Writing Decision Rules So the Office Stops Asking You
Why this matters
Handing the CSR the authority to discount does not stop her calling you. She calls because she does not know what you would say, and guessing wrong costs her more than interrupting you does. A decision rule is your answer written down in advance: for this kind of situation, do this, within this limit, except in these cases, and log it. A good rule lets the office make the same call you would make, at the moment it comes up, and gives you a record to check whether it was the right call. A bad rule is either so vague it gets ignored or so rigid the first odd case breaks it, and then everyone goes back to asking.
Who gets the authority is a separate question, answered first; see related: Decision Rights: Who in the Shop Decides What. This article is about what you hand them along with it.
A rule is not a policy
A policy states an intention: "we take care of our repeat customers." A rule states an action: "a customer with two or more paid visits in the last 24 months gets up to 10 percent off labor on request, once per 12 months." The policy tells the CSR what you value; the rule tells her what to type. Office staff ask you questions because they have the first and not the second.
Every rule worth writing has six parts. Write them in this order, because each one depends on the one before.
Step 1: Write the default answer as an action
Start with what happens in the ordinary case, as a verb the person can do without interpretation. "Offer next available slot, no charge for the visit" is a default. "Handle it fairly" is not.
Why first: the default is what gets used 80 to 90 percent of the time, and if it is wrong, no amount of careful exception-writing rescues the rule. What breaks if skipped: people write the exceptions first, because the exceptions are what they argued about, and end up with a rule that says what not to do and never says what to do.
Step 2: Put a number on the limit
The limit is the edge of the authority, and it has to be a number or a countable condition: a percentage of the invoice, a count of uses per customer per year, a number of days since the original job. Use units that do not go stale with inflation, so percentages and counts rather than amounts.
What breaks if skipped: "a reasonable discount" means 5 percent to one person and 25 to another, and the first time someone reads it generously you take the authority back, which teaches the office that written rules are provisional.
Step 3: List the exceptions, and close the list
Exceptions are the cases where the default does not apply. Write them as a short, closed list: "does not apply to after-hours emergency rates, to equipment sales, or to a ticket already under a maintenance plan price." Then add one line: anything not on this list follows the default.
Why closed: an open list ("except where inappropriate") hands the judgment back to the person and puts you back on the phone. What breaks if skipped: the CSR treats every unusual case as a possible exception and escalates it.
Step 4: Name the escalation, with a clock
For anything outside the limit or genuinely ambiguous, say who decides and how fast. "Over the limit, ask the office manager; if she has not answered in 2 hours, tell the customer we will call back tomorrow by noon." The clock matters more than the name: an escalation that can sit unanswered all afternoon leaves a customer on hold and teaches staff to bypass the chain and call you directly.
Step 5: Make every use leave one log line
Each time the rule is used, one line goes somewhere you can count: a reason code on the invoice, a tag on the job, a line in a shared sheet. Date, ticket, which branch of the rule (default, exception, escalated), and the person. This is the part owners leave out because it feels like paperwork, and without it Step 6 is impossible.
What breaks if skipped: you will review the rule from memory, which means from the two cases that annoyed you, and you will change it in response to anecdote.
Step 6: Set a review trigger and a cadence
Write down two things: when you look at the log on a schedule (monthly for the first three months, then quarterly is a sound starting point) and what number forces an early look. A discount rule might say "review early if it is used on more than 12 percent of tickets in a month." A warranty rule might say "review early if callbacks under the rule exceed 4 percent of completed jobs in a month." The trigger catches the rule drifting before the scheduled review would.
Three rules, written out
The discount rule (CSR). Default: a customer with two or more paid visits in the last 24 months, or aged 65 or over by their own statement, gets up to 10 percent off labor on request. Limit: 10 percent of labor, once per customer per 12 months. Exceptions: not on after-hours emergency rates, equipment sales, or maintenance-plan tickets. Escalation: anything over the limit to the office manager, 2-hour clock. Log: reason code DISC-REPEAT or DISC-SENIOR on the invoice. Review trigger: used on more than 12 percent of tickets in a month.
The warranty callback rule (office manager). Default: the same complaint on the same equipment within 90 days of our repair is booked as a no-charge callback, next available slot, assigned to a different tech from the original where possible. Limit: covers our labor and parts we supplied; does not cover a new, unrelated failure on the same unit. Exceptions: customer-damaged equipment, equipment we advised replacing in writing and the customer declined, and anything past 90 days. Escalation: "is this the same failure?" disputes go to the lead tech on site, and he calls it before leaving. Log: job type CALLBACK linked to the original ticket. Review trigger: callbacks over 4 percent of completed jobs in a month.
