Act on reorder suggestions
A suggestion that nobody acts on is a report. This article is about the three routes from a proposed reorder to a real order, and which accounts belong on which route.
Route 1: the buyer accepts it in the storefront
The cheapest route, and the one to design for first.
The buyer logs in and sees the suggestion on their account overview: the articles, the quantities, the date it is based on. They can accept it whole, adjust the quantities, drop a line, or dismiss it. Accepting puts the lines into a cart, not straight into an order, so the buyer keeps the last look before checkout, along with their purchase-order number, cost centres and delivery address.
Two details decide whether buyers use it:
- The quantities have to be orderable. Rounded to the packaging unit, at or above the minimum order quantity. A suggestion that a buyer has to correct every time trains them to ignore it.
- The suggestion has to say why. "Based on your orders since February, roughly 26 days of cover left" is a sentence a buyer can check against their own shelf. A bare list of quantities is a guess they have to verify from scratch.
If the buyer's approval path applies, the resulting order goes through it exactly as a manual one would. See Submit for approval.
Route 2: your rep uses it as a call reason
For accounts that do not use the storefront, this is where the value is, and it works before any digital adoption at all. Most of it works today.
Your Innendienst opens the account in CRM › Organizations and reads the Order history on the Orders tab: Last order, Orders 30 d, Orders 90 d, Revenue 30 d, Revenue 90 d. That tells them whether this account is early, late or on time against its own rhythm, which is a specific reason to phone and beats "checking in" by a wide margin. The planned suggestion adds the third piece: the articles predicted to be running low.
Give the rep three things and the call runs itself:
- The suggested lines, with quantities and the customer's own article numbers. Until suggestions ship, the customer's order list from last time is the list.
- The account's rhythm, from the Order history above. See The metrics that matter in B2B.
- One action that turns the accepted lines into an order without retyping them. Today that is converting the order list.
Log the outcome on the customer timeline, including a no. "They have six months of stock, they bought ahead of the price increase" is exactly the correction a prediction needs, and it is not in any order record. See Log customer activity.
This route is also your migration lever. A buyer who accepts three suggestions by phone is a buyer you can ask to accept the fourth in the shop.
Route 3: auto-conversion
The suggestion converts to an order on its own unless the customer declines within a set number of days.
This works, and it only works under conditions. All of them, not most of them:
| Condition | Why it is not optional |
|---|---|
| The customer agreed to it in writing | Otherwise you are ordering on their behalf |
| The article list is agreed and closed | A new article appearing in an automatic order is a return |
| A quantity ceiling per line and per order | Caps the damage when the prediction is wrong |
| A notice that arrives before the deadline, to a person | An automated mail to info@ is not a notice |
| One obvious way to decline, in one click | If declining is hard, the next escalation is a phone call to your boss |
| A stop for anything unusual | A quantity far outside the account's history should route to a human |
An account that has run manually for a quarter with a high acceptance rate is a candidate. An account that accepts half the suggestions and edits the rest is not. That edit rate is the model telling you it is not ready.
When a suggestion is wrong
Wrong suggestions are information, not only noise. Three things to do with them:
- Record the decline reason where you can. "Bought elsewhere", "project ended", "switched article", "we hold more stock than you think" each point at a different fix.
- Exclude the article if it is structurally unpredictable: a project item, a seasonal item, something they resell.
- Ask for the real consumption rate for the account's top articles and use it instead of the inferred one. One conversation beats a model change.
If a whole account's suggestions drift at once, that is usually substitution, stockpiling or a change in their production volume. See How smart reordering works.
Promote the reliable ones
An article an account has ordered like clockwork for two years does not need a prediction. It needs a schedule. Move it onto a standing order: the run is mechanical, the buyer stops being asked the same question every six weeks, and the revenue becomes forecastable instead of merely likely.
Doing that on the ten or fifteen most predictable article-account pairs is usually worth more than any improvement to the prediction itself.
What to check
- The acceptance rate per account. Below roughly a third, raise the confidence floor for that account or move it to the rep route.
- The edit rate on accepted suggestions. High edits mean the quantities are wrong, not the timing.
- Declines with reasons attached, rather than silent dismissals.
- No auto-converted order waiting for release past your approval SLA.
Next
- Standing orders — where the reliable ones should end up.
- Recurring revenue health — measuring whether any of this is working.