Adaptive Planning fails at the first close, not at go-live
Eight decisions decide the outcome. Most get made before anyone opens the tool.
- The build is not the risk. A competent partner stands up a working budget model in a quarter.
- The risk is month end. Plan structure and actuals structure do not line up, and the variance report becomes a two-day reconciliation.
- Mapping is where the effort actually sits. Worktags, cost centers, companies and ledger accounts to Adaptive dimensions. Not the modeling.
- Five of the eight decisions belong to people who do not think they are on the project: the CFO, the controller, HR and IT security.
- Plan 7 months, not 4.5. Workday's published average does not survive contact with a real chart of accounts.
Start with the real timeline
Workday puts average deployment at 4.5 months. I have not seen one land there.
Add workforce, sales or consolidation and you are at 8 months. The gap between the brochure and the field is not partner incompetence. It is the mapping work nobody scopes, plus one full planning cycle in parallel before the old process can be retired. Neither of those is optional, and neither is in the estimator.
The eight decisions
Decisions, not tasks. Each has an owner who does not work for your implementation partner. Open any one for the detail.
Who owns the modelVP of FP&A
Not who administers it. Who decides what it does, approves a change, and answers for the number in front of the CFO. Name a person and a backup.
The model becomes an IT ticket queue. Mid-cycle changes wait on a sprint, and planners route around the system instead of through it.
The level structure, designed against FinancialsController with FP&A
Levels carry cost centers, departments, entities and regions. They drive sheets, reports, security and workflow at once, which is why a sloppy level design surfaces in four places.
Two things worth knowing. Adaptive supports ragged hierarchies, so branches do not all need the same depth. And level attributes handle most grouping needs without adding a dimension.
Actuals need translating on every load. The mismatch never gets fixed. It gets absorbed by a person doing it by hand.
How few dimensions you can live withFP&A holds the line
Five custom dimensions at 100 members each creates roughly 10 billion possible intersections. Practitioner guidance puts the workable account range at 500 to 2,000, with calculation performance degrading past about 3,000.
Check whether an attribute, an account structure or a report filter would do first.
Sheets get slow. Slow sheets send planners back to Excel, and adoption is the one thing configuration cannot fix afterward.
The actuals feed, mapped and tested to the dollarController
Two patterns are normal. The native Workday to Workday connection when both products sit on one tenant, or an EIB load when you need transformation control or actuals from more than one ledger.
Neither is zero configuration. Somebody decides which worktags, cost centers, companies and ledger accounts map to which Adaptive dimensions. That mapping is most of the real work.
It also depends on the chart of accounts being stable. An Adaptive rollout running next to an unfinished COA redesign will map twice. And the refresh belongs on the close calendar, because actuals loaded before the close is final show numbers that will change.
This is the one that shows up at the first close. See the diagram below.
Cube or sheet, for each thing you planModel owner
Cubes handle high-volume, multi-dimensional actuals. Sheets suit driver logic at a summary level. Most mature models are hybrid, which is fine when it was chosen.
Retrofitting a cube after several cycles of loading into sheets is a rebuild, and it takes the reports with it.
Version strategy, and who declares the official numberCFO defines
Actuals, budget, working forecast, prior forecast, long range, scenarios. The ceiling is 200 versions. Performance and human comprehension both live near 15 to 20 active. Virtual versions keep reporting views stable while the plan versions underneath them change.
Two people bring different forecasts to the same meeting. After that happens once you are repairing trust, not implementing a platform.
Security matched to the planning processController, HR, IT
Actuals carry different sensitivity than plan data. Access should follow who enters, reviews and consolidates, because those three people are often in three different places on the org chart.
I watched one over-broad role on a compensation object stop an entire end-to-end test cycle for a week. Every script that touched pay had to pause. The build was fine. The design decision was not.
A week of testing, at the point in a program where a week is most expensive.
The last cycle in the old worldCFO
Pick the date the spreadsheet process stops. Run one full cycle in parallel, prove the new one, retire the old one on the announced date. A date on the finance calendar, not an intention.
Both processes run indefinitely and the spreadsheets win. They are already trusted, and nobody needs permission to change one.
Model constraints above come from Workday documentation and published practitioner guidance. Sources at the end.
Where the variance report dies
Two structures have to face each other. Designed together, actuals land and variance reporting is a version-and-period selection. Designed apart, something translates between them every month, and that something is a person with a spreadsheet.
What is still recoverable, and when
Decisions do not fail. They harden. Each one stops being cheap at a different point in the program.
Drag the slider to where your program is now.
Design: everything is still cheap
Cost to change any of the eight right now: one meeting and a decision log entry. This is the only stage where that is true.
Score your own program
Eight questions. Decisions 1, 2 and 4 are gates: a No on any of them outranks the total.
Where does your program actually stand?
Practices that hold
Most published advice for this platform is about attitude. These are the ones I would hold a program to.