The "different tech" clause is there because the original tech has a stake in classifying the failure as new. Rules should not ask anyone to judge their own work.
The schedule override rule (dispatcher). Default: a no-heat or no-cool call with a person over 70, an infant, or a medical device in the home may bump a scheduled maintenance visit. Limit: no customer is bumped more than once in 30 days. Exceptions: never bump another urgent call; never bump a job with a crew already en route. Escalation: if no maintenance visit can be bumped within the limit, the dispatcher calls the lead tech for an after-hours slot. Log: tag BUMPED on the moved job, with the reason. Obligation that comes with it: the bumped customer is called within 15 minutes and offered the next two open slots.
Worked example: the first two discount reviews
Month one. The shop ran 212 tickets. The log showed the discount used on 31 tickets, 14.6 percent of 212, over the 12 percent trigger. Six requests went to the office manager as over-limit; she granted two.
Reading the log by reason code: 22 of the 31 were DISC-REPEAT, 9 were DISC-SENIOR. The owner pulled the 22 repeat cases and checked visit history. Nine of the 22 had only one prior paid visit, which the rule does not cover. The CSR had read "repeat customer" in her head as "has been here before," which is what the old verbal policy had said. That is a definition failure, not a judgment failure; the written rule already said two visits, but the CSR was working from memory of the old wording.
Remove those 9 and the rule as written was used on 22 tickets, 10.4 percent of 212, under the trigger. So the rule's limit was fine. The fix was to put the visit count on screen: the office added a visit-count field to the customer record so the CSR sees it before she offers anything.
Month two. 198 tickets, 19 discounts, 9.6 percent of 198, under the trigger. Three over-limit escalations, one granted. Two of the 19 fell under exceptions (both maintenance-plan tickets discounted anyway), so the exception list got a line on the booking screen too.
What the owner did not do: he did not lower the limit to 5 percent in month one, which was his first instinct on seeing 14.6. The log showed the overrun came from a misread definition, not from a limit that was too generous. Changing the limit would have punished the right uses to fix the wrong ones. That is the job the log line does: it tells you which part of the rule failed.
How a rule rots, and how to catch it
The exception list grows by word of mouth. Someone grants an exception once, it becomes known, and it is applied again without ever being written. Catch it in the log: count the uses that do not match any written branch. More than one or two a month means an unwritten exception exists; either write it into the rule or stop it.
The escalation goes stale. The office manager leaves, the rule still names her, and requests go nowhere. Name roles, not people, and check every rule's escalation line when anyone changes seats.
The rule outlives its reason. A rule written when you were short of techs (bump maintenance freely) becomes wrong when you are not. Put one line of reasoning in each rule, "because we are short of techs this season," so the person reviewing it later can tell whether the reason still holds. A rule with its reason attached can be retired; a rule without one just accumulates.
You override it. If you step in on a case the rule covers, either the rule is wrong and you change it at the next review, or you were wrong and you leave it alone. Overriding it ad hoc and leaving it written is the fastest way to teach the office that the rules are decoration, and they will go back to asking you.
What changes the answer
High-stakes, low-frequency decisions should not get rules at all. Firing a customer, writing off a large balance, a refund on an installation: these come up rarely and every case is different enough that a rule would either be too broad to be safe or so narrow it never fires. Leave them as named decisions with a consult, not as rules.
A brand-new person in the seat should get the same rule with a lower limit for the first 60 days. The rule is the same; the trust level is not yet.
Rules that touch consumer protection or warranty law (a written warranty on a product, a cancellation right on a contract sold in the home) need the legal text checked before they are written as shop rules. The rule can promise more than the law requires, never less.
How to know the rules are working
Three numbers, checked at each review. The share of rule decisions that were escalated should settle below about 1 in 5. The number of uses matching no written branch should be near zero. And your own interruption count on those topics should fall; if the office still calls you about discounts after a rule exists, ask them which case the rule did not answer, and that case becomes the next line.
References
- See related: Decision Rights: Who in the Shop Decides What
- See related: Delegating the Decision, Not Just the Task
- See related: We Made a Rule After One Bad Job
- See related: Second-Order Effects of a Policy Change