Every vendor pitching “predictive planning” is pitching it to the CFO. Oracle, Anaplan, Board, Workday Adaptive, Pigment, all of them put the phrase on the product page and all of them sell into finance. That makes the CFO the person who decides what the phrase means inside the company. If it means a feature set, FP&A gets better plumbing and the same plan. If it means a discipline, four things about how finance works have to change, and none of them are in the demo.
The budget becomes a scenario set
The annual budget is a single world with numbers in it. Predictive planning replaces the single world with three to five named ones and attaches a budget view to each. Not good-base-bad, which is one forecast with error bars drawn on. Distinct worlds on real cleavages. A world where the Fed keeps hiking into the spring and demand softens. A world where rates stop in December and the consumer holds. A world where input costs move and rates do not. Each world has a revenue line, a cost line, and a capital envelope that make sense inside it.
What the board approves changes shape. Instead of one plan with a variance tolerance, it approves a base posture plus the pre-agreed moves for each other world. The November meeting stops being the only moment strategy gets discussed and becomes the moment the scenario set is adopted. Finance still owns the model. The model just has more than one column, and the columns have names the executive team uses in conversation.
Re-forecasts become re-stakes
FP&A already runs a rolling forecast or a quarterly reforecast, and that cadence is worth keeping. The problem is what it is allowed to touch. A reforecast updates the numbers inside a fixed set of commitments. It can take revenue from one figure to a lower one and call that an adjustment. It cannot take a capital commitment off the table, because the commitment was approved in November and the reforecast was not given the authority to re-open it.
Predictive planning adds a second layer on top of the reforecast: the monthly Convergence Review, where every active scenario is sorted into converging, holding, or collapsing, and the stakes attached to any scenario that moved get re-examined. The reforecast tells you the number moved. The Convergence Review decides whether the commitment should. Finance brings the first one. The executive team runs the second. The Stake Sizing Matrix is the shared language for the resizing, so a commitment that leans on a collapsing scenario gets staged or hedged rather than defended.
Assumptions get a register, and the register gets expiry dates
This is the change that costs the least and pays off first. Pull every load-bearing assumption out of the model and put it in one register: the rate assumption, the demand growth rate, the key input costs, the hiring cost and availability, the FX line, whatever else the plan cannot survive without. For each one, record the value, the date it was written, the public signal that would tell you it broke, the owner who brings that signal to the Convergence Review, and the date by which it has to be argued again.
The expiry date is the part most finance teams skip and the part that matters most. An assumption with no expiry is defended by inertia. Nobody re-argues it because nobody was assigned to. This cycle’s rate line is the live example: the Fed raised its target ceiling to 4.00% on September 17, the market has a December hike priced near 97%, and most 2027 budgets still carry a rate assumption written the first week of September. That line would have been caught by a register with a sixty-day expiry. It was not caught by the reforecast, because the reforecast does not own assumptions.
The same logic runs at a larger scale. The book’s 18-Month Sunset is the expiry date on the annual cycle itself, a sequenced plan to retire the November marathon while the Loop stands up beside it, with finance owning the controls and audit mapping. The assumption register is the Sunset in miniature. Nothing in the plan lives without a date on which someone has to defend it again.
The vendor conversation changes
Buy the platform. The data plumbing alone usually justifies it, and two phases of the Loop, Scan and Steer, genuinely need the plumbing. But buy it knowing which phases it serves and which it does not. Run the Software/Discipline Gap Audit before the renewal: for each of Scan, Story, Stake, and Steer, score what the current software does for that phase from zero to ten, and score what the discipline does for it. The pattern in most companies that already own a planning platform is high software, low discipline, in the middle two phases. That reading tells you the next dollar belongs to the discipline, not the next seat.
Then ask the vendor four questions the demo will not answer on its own.
- Can the model carry three to five named scenarios as first-class objects, with a budget view each, rather than as sensitivity runs on one base case?
- Can an assumption carry metadata: the date it was written, a linked external series, an owner, and an expiry date that surfaces when it passes?
- Can a commitment be tagged to the scenario it was sized for, so when that scenario collapses the commitments that depend on it are one query away?
- Where does the platform expect the judgment to happen, and who in the workflow is named on it? If the answer is “the AI suggests and the planner accepts,” that is the AI/Human Split run backwards.
None of these are gotchas. The better platforms can do most of it. The point is to know what you are buying and to keep the discipline on your side of the line. The software-or-discipline question is worth reading in full before the renewal meeting.
What this does to the FP&A team
It raises the job. Variance analysis against a fixed plan is a reporting function. Maintaining a scenario set with a live assumption register and preparing commitments for sizing is a planning function, and it is the work the best FP&A people already wish they were doing. The role that owns the Loop, the Predictionist, often comes from this bench. The CFO who builds the register and runs the Gap Audit has usually already identified who that person is.
What to do Monday
Build the assumption register for the 2027 plan before it goes to the board. Five to ten rows, each with a value, a date, a signal, an owner, and an expiry. Pull today’s value for each signal from the Signal Library and compare it to the date the assumption was written. The full afternoon version sorts what you find into holding, drifting, and broken. Then put the first Convergence Review on the calendar for December with the CEO in the room. The full method, including the Stake and Steer chapters and the chapter on choosing a platform, is in the book.