Customers

The /customers page lists every customer the predictions worker has built a baseline for. Each row tells you four things at once: who they are, how often they pay you, what they spent in the last 90 days, and roughly when (and how much) you should expect from them next.

Columns

Column	What it shows
Customer	Display name (Xero contact name, or the contact's external id when none is recorded).
Cadence	Detected payment cadence (Weekly / Fortnightly / Monthly / Quarterly / Annual / None) + confidence %.
Last 90d	Sum of inflows attributed to this customer in the trailing 90 days.
Last seen	Days since the last invoice or payment from this customer.
Next Purchase Due	Date range pill ("May 18 – 24") + $X ± $Y amount pill — when and how much we expect from them next.
Risk	Churn-risk badge: Active (≤ 30 days), At risk (31–90 days), Churned (> 90 days or no signal).
ABC	Recursive 80/20 tier — A is the top 20% by revenue contribution, then B, then C.

The Next Purchase Due column mirrors the dashboard's Top-customers card so the two surfaces read as one family — same chip pattern, same colours, same rounding. The amount pill reads as a real range: the central $X is the per-period baseline (median when we have one, mean otherwise — the same "robust centre" rule the deviation classifier follows), and the ± $Y is the half-IQR (or stddev on legacy rows). A row whose cadence is "None" (or whose projected next emission falls outside the 13-week horizon) renders an em-dash in this column rather than a fabricated guess.

Sorting

Every column header is sortable — click once to sort ascending, click again to flip to descending. The active sort key shows a ▲ / ▼ arrow next to the header label, and the URL captures the state in ?sort=...&dir=... so a refresh or share preserves the view.

Sort key	Asc reads as	Desc reads as
name	A → Z	Z → A
cadence	Most-frequent first (Weekly → Annual → None)	Annual first, Weekly last
spend	Lowest last-90d-spend first	Highest last-90d-spend first (default)
churn	Most-recently-seen first	Most-stale first
nextPurchase	Soonest-due first (projected horizon week)	Latest-due first; rows with no projection on top
risk	Active customers first (then At risk, then Churned)	Churned first (then At risk, then Active)
abc	A tier first	C tier first

risk is different from churn — risk clusters customers by tier (Active / At risk / Churned), churn is a straight days-since-last-seen timeline. Use risk to find every at-risk customer regardless of exact age; use churn to find the single oldest unpaid contact.

Filters

A chip bar above the table lets you narrow the list by ABC tier, Cadence, and Risk — each group is multi-select within itself, and the three groups are intersected so an active filter set always reads as "show me only customers who are X and Y and Z."

Each chip is a link — clicking an active chip removes it from the filter, clicking an inactive one adds it. The URL is the source of truth (?abc=A,B&cadence=WEEKLY&risk=AT_RISK), so a filtered view is bookmarkable and shareable.

Filters survive sort clicks and pagination — the chip state never silently resets when you change the column order or step to the next page.

Search

The search input above the chip bar filters rows by display-name substring (case-insensitive). It submits as a plain form, so the search term lives in the URL (?q=acme) and a refresh / bookmark / share keeps the view. The Clear search link next to the input drops the search term while leaving the filter and sort state intact.

Drill-in

Clicking a customer's name opens the per-customer page — full history, manual payment-delay shifts, and the projected next 6 invoices with concrete dates. The per-customer baseline (mean / stddev) is computed automatically by the nightly worker; there is no manual override surface here. (Admins who need to override account-level baselines can do so on the Accounts → [account] drill-in.)

Excluded contacts

The "N excluded →" link in the page header opens the list of contacts an admin has manually excluded from customer analytics (refund-only contacts, internal transfers, etc.). Re-including a contact triggers the next nightly worker run to rebuild their baseline; until then the row reads "—" for cadence / spend / next purchase.