Skip to content
The Escalation Desk

case 3 of 3 · unit economics

Why did you pick the expensive courier? An honest answer, zone by zone

sev3Harvestline Groceryopened May 4, 11:05 AM ETdetected via: Customer finance lead flagged provider mix on the monthly invoice

01Symptom

Priya Raman, Harvestline Grocery’s finance lead, flagged one line on the monthly invoice: 37.9% of orders route to FleetX, the priciest provider on the account. Fair question — and this case answers it with arithmetic, not defense. Nothing is broken; a customer is asking us to justify our own routing with their money.

The honest answer splits. In part of the city the premium is buying something the invoice cannot see; in the rest it is a config decision nobody re-checked. The case keeps those two separate, because blending them would flatter us and cost the customer money.

Provider mix over the trailing 30 days
providerdeliveriesshareon-timecost / delivery
FleetX5,44537.9%95.8%$16.62
Parcelet4,47431.1%91.6%$12.16
DashNow3,76526.2%91.8%$10.73
CourierCo6974.8%93.5%$12.96

Trailing 30 days, 14,381 deliveries. The FleetX row is the eyebrow-raiser: highest share, highest on-time, highest cost.

02Evidence trail

Before any arithmetic, the router’s receipts — one order per zone, every quote it saw at dispatch time. The router’s score is quote price plus an expected-lateness cost built from each provider’s trailing on-time in that zone (A4). Cheap and unreliable loses to expensive and reliable downtown. And in the third panel, watch the quote that clears the reliability floor at a lower price and still loses. That is not a score. That is a pin.

HVL-20260408-976978701Apr 8, 1:03 PM ET
Quotes received for order HVL-20260408-9769
providerpriceetaon-time
DashNow$8.2245 min77.0%
Parcelet$9.5435 min86.5%
CourierCo$11.4638 min87.6%
FleetX$14.5234 min97.0%← dispatched
HVL-20260408-810678704Apr 8, 3:10 PM ET
Quotes received for order HVL-20260408-8106
providerpriceetaon-time
DashNow$8.0640 min77.5%
Parcelet$9.6040 min87.2%← dispatched
CourierCo$11.1439 min87.8%
FleetX$14.9940 min97.6%
HVL-20260409-800078757Apr 9, 6:05 PM ET
Quotes received for order HVL-20260409-8000
providerpriceetaon-time
DashNow$8.6041 min95.5%
Parcelet$10.6647 min96.9%
CourierCo$12.2142 min96.2%
FleetX$14.2643 min97.8%← dispatched
HVL-20260409-142478759Apr 9, 4:24 PM ET
Quotes received for order HVL-20260409-1424
providerpriceetaon-time
DashNow$10.5048 min95.9%
CourierCo$12.7453 min94.3%
Parcelet$12.8158 min97.9%
FleetX$16.0545 min97.1%← dispatched

On-time is each provider’s trailing rate in that ZIP at quote time (A4).

The arithmetic, zone by zone. For each zone: what FleetX costs, what each alternative would save on the invoice, and what that saving costs back in refunds, support contacts, and SLA credits. The economics are illustrative (A7/A14) and every constant is printed below the tables:

78701Downtown Austinpremium justifiedFleetX 1,143/mo · avg all-in $16.18 · on-time 96.8% · zone spend $21,550
Alternative-provider economics for zone 78701
alternativeavg quoteon-timespend savedlates addedrefundssupportsla creditnet / mocall
Parcelet$11.8285.3%$4,983132$1,795$799$2,155$235below materiality
DashNow$10.3174.3%$6,709258$3,509$1,561$2,155-$515keep FleetX
CourierCo$12.9186.0%$3,738124$1,686$750$2,155-$854keep FleetX
78704South Congresspremium justifiedFleetX 1,370/mo · avg all-in $15.84 · on-time 96.0% · zone spend $28,061
Alternative-provider economics for zone 78704
alternativeavg quoteon-timespend savedlates addedrefundssupportsla creditnet / mocall
Parcelet$12.0686.5%$5,179131$1,782$793$2,806-$202keep FleetX
DashNow$10.4775.3%$7,357284$3,862$1,718$2,806-$1,030keep FleetX
CourierCo$12.9186.2%$4,014135$1,836$817$2,806-$1,445keep FleetX
78757Crestviewpremium not justifiedpin activeFleetX 1,538/mo · avg all-in $16.29 · on-time 96.0% · zone spend $25,087
Alternative-provider economics for zone 78757
alternativeavg quoteon-timespend savedlates addedrefundssupportsla creditnet / mocall
Parcelet$12.2995.7%$6,1524$54$24$0$6,073switch
DashNow$10.6792.3%$8,64456$762$339$2,509$5,035switch
CourierCo$13.3594.8%$4,52218$245$109$2,509$1,659switch
78759Great Hillspremium not justifiedpin activeFleetX 1,391/mo · avg all-in $18.19 · on-time 94.6% · zone spend $24,973
Alternative-provider economics for zone 78759
alternativeavg quoteon-timespend savedlates addedrefundssupportsla creditnet / mocall
DashNow$12.4491.8%$7,99839$530$236$2,497$4,735switch
Parcelet$14.0696.1%$5,7450$0$0$2,497$3,248switch
CourierCo$14.8193.7%$4,70212$163$73$2,497$1,968switch

