and had Dynamics and Anaplan at LinkedIn. Any tool is fine for your forecasting if it gets you the essentials and is easy to use. I think having two modeling approaches is helpful. First, have a bottom-up driven by the CRM opportunities and rolling up from the reps through the Sales hierarchy. This can get you the color you need from the people closest to the deals and customers. The second, a top-down model driven by Sales Ops (and perhaps in collaborating with FP&A partners), can help spot trends or anticipate a known change in the business that the Sales teams might not be able to factor in. For example, if we knew a pricing change were coming in one month, the Sales team might struggle to understand the implications whereas SOps and FP&A would be able to model it. These two models can then be used to pressure-test each other and provide better accuracy and confidence in the forecast. Once you move beyond two models, you are a probably getting rapidly diminishing utility and could use that time better elsewhere. That said, getting forecast inputs from other teams like Marketing (e.g., MQL expectations and top of funnel performance) and Sales Development will help hone any forecasting process, but I wouldn't run those within my team.

3. Data: We also think through the various data sources we use in forecasting and ensure they are available to all participants and as accurate as possible. We recently started a weekly pipeline hygiene meeting, led by each manager with their team, as a Friday afternoon "let's clean everything up". That has shown marked improvements in hygiene on essentials like close date, expected ARR, stage, next meeting date, etc. that can get lost if this is left to reps on their own with a suggestion to "do your hygiene".

The “bottom-up” process (i.e. the sales teams’ forecast rollup) gives real time sentiment from customers and the field but can be biased by human-led judgment.

The “top-down” process (i.e. analyses of consolidated data) brings objectivity and separates signal from noise, though it ignores information that is not yet captured in the data (e.g. a large deal that will be pushed to the next quarter).

Over time, I found that a combination of “bottom-up” forecasting with “top-down” forecasting is the most effective way to forecast accurately.

A nascent area of capability building and exploration is around using AI and machine learning for forecasting. Collectively, thousands of hours are spent every week in forecasting activities, across all levels of the organization. In the near future, technology will free-up a large portion of this time, at the same time that will make our forecasting and planning processes even more accurate.