N NoBullshit.wiki 中文

Coffee customer service - "tastes different from last time" complaint routing SOP - five intake fields, three routing lines, deadlines on every loop

Works with:Claude / GPT-4 classCategories:Coffee ShopLevel:Intermediate Updated

You are the service-quality supervisor of a coffee shop.
Turn this "flavor inconsistency complaint" SOP into an
archivable document: keep the structure, leave deadlines
and trigger lines as [ ] for me to set by staffing, and
lock everything else.

State the SOP's stance up front: a customer's "different
from last time" is not a misunderstanding to rebut, it is
a free QC sensor. The agent does not argue, does not
diagnose the cause on the spot, and does exactly three
things - log every field, route to the right line, chase
the loop closed.

[Intake: five fields, none optional (a missing field is
logged as "not captured")]
1. Drink and cup type
2. Bean lot: check the roast lot number on the bag in
   use during that shift
3. Barista on bar: who made the cup
4. Grinder setting: the machine's most recent calibration
   record for that shift
5. The customer's own words: verbatim excerpt (e.g.
   "sour," "thinner than before," "milk drowned it") -
  never translated into the agent's phrasing

Intake script:
"Got it - let me note down what you're tasting. We have
no record of a recipe change, but the difference you're
tasting is real. Let me check the beans and the equipment
from the last couple of days and get back to you within
[24] hours."
- Neither "we never change" nor "must be a new lot" -
  both sentences slam the door on the investigation that
  follows.

[Routing: by the record; one complaint may take more
than one line]
- Line 1, bean lot (lot changed within the last [7]
  days / other customers on the same lot said similar):
  to purchasing - verify whether the lot changed and
  whether the supplier flagged any process change;
  deadline [ ] hours; return field: one-sentence lot
  conclusion
- Line 2, equipment (grinder setting far from its last
  calibration / unlogged adjustments between the
  complaint and the last calibration): to maintenance -
  recalibrate, check water temperature; deadline [ ]
  hours; return field: readings before and after
- Line 3, person (no lot change, no equipment gap,
  description pointing at mouthfeel): to the trainer -
  compare that day's output against blind-tasting
  records; deadline [ ] days; return field: whether the
  barista enters the coaching watch list
- No assignable line: defaults to the manager, who
  assigns; "investigated, nothing found" straight-closing
  the ticket is banned

[Closure: callback and escalation]
- After all routed lines return, the agent calls back
  within the promised window: what was found + what was
  done + what's different next visit - all three from
  the return fields, with no "we'll keep improving"
  bolted on
- Escalation: the same drink with the same description
  [3] times in one week - regardless of what the three
  lines each concluded - auto-escalates to the manager
  for a cross-cutting review, laying that week's bean,
  equipment, and scheduling records on one timeline
- Weekly ledger report: complaint count / routing
  distribution / on-time return rate per line - when
  the return rate falls below [80%], the SOP's
  execution is the problem to fix before the coffee

Formatting rules:
- Keep the three sections: intake, routing, closure
- Each line's deadline and return field as one row of a
  horizontal table
- No "customer satisfaction first" opening or closing

Why it is written this way

The five-field record is the fulcrum of the whole SOP. The truth of "different from last time" hides in four objective coordinates (drink, lot, person, grinder setting) and one subjective one (the customer's verbatim) - drop any one and the routing line is guesswork. Logging the verbatim is the easiest step to skip: the moment an agent translates "sour" into "a preference issue," the complaint degrades from QC signal to mood-soothing and all three lines go dark downstream. The verbatim is the only thing comparable across complaints.

Each routing line carries a hard return field and deadline because the way routing actually dies in a store is routing into a vacuum: the complaint goes to purchasing, purchasing glances and never answers, the customer never gets a callback, the agent stops daring to ask - and eventually everyone stops routing at all. Return fields compress each line's output to one-sentence scale (lot conclusion, calibration readings), cheap to produce but giving the loop something checkable; putting the return rate into the weekly report puts the SOP's own execution under measurement too - fix the process before debating the coffee.

The escalation trigger "same drink, same week, three times" rather than "when the manager feels it's a lot" is the same design as the spill checklist's review line: clustering is the most reliable signal of a lot problem - a single complaint can be one palate drifting, but three matching descriptions on one drink is almost certainly beans, equipment, or training having moved. And banning the "investigated, nothing found" close exists because that's precisely the most suspicious shape: all three lines reporting clean while the customer tastes a difference means one line's checking standard itself is leaking.

Variables

Variable Notes
Per-line deadlines Return deadlines for each routing line, set by staffing and shift pattern
Escalation trigger Same-drink same-description count per week that triggers the manager review
Callback window The reply time promised in the intake script; must be shorter than the sum of line deadlines

flavor complaintsticket routingclosure management