Illustrative economics (A7/A14): 40% of added lates refund at $34 each; 55% raise a $11 support contact; when the projected zone on-time falls below the 95% floor, an SLA credit of 10% of zone spend applies; a switch only counts as savings above $250/mo materiality. All table figures are per month.

cost and on-time by provider and zoneops warehouse · illustrative schema (A8)
-- Realized share, on-time, and cost by provider and zone, trailing 30 days.
-- On-time per the A6 definition: delivered within 15 minutes of promise.
SELECT
  zip,
  provider_id,
  COUNT(*) AS deliveries,
  ROUND(100.0 * COUNT(*) / SUM(COUNT(*)) OVER (PARTITION BY zip), 1)
    AS zone_share_pct,
  ROUND(AVG((delivered_at <= promised_at + INTERVAL '15 minutes')::int), 3)
    AS on_time_rate,
  ROUND(AVG(cost_cents)) AS avg_cost_cents
FROM deliveries
WHERE account_id = 'acct_harvestline'
  AND created_at > now() - INTERVAL '30 days'
  AND delivered_at IS NOT NULL
GROUP BY zip, provider_id
ORDER BY zip, deliveries DESC;

03Diagnosis

Two mechanisms produce the 37.9%, and they deserve opposite treatment.

Merit. In 78701 and 78704, scored routing picks FleetX because every cheaper provider’s trailing on-time in those ZIPs sits far below it — the cheapest quotes downtown come from a provider running 74.3% on-time against FleetX’s 96.8%. Even the best alternative projects those zones to 85.6% and 86.1% — both through the 95% SLA floor. The router is doing its job.

Config debt. In 78757 and 78759, FleetX is not winning a score — it is pinned. The pin dates to a February decision after a bad DashNow week, and nobody revisited it. Today Parcelet runs 95.7% and 96.1% trailing on-time in those ZIPs, and the pin just burns money — $10,808 a month of premium that is not justified.

what the operator did not do

  • Defend the router wholesale — the job is honest accounting, not advocacy for your own routing.
  • Blend the four zones into one average — a blended number would show a modest premium and hide both truths.
  • Recommend switching the dense core to look responsive — those paper savings reappear as refunds and SLA credits.

04Resolution & prevention

where the customer is right

$10,808 /month buying nothing

In 78757 (Crestview) and 78759 (Great Hills), the FleetX premium was not purchasing reliability those zones needed: net overspend of $6,073 and $4,735 per month respectively, after charging the switch for every added late, refund, support contact, and SLA credit. Roughly this much of the premium was buying nothing, and the finance team was right to flag it.

Recommended change: Remove the February FleetX pin in 78757 and 78759 and let scored routing pick per zone (Parcelet in 78757, DashNow in 78759); keep FleetX for baskets over $120 or sub-45-minute windows. Projected savings: $10,808/mo.

Prevention is a calendar rule, not a dashboard: routing pins now carry a quarterly review and expire unless re-justified with current numbers. A pin is a bet that the router is wrong; the February bet was right for a week and wrong for a quarter. The query below is the standing check — zones where we dispatch above the cheapest quote even though the cheapest provider is clearing the reliability floor.

finding money left on the tableops warehouse · illustrative schema (A8)
-- Stale-pin detector: orders dispatched above the cheapest quote even though
-- the cheapest provider is clearing the 95% on-time floor in that zone (A4).
WITH ranked AS (
  SELECT
    order_ref,
    zip,
    provider_id,
    price_cents,
    chosen,
    MIN(price_cents) OVER (PARTITION BY order_ref) AS cheapest_cents,
    FIRST_VALUE(trailing_on_time) OVER (
      PARTITION BY order_ref ORDER BY price_cents
    ) AS cheapest_on_time
  FROM quotes
  WHERE account_id = 'acct_harvestline'
    AND ts > now() - INTERVAL '30 days'
)
SELECT
  zip,
  COUNT(*) AS orders,
  ROUND(AVG(price_cents - cheapest_cents)) AS avg_overpay_cents,
  SUM(price_cents - cheapest_cents) AS overpay_cents_30d
FROM ranked
WHERE chosen
  AND price_cents > cheapest_cents
  AND cheapest_on_time >= 0.95
GROUP BY zip
ORDER BY overpay_cents_30d DESC;

the outreach — same facts, two altitudes

To: Tom Delgado (Eng, Harvestline)

Subject: Routing analysis behind the provider-mix answer for finance

Hi Tom, Backup for the note I sent Priya. Two different mechanisms are producing the FleetX share, and they deserve different treatment. 78701/78704: scored routing (quote price + expected-lateness cost from trailing zone on-time) picks FleetX on merit. The cheapest quotes there come from providers running 74.3% on-time vs FleetX at 96.8%; switching breaches the 95% zone SLA floor. Leave it alone. 78757/78759: FleetX is not winning on score — it’s pinned. The pin dates to the February DashNow wobble and was never revisited. Best alternatives run Parcelet at 95.7% in 78757 and DashNow at 91.8% in 78759; the projected mix books $2,497/mo of SLA credits, already charged inside the net figure. Proposed change (not yet applied): remove the pin in 78757/78759 and let scoring run. Projected ~$10,808/mo saved, on-time impact within noise. Happy to walk through the query set.