The first site is a pilot, and pilots succeed for reasons that do not transfer. There is a champion. There is attention from above. There is tolerance for teething problems. None of those are available at site four.

The question that decides a multi-site rollout is which parts you standardise and which you let each plant own. Get it backwards and you either impose a template nobody uses, or end up with five incompatible systems and no group-level view.

Standardise the structure, not the content

The rule that holds up: anything you want to compare across sites must be identical. Everything else should be local.

Identical everywhere

  • The scoring scales. Impact and frequency, three levels each. If Verona uses a 1–5 scale and Lyon uses high/medium/low, your group view is decorative.
  • The waste categories. Same seven, same definitions, same decision rule for ambiguous cases.
  • The status model. Open, in progress, resolved, and — critically — the same rule about who is allowed to close.
  • The cadence types. A "daily supervisor walk" should mean roughly the same thing in every plant.

Local everywhere

  • The route and checkpoints. Obviously. A stow aisle is not an assembly cell.
  • Who walks and when. Shift patterns differ. Imposing 06:15 on a site that starts at 05:00 just guarantees the walk gets skipped.
  • Which improvements get capacity. Local constraints, local call.
  • Language. People log issues in the language they think in. Force English on a floor that does not use it and you get three-word entries.
The failure mode to watch
Central teams standardise the route (which should be local) and leave the scoring loose (which should be central). It feels like the opposite of interference. It produces walks that annoy people and data that cannot be compared.

Sequence matters more than speed

The most common rollout mistake is going wide before the second site works. Site one succeeded with a champion; you do not know yet whether the routine works without one.

A sequence that tends to hold:

  1. Site one, with a champion. Prove the habit exists. Three months minimum.
  2. Site two, deliberately without one. This is the real test. If it only works where someone is passionate about it, you have a person, not a system.
  3. Fix what broke at site two. Almost always: the cadence was unrealistic, or nobody owned the closing of issues.
  4. Then go wide. Three or four at a time, with the fixed version.

Data separation is a rollout feature

Plants do not want their open-issue count visible to a sister site, and they are right not to. Early visibility across sites turns the tool into a league table, and league tables produce staged floors just as reliably as audits do.

Scope data to each organisation and site with role-based access, and give group-level roles a deliberate, aggregated view rather than raw access to everyone's issues. The counts roll up. The individual observations stay where they were made.

What the group view is actually for

Not ranking. The useful group-level questions are about the system, not the sites:

  • Is the same waste type dominant across plants? That points at a design or procurement decision, not a local one.
  • Is a fix that worked at one site being tried at others? Transfer is the entire economic case for a group.
  • Is the cadence holding everywhere, or only where someone is watching?

The last one is the only leading indicator worth putting on an executive dashboard. Walk completion rate tells you whether the system is alive. Everything else tells you what it found.