Fifty accounts, one ladder

Every client was invoiced on time. Roughly one in four paid on time. Nobody could say which ones, or why, until the month was already gone.

Shine Logistics moved freight for fifty-odd business clients on credit terms. Invoices went out promptly — this was not a billing problem. Payments came back on the client's schedule, not the contract's, and the only chase mechanism was a finance executive's memory and a spreadsheet she updated when she could. The founders' question: why does everyone owe us, and what would it take to be paid when we agreed?

An invoicing register (invoice date, amount, agreed terms), a bank statement that had to be matched to it by hand, and a client list with no memory of its own — no record of who had been reminded, when, or by whom. Three sources; zero joins.

-- the table that did not exist: one row per invoice, with its promise and its reality
SELECT i.invoice_no, i.client, i.amount, i.invoice_date,
       i.invoice_date + i.terms_days          AS due_date,
       p.paid_date,
       p.paid_date - (i.invoice_date + i.terms_days) AS days_late,   -- negative = early
       r.last_reminder, r.reminder_step
FROM invoices i
LEFT JOIN payments  p ON p.invoice_no = i.invoice_no
LEFT JOIN reminders r ON r.invoice_no = i.invoice_no;

Fifty accounts, placed by how late they paid. Then the same fifty, eight months later.

Placed on one axis, the accounts told on themselves. Lateness was not random — it was habitual. The same dozen accounts paid 30–90 days late every single month, and had been doing so long enough that nobody remembered it being otherwise. They were not struggling; they had simply learned that Shine did not chase. A handful of genuinely stretched clients sat among them, indistinguishable until someone looked.

The fix was not a threat. It was a cadence — polite, predictable, and impossible to forget.

Every invoice now climbed the same six rungs unless it was paid. Each rung had a day, an owner, a channel and a tone that was written down once and never improvised again. Click a rung.

Two kinds of flexibility travelled with the ladder. For clients who were genuinely stretched, part-payments against an agreed schedule — half now, half in fifteen days beat all of it in ninety. For clients who wanted to pay but kept losing invoices, a shorter cycle: fortnightly statements with every open invoice attached, so the question "which one?" stopped being an excuse.

# the tracker: who is on which rung today (the daily job, simplified)
inv["days_late"] = (today - inv["due_date"]).dt.days
inv["rung"] = pd.cut(inv["days_late"], [-99, -4, 0, 6, 14, 29, 999],
                  labels=["quiet", "courtesy", "nudge", "call", "statement", "escalate"])
due_today = inv[(inv.paid_date.isna()) & (inv.rung != inv.last_rung_sent)]
for _, row in due_today.iterrows():
    send(TEMPLATES[row.rung], to=row.client_contact, attach=statement_for(row.client))
    log_reminder(row.invoice_no, row.rung, today)        # the memory the business never had

The watch had a second audience. The rebuilt tracker became the client reporting itself — every account got a clean statement on a schedule instead of a call when things went wrong — and client satisfaction scores rose 30% over the same period. Being chased politely and predictably turned out to be something clients preferred to being chased badly and late.