The discount nobody approved

Every quote was priced by a human, negotiated by another, and closed by a third. Nobody could say what a product actually sold for.

The platform sells medical equipment — diagnostic imaging, patient monitoring, surgical instruments, consumables, refurbished devices — to small and mid-size hospitals that buy on quotation. A hospital asks; a category team sources and costs it from vendors; a pricing team sets a quote; sales negotiates; finance invoices. Four hands, four judgements, and no line in any system that said this is the price we meant to charge. The founders' question came from the board: are we losing deals on price, or winning them at any price? The honest answer was: both, and nobody could tell which was which.

Three systems again. The quotation workflow — every request stamped as it moved Quotation Requested → Assigned to Pricing → Assigned to Category → Quotation Generated, or died as Cannot Procure — with a turnaround clock on each hand-off. The CRM, where sales logged the negotiation and, when a quote was lost, a Not Accepted Reason — free text, including "price". And the accounting ledger, where vendor bills carried cost and invoices carried the selling price, so margin could be computed per line for the first time.

-- margin per invoice line: what we paid the vendor against what the hospital paid us (real shape)
SELECT i.date                 AS bill_date,
       i.payment_terms_label,
       li.sku, li.product_category,
       li.quantity * v.bcy_rate  AS cost_price,          -- vendor bill, base currency
       li.quantity * li.rate     AS selling_price
FROM zoho.invoices i
JOIN zoho.invoices_line_items li ON li.invoice_id = i.invoice_id
LEFT JOIN zoho.bills_line_items v ON v.sku = li.sku AND v.bill_date <= i.date;

# then, in pandas — the number nobody had seen per line before
df["margin"] = (df.selling_price - df.cost_price) / df.selling_price
df = df.replace([np.inf, -np.inf], np.nan)              # a ₹0 selling price is a data bug, not a 100% discount

List price is a story you tell yourself. Pocket price is what reaches the bank. The gap between them had five steps, and only one was a decision.

The waterfall at the top of this page is the TRACE. It follows one average deal from the price list to the money that actually landed, and names every step where value fell away: the category discount (published, deliberate), the negotiated discount (sales, unrecorded, the largest leak), freight and installation absorbed to "get the deal done", and the cost of extended payment terms — 60 and 90 days granted as a closing gesture, which is a discount wearing a calendar. The framework is McKinsey's pocket-price waterfall; the finding was local: only the first step had an owner.

Small hospitals were getting the deepest discounts. Large ones were paying closer to list. And the deepest discounts did not win more often.

Cohorting every quote by pocket price and outcome produced the chart above, and the two findings that changed the conversation. First, the pocket-price band was 3.1× wide for identical products: some hospitals paid under half of list, some nearly all of it, and deal size explained almost none of it — long relationships and whichever rep picked up the phone explained most. Second, and this is the one that matters, discounting deeper did not lift the win rate. Past a point, quotes were lost anyway, for reasons the CRM had been recording all along: delivery time, a missing demo, a competitor's brand. The discount was buying nothing.

Not a price list — a corridor per category: a floor that protects margin, a target that wins, a ceiling that flags. Pick a quote.

# the corridor, fitted per category from won and lost quotes (simplified)
bands = quotes.assign(band=pd.cut(quotes.pocket_pct, bins=range(40, 105, 5)))
curve = bands.groupby(["category", "band"]).agg(win_rate=("won", "mean"), n=("won", "size"), margin=("pocket_margin", "median"))

def corridor(cat):
    c = curve.loc[cat]
    floor  = c[c.margin >= MIN_MARGIN[cat]].index.min().left          # protects margin
    target = c[c.n >= 12].win_rate.idxmax().left                       # where it actually wins
    ceil   = c[c.win_rate >= 0.25].index.max().right                   # beyond this, quotes die
    return floor, target, ceil

# the rule in the quote tool: below floor → approval; above ceiling → review; outside either → logged, always

The watch was three numbers on one page, refreshed from the same daily job that ran the collections tracker in case 01: conversion, the pocket price index, and — the one that kept the corridor honest — the share of quotes going out outside it. When that share crept up, it was never the corridor that was wrong; it was a new rep, or a vendor cost that had moved. The monthly read said which.