Predictive Insight Packs & “Why This?”
Seven packs that score risk from your own records with stated rules and statistics, each with an explanation of what drove the score, actions that go through the normal review, and a record of whether each insight turned out to be right.
The Seven Packs
| Pack | What it looks at |
|---|---|
| Stockout risk | Days of stock left from on-hand and recent daily sales (30 and 90 days, with a seasonal uplift in peak months), against lead time and safety stock. |
| Supplier delay | On open purchase orders: the supplier’s late share over 12 months, how close or past the expected date it is, and whether the supplier has acknowledged the order. |
| Late payment | On unpaid invoices: days past due, the customer’s late-paid share over 12 months, and the size of the balance. |
| Cash pressure | Cash in the ledger plus receivables due in 30 days at your usual collection rate, less supplier bills due in 30 days. |
| Till shrinkage | Per till over 60 days: cash shortages as a share of cash sales, and the share of sessions that closed short. Overs do not cancel out shorts. |
| Approval SLA breach | A pending approval’s age against the usual time for that kind of document over the last 90 days. |
| Asset custody risk | The asset risk score: value, risk level, time since last verified, overdue return, and how often and how widely it moves. |
Each pack lists its top 50 rows, banded High at 75 or above and Medium at 50 or above. Packs refresh every morning, and can be refreshed by hand. If a refresh has not run for more than 30 hours, the pack and its “Why this?” panel say the figures are stale and when they were last refreshed. Each pack needs the view AI insights permission plus the permission for its module, and its module’s AI switch on.
“Why This?”
Every pack row, demand forecast and anomaly has a Why this? panel, on the web and in the mobile app. It shows the drivers and their values, the records and time window used, how many records it is based on, the confidence and the reason for it, the method, and the recommended action. If the reasoning does not hold for your situation, you can see exactly which number to question.
Acting On An Insight
- Raise a procurement request or request a stock transfer — created as pending, and nothing moves until it is approved the usual way.
- Create a follow-up task, alert a colleague, pin a report to the dashboard or open a support note.
Each action needs the same permission it would need anywhere else. A request or transfer worth more than a threshold you set under Settings → AI & data waits in Insight actions for someone with the approve AI actions permission. The person who raised it cannot approve it.
Feedback And Outcomes
Mark an insight Helpful, Not helpful, False positive, Accepted or Dismissed; false positive and dismissed need a reason. Fourteen days later AWRA records what happened to it: resolved, improved, worsened or unchanged. Usage is broken down by module, role and model, so you can see where AI is used and where its advice is acted on.
Filters On Every AI Answer
Before an AI answer reaches you, AWRA removes anything that looks like a card number, key or password, withholds an answer that names another workspace, and drops instructions such as deleting everything or switching off multi-factor. In the assistants that cite sources, a sentence claiming a price, a certification or a guarantee the sources do not support is dropped. Each catch is logged.
If one AI provider fails, the request moves to the next one, and if none answers you get the figures without the AI summary. A request made for a different workspace than the one you are signed in to is refused. Settings → AI & data counts both: answers that came from a fallback provider, and requests refused.
Need help with this topic?
Capture the module, record number, branch or warehouse, user, visible message, and what you expected to happen before contacting support.