What's actually changed in planning tools over the past few years
Anyone listing release notes is selling. The useful reading of a roadmap is different: which changes alter how the project gets done — and which don't yet, however good the demo looks.
Below, five moves that actually happened over the past few years in the platforms we work with, and an honest read of where each one stands.
1. AI moved out of forecasting and into the model
The first wave was machine learning applied to forecasting: PlanIQ on Anaplan (now evolving into Anaplan Forecaster), predictive forecasting inside PAW on the IBM side. It's real and it works — but it's narrow: needs history and improves one number.
The second wave is the one that changes the project: assistants that read, explain, and build the model. CoModeler for structure creation via natural language, CoPlanner for conversational analysis, and on the IBM side, agents built on watsonx.
Where this delivers value today:
- Exploration and documentation. Asking "what feeds this account" and getting an answer in seconds is worth a lot in a ten-year model nobody documented. Most of the cost of sustaining a legacy model is archaeology.
- Variance explanation. "Why did this number change between versions" is the most repeated question in any planning cycle, and it finally has a shortcut.
Where it's still a demo:
- Generating production-grade structure from a prompt. It generates something plausible. And plausible is the worst possible outcome in a planning model — it passes the smoke test and fails at close, when people are already trusting the number.
Net effect today: less time documenting and navigating, the same time modelling. That's already a lot. It just isn't what the demo promises.
2. Calculation engine is no longer one thing
For decades, choosing the tool was choosing the engine. That has changed.
Anaplan now has Polaris alongside HyperBlock, with different sparsity handling — and dimension explosion is no longer an automatic "no". IBM kept the TM1 engine and changed the architecture around it (subject of the v12 post).
Practical effect: "does it fit the tool?" has more than one answer per tool — and it's a question that has to be asked earlier, because the engine choice isn't easy to reverse after the model is built.
3. The model lifecycle became software
Versioned promotion between environments, traceable deployment, parity between dev, QA, and production. ALM on Anaplan, Git integration in Planning Analytics v12.
It's the least glamorous change on the list and probably the one with the biggest real impact on delivery quality. Anyone sustaining a large model knows most serious incidents don't come from formula errors: they come from someone publishing the wrong thing to the wrong environment.
4. The data layer moved out of the planning project
Five years ago it was normal for the planning tool to also be the place where data was cleaned, deduplicated, and reconciled. Today the standard is the lakehouse or corporate data warehouse as the source, with the planning tool consuming already-processed data.
Data orchestration tools inside the platforms themselves push in this direction, and client expectations have followed.
Practical effect: the integration discussion moved — earlier than the project, and outside it. And a planning project without a defined data layer today looks obviously incomplete in a way that it didn't before.
5. Excel is back — and this time with governance
There was a period when every platform wanted to take users out of the spreadsheet. It didn't work, and platforms stopped trying.
PAfE, Anaplan's add-ins, and the general evolution of interfaces accepted the inevitable. The real change isn't that Excel came back: it's that the spreadsheet became a window into a governed model, instead of being the model. The number still has an owner and an origin, even when the user consumes it in a cell.
What hasn't changed
The hard part is still exactly where it was: agreeing what to plan, at what level, with whom, and against which number.
Nothing in recent years has made that conversation easier. Tools have gotten much better at everything around the decision, and exactly as good as before at the decision itself.
Worth remembering when a roadmap is presented as the solution to a process problem.
How to read a vendor roadmap
Three questions that save frustration:
- Is it GA in my region and on my licence tier? A feature announced at a conference and a feature available in your contract are different things — sometimes separated by years.
- Does it work on my model or on an ideal model? The demo runs on a clean base, with small dimensions and consistent naming. Your model isn't like that.
- If I depend on this, what happens if it's discontinued? Platforms are launching and retiring features faster than they used to. Worth knowing the fallback plan before building on top.
We work with IBM Planning Analytics and Anaplan side by side, which lets us assess platforms without team-cheering — see the TM1 vs Anaplan comparison or get in touch.