Set up recurring orders

Configure reorder suggestions

The consumption window, the confidence floor, the channel and the per-customer opt-in — the four settings that decide whether buyers read your suggestions or learn to ignore them.

Reorder suggestions propose the next order to a customer before they ask for it. This article is the settings behind that; the arithmetic is in How smart reordering works.

Before you begin. Suggestions need history. An account that migrated last month has nothing to predict from, and a customer who orders four times a year needs two years of it. Check that the accounts you are switching this on for have at least four orders of the articles you care about, and that those orders are in the platform rather than only in your ERP — see What counts as an order.

Not to be confused with your own reorder point

Two different things use the word "reorder", and mixing them up wastes a week.

Reorder point (Meldebestand)Reorder suggestion
Whose stockYours, in your warehouseYour customer's, inferred
Data sourceon_hand in a locationThe account's order history
What it producesAn alert that you should buyA proposal that the customer should buy
Where it livesOrder Management › StockThis area

Your reorder point is a purchasing signal — see Manage stock day to day. Everything below is about the customer's side.

The four settings that matter

Consumption window

How far back the consumption rate is calculated over.

WindowRight forWrong for
90 daysFast-moving C-Teile ordered monthly or more oftenAnything quarterly — you will see one order
180 daysThe usual defaultStrongly seasonal goods
365 daysSlow movers, quarterly ordersAccounts whose volume changed this year

Longer is not safer. A twelve-month window on an account that doubled its production in spring predicts last year's demand with great confidence. If you have to pick one number for everything, pick 180 days.

Minimum confidence

The floor a prediction has to clear before it becomes a suggestion. It combines the number of past orders, the length of the history and how regular the intervals are.

Set it high to start with. A high floor produces a handful of suggestions that are nearly all right, and buyers learn that your suggestions are worth opening. A low floor produces many suggestions that are mostly noise, and buyers stop reading them inside a month — which is a much more expensive mistake than sending too few, because you cannot easily undo it.

Channel

Where the suggestion appears. You can run more than one, and for different accounts.

ChannelReachesRight when
StorefrontThe buyer, next time they log inThe account already uses the shop
EmailThe buyer, whether or not they log inYou want the suggestion to be the reason they log in
Sales repYour Innendienst or Außendienst, as a call reasonThe account still orders by phone — this is the migration lever

The third one is underrated. An account that has never logged in is exactly the account where "you are probably about to run out of 8815-F" is a good reason to phone, and it works before any digital adoption at all.

Per-customer opt-in

Whether an account receives suggestions is per account, and it should be. Some buyers want them; some read them as a supplier telling them how to do their job.

Decide the default deliberately. Opt-out (everyone gets them unless switched off) reaches more accounts and produces the occasional irritated call. Opt-in (nobody gets them until agreed) is slower and is the right default for accounts where a named rep owns the relationship.

Set it up

  1. Choose the accounts. Start with a handful that order regularly and that you have a working relationship with. A segment is the natural way to hold the list.
  2. Set the consumption window — 180 days unless you have a reason.
  3. Set the confidence floor high, then lower it only after you have looked at a month of real suggestions.
  4. Set the lead time and safety days. Your fulfilment time plus the customer's own internal delay, and a buffer in days of cover. Ten days of safety is a reasonable start for cheap consumables.
  5. Set the coverage target — how many days a suggested delivery should cover. Match it to the interval you want them ordering on.
  6. Exclude what should not be predicted. Project articles, seasonal goods, anything bought once or twice, anything with a shelf life shorter than the coverage target, and articles a customer resells rather than consumes.
  7. Turn on packaging rounding so a suggested quantity is always orderable — rounded up to the packaging unit and never below the minimum order quantity.
  8. Choose the channel and the notification lead time.
  9. Set the acceptance behaviour: suggestion only, or suggestion that auto-converts after n days unless declined. Start with suggestion only.
  10. Switch it on for the pilot segment and read the first month's output yourself before anyone else sees it.

Sanity-check the output before a customer does

Take the first batch of suggestions and check them by hand:

  • Does every suggested quantity round to a shippable pack?
  • Is any suggestion for an article the account bought exactly once?
  • Do the dates cluster absurdly — twenty articles all due the same Tuesday? That usually means the last order is being read as the only order.
  • Would you defend this suggestion on the phone? If not, raise the floor.
Do not switch on auto-conversion during the pilot. A suggestion that is wrong costs you an ignored email. An auto-converted order that is wrong costs you a return, a credit note and a conversation about whether you are ordering on the customer's behalf. Earn the second one with a quarter of correct suggestions first.

What to check

  • Suggestions exist for the pilot accounts, and their count is in the tens, not the thousands.
  • Excluded articles produce none.
  • A suggestion carries the account, the articles, the quantities, the date and the assumption it was based on — a suggestion you cannot explain is one your rep cannot defend.

Next