Moving from S&OP to touchless forecasting

First question: what does “touchless” actually mean here?

The term gets used for three very different things, and the difference decides the size of the program. At the modest end, touchless means the statistical baseline runs without manual adjustment for the bulk of SKUs, and humans see only the exceptions — the launches, the end-of-life items, the lines where the model is not trusted yet. In the middle it means the whole demand plan, not just the baseline, moves to machine-generated with planners approving by exception. At the far end — the one vendors demo — no planner touches a number at all and the S&OP meeting reviews outcomes instead of spreadsheets. Most practitioners who have lived through this will tell you the far end is a direction of travel, not a project end state, and that programs which started by defining which decisions stay human went far better than programs that started with a tool selection.

The useful exercise before anything else: take last year’s planning calendar and mark every touch — every spreadsheet edited, every consensus call held to argue a forecast by a point or two. Then ask, touch by touch, what that touch changed. The practitioners who ran this honestly found a large share of touches changed nothing, and that inventory of no-op touches is the real business case.

Measure forecast value-add, not model accuracy

The fastest way to lose an intelligent-planning program is to run it on accuracy metrics alone. A model can improve MAPE against a naive forecast while making the plan worse where it matters — on the promotable SKUs, the long tails, the seasonal peaks that drive the capacity decisions. The practitioners who trust their numbers moved to forecast value-add: does the machine forecast, and does each human-adjusted version of it, beat a naive baseline at the aggregation levels where actual decisions get made — revenue, volume, capacity by month? Where a touch adds value, keep the touch and make it deliberate. Where it does not, removing it is not a job loss; it is the point. This is also the metric that survives the political fight, because it protects the planner whose judgment genuinely adds value from being flattened along with the noise.

The process redesign is the hard part, not the tool

The technology of a demand forecast has been a solved problem at most companies for years; the process around it is not. Touchless planning breaks the monthly cadence the organization is built around: if the numbers refresh weekly or daily, the demand review can stop being a meeting where last month’s forecast is re-litigated and become an exception review with a short agenda. Consensus changes shape too — instead of every function submitting its own number and negotiating, there is one baseline and people are invited only where they disagree. Practitioners report that the surviving planners move from producing numbers to investigating exceptions, which is a career change, not a tooling change — and that the programs which said this out loud early, and retrained for it, kept their best planners through the transition. The ones that announced a “planning transformation” and left the org chart in a drawer discovered that a sceptical planner can quietly re-touch every number and make the whole program read-only theater.

Master data decides how far you get

Nothing exposes data debt like asking a model to forecast it. Item codes that changed at an acquisition, a customer hierarchy where one retailer lives under three names, promotional calendars that live in a buyer’s inbox, lead times that are fiction — each one caps how much of the portfolio can go untouched. The practitioners who got furthest scoped the first release to the cleanest part of the portfolio and let the program’s credibility fund the data cleanup, rather than pausing the program for an eighteen-month data project that would have been cancelled at the first budget review.

Three questions worth asking someone who has done it

1. Which planner touches did you remove first, which did you defend, and how did the organization react to each?
2. Where did the machine forecast genuinely beat your planners — and where did you have to build an exception lane for the model to stay trusted?
3. What happened to your monthly S&OP calendar, and what would you keep from the old cadence if you started again?

A software vendor answers these with a demo. A peer at your scale answers them with the org chart they actually changed, the metric their CFO accepted, and the month it got worse before it got better. That difference is the entire product.