Phase by planning process
Not by department. No single department's plan is a complete cycle. Each phase should end with something Finance can run.
Answer the CFO's five questions
Build for those and nothing else in phase one. A model that answers five questions well gets used. One that could answer forty gets abandoned.
Second a real planner to the build
Someone entering numbers next quarter, full time, with authority to say no planner will do that.
Test with your ugliest data
Three mock loads minimum. Under 2% error on the final one. Control totals reconciled to the dollar by Finance.
Produce a variance report from the mock
The step everyone skips. A feed that loads clean and still yields an unreadable variance report has passed the wrong test.
Parallel run once, then stop
One cycle, both ways, differences explained. Parallel running is a proof, not a lifestyle.
Put the release calendar on the FP&A calendar
Two Workday releases a year, roughly a five-week preview each. Someone owns regression-testing the model against both.
Schedule the quarterly cleanup
Stale versions, unused dimension values, test accounts. Retire a level with version availability, never deletion, or you lose history.
Workday announced Adaptive Decision Intelligence in May 2026, and the Illuminate planning agents are landing through the year. Useful, and it raises the stakes on everything above. An agent answering questions against a badly mapped model produces confident wrong answers faster than a person could. Governance first.
Seven lessons, and what each cost
Every one was preventable at design for less than it cost to fix
The model reproduced the spreadsheets
Every tab became a sheet. The approval chain stayed as it was. Scout24 went in with a budget cycle spread across 90+ Excel files, and rebuilding that faithfully is the most common way to change nothing.
- Cost
- The full spend, none of the cycle time back
- Fix
- Design the process first. The tool is the second decision.
Actuals did not tie at the first close
Account groupings did not match, cost center hierarchies diverged. By the third cycle the variance report was being rebuilt in Excel every month. Live and irrelevant at the same time.
- Cause
- Level structure designed in isolation from the Financials hierarchy
- Cost
- Two to three days of senior finance time every close, indefinitely
- Fix
- Mirror the structures in design, prove it with a variance report off a mock load
Nobody owned the mapping
On a program this year, finance and procurement loads kept failing. No single person owned validation rules across the modules, so every team assumed another had it.
- Cost
- Weeks of build time, and a workstream that entered testing behind
- Fix
- One named owner, a date on every item. The failures stopped. It cost nothing.
The report library came over whole
On one program, 400 "critical" legacy reports collapsed to 90 against the actual run logs. Nobody asked for the other 310 again.
- Cost
- Build time on 310 reports, plus regression on them forever
- Fix
- Rationalize against usage evidence before you migrate
Complexity nobody remembers choosing
At one client, 80 small customizations added three weeks of regression to every release. Two releases a year. In Adaptive the equivalent is a dimension added for one request and account links chained three deep.
- Cost
- Six weeks a year, permanently
- Fix
- A change gate with a named approver, and cleanup with authority to delete
Security got retrofitted
One role exposed compensation data wider than intended. Every test script touching pay had to wait while it was redesigned.
- Cost
- A week of the end-to-end test cycle
- Fix
- Design access with the model, test it before UAT
The old process never died
Parallel running was meant to last one cycle. Nobody set the date, so it lasted four. The new tool became the place you entered numbers a second time.
- Cost
- Double entry across a planning year, and adoption you have to win back
- Fix
- The CFO announces the last old cycle at design, not after go-live
The 90-day test
Ninety days after go-live, five observable things. Not adoption percentages.
- Actuals landed on the close calendar without anyone asking, and they tie to the general ledger.
- The variance report came from selecting a version and a period. Nothing was exported.
- A planner outside Finance entered their own numbers without calling FP&A.
- Someone asked a scenario question in a meeting and got the answer in that meeting.
- The old spreadsheet process has an end date, and it has already passed.
Four of five means it worked and the fifth is fixable. Two of five means the model gets quietly abandoned inside a year, and the time to intervene is while the partner's design decisions are still recent enough to trace.
Get the scorecard
The same eight decisions on two printable pages, with an owner column and a due-by column. Print it and argue over it with the people whose names go in the middle. That argument is the deliverable.
Mine, from 20+ Workday programs: the 7-month average, 16 weeks for financial planning only, 8 months with workforce or consolidation added, the conversion gate thresholds, 400 reports to 90, the 80 customizations, the security role, the validation ownership fix. I have not personally run an Adaptive Planning implementation. The Adaptive-specific failure patterns below are attributed, not claimed.
Product and model constraints: Workday Adaptive Planning documentation; EPMLogic model design guide; Sama, actuals versus plan integration patterns; QBIX level structure primer.
The 4.5 month claim: Workday implementation estimator. Customer cycle times: Scout24, reported by Workday. FP&A time allocation: Cube FP&A statistics. AI direction: Workday Illuminate agents and the May 2026 Adaptive Decision Intelligence release.