Annual Planning Is a Compromise
I used that planning method for years before I understood that every step of it was a compromise forced by expensive judgment.
Almost every large company has a team that does annual planning for the business; sales operations, pricing and supply chain planning all count. The kind I worked on was pharma SFE, sales force effectiveness. Customer tiers, territories redrawn once a year, quotas fixed at the start of the year: I used all of these as methodology. It took me a long time to realize they weren’t the best way to do things, only workarounds for when judgment was too expensive to use often.
Now that judgment is cheap, there’s no need to keep those workarounds exactly as they are. These teams won’t disappear, but the way they work will change, from running one big project a year to handling individual decisions as they come up. That’s my own view, not an industry consensus.
Every step is a compromise
SFE has a mature methodology, often described in the industry as ten steps: data foundation, potential assessment, segmentation, targeting, call frequency, sizing, territory alignment, quota setting, incentive design and performance management. Take four of them.
Segmentation sorts tens of thousands of doctors into four tiers, A to D. Doctors in the same tier are actually quite different; nobody can look at them one by one.
Territories are redrawn once a year. The market obviously changes more often than that, but every redraw keeps several people busy for weeks, followed by round after round of negotiation with each region.
Quotas are set at the start of the year and barely change. If the market shifts mid-year, quotas rarely follow, because recalculating them costs too much.
Bonuses are paid quarterly. What a rep does this month shows up in their pay three months later.
At heart, the ten-step method compresses countless small decisions into a few big ones, batches them, and hands them to a small group of analysts and consultants who process them by rule. I was one of those consultants. When judgment was scarce, this was sensible, even clever.
The cost of the compromise
When this year’s plan is built on last year’s data, by the time it has run to year-end the data behind it is two years old. Two doctors in the same tier, one just promoted to department head and one about to retire, get the same visit frequency. The front line can’t see how their quotas were calculated. What they can’t see, they don’t trust, and they spend a great deal of energy negotiating quotas instead.
What’s more of a shame is that hardly anyone checks afterwards. There’s no systematic answer to whether the plan was right, and the next year the same rules run again.
The premise has changed
Looking at each doctor’s data individually and making a recommendation used to cost too much. Now it costs almost nothing. The old situation where CRM, flow data and market data each lived in their own silo is also easing: after years of work on master data and data platforms, most companies are at the stage where the pieces can connect but haven’t been connected well yet. And how a decision turns out can now flow back into the system automatically and inform the next one.
What I’m trying instead
I call what replaces SFE a CIU, a Commercial Intelligence Unit. I built an MVP that runs 11 operating scenarios end to end, from redrawing territories to recovering at-risk quotas mid-month.
The basic difference from SFE is the unit of work. SFE works in projects: one territory realignment, one round of quota setting. A CIU works one decision at a time. Should this doctor’s visit frequency change next month? Should this territory’s quota be revised after a VBP (volume-based procurement) result? The question shifts from “how many visits a month for Tier B customers?” to “how should we follow up with this doctor this month, given the current policy situation?”
The core of the ten-step method is all still there: potential, segmentation, sizing, territories and quotas. It has just turned from a once-a-year project into an engine that can be rerun at any time. And reruns aren’t triggered only by the annual calendar but also by signals, such as VBP results, competitor moves, changes in hospital leadership, reps leaving and unusual attainment. Every plan is simulated before it goes live. If these two territories are merged, what happens to workload? If this quota changes, how many people’s bonuses does it affect? All of this rests on one shared set of business definitions and a decision log, so every decision can be traced and rolled back.
There are things it deliberately doesn’t do. The language model doesn’t do arithmetic; quotas, bonuses and workloads are all computed by deterministic code and can be recalculated. Compliance red lines are hard-coded and can’t be configured. At first the AI only makes proposals, and every decision that enters the log needs a person’s approval; more autonomy comes one level at a time. Judgments about people, such as who needs coaching or who has seemed off lately, stay with the manager.
What happens to the planners
They won’t disappear, but the job will change. They used to spend their time pulling data, building models, making slides and negotiating with regions. More of it will go to three things. One is setting the rules, such as which customers are worth covering and what counts as a balanced territory. Another is reviewing exceptions: when the front line overrules the system, someone has to look at the case and decide. The third is deciding how much authority to give, meaning which kinds of decisions the system can take on more of, and which to pull back.
Doing this well takes both a grasp of the business rules and fluency with data. Few people in a company have both. The people who did annual planning are among the few who do.
