Denise puts in a request for two weeks of vacation in September. Nine years in customer service, never asks for much, and she gives you six weeks notice because that is who she is. You approve it in about four seconds.
Then somebody has to cover her accounts, and the coverage conversation goes like this…
Halvorsen sends POs with their own part numbers, not yours. Denise translates them. Not from a cross-reference table, from memory, because the table was never built. Corbett's blanket releases have to go in as two lines, one for the material and one for the cut charge, or the invoice comes out wrong and their AP holds the whole thing for a month. Ferro-Line's POs list a ship-to that closed in 2023, and the real dock is the one on the second page of the contract. There is an account whose freight terms on the PO are wrong every single time, and Denise knows to price it off the agreement instead, and she has never mentioned this to anyone because it stopped feeling like a decision years ago.
None of it is in the ERP. None of it is in a work instruction. It is in her head, and its only external record is the pattern of ten thousand orders she has entered, which nobody has ever looked at that way.
WHAT THE COVERAGE PLAN ACTUALLY TELLS YOU
You are treating it as a scheduling problem. Two weeks, four accounts, somebody sits in her chair.
It is actually the only audit anybody at your company has ever run on undocumented order-entry rules, and it covers exactly the accounts somebody happened to remember in the meeting.
What it will miss: every rule that only fires a few times a year. Those do not come up in the meeting and they do not surface during a two week absence. They surface eleven months later as a credit memo nobody can explain.
The reflex is to say this should have been documented, and sure. But look at who you are asking. The person who holds the rules is the person carrying the most accounts, and the reason she holds them is that she is fast, and the reason she is fast is that she stopped writing things down years ago. Documentation is the one task on her list with no customer waiting on it. So it does not happen, and everyone involved is behaving rationally.
If you want to run this on your own accounts, we built the worksheet for it. Fourteen prompts for where these rules hide, a page per account, and a print view you can take into the coverage meeting. Nothing to sign up for.
Get the Account Rulebook Worksheet
It Is Not Just Your Company
Denise is not unusual. She is the statistical center of this business.
BLS puts median tenure for workers aged 55 to 64 at 9.6 years. For workers 25 to 34 it is 2.7 years. More than half of workers in their early sixties have been in the same place a decade or more. Two percent of workers in their late twenties have.
So the rules are held by people with nine years in, and they will be replaced by people who turn over every three. That is the whole problem in one sentence. Every undocumented rule is a bet that the person holding it stays longer than the rule matters, and the odds on that bet are getting worse, not better.
One note on timing. Work Institute's read of BLS quit-rate data finds the same rhythm every year of the series: turnover bottoms out in January, climbs through spring, and peaks in mid to late summer. Mid August is the top of the curve. Not a scare tactic, just the calendar.
This Week, Try This: The Delta Sweep
Most attempts at this problem start with a meeting where you ask people to write down their special rules. That produces the top ten and stops, because self-reported process knowledge is capped at what someone can remember while being watched.
Different approach. Do not ask the person. Read the record.
Every undocumented rule left a fingerprint. Somewhere there is a customer PO, and somewhere there is the order that got entered from it, and the two do not match. A part number was translated. A line was split. A ship-to was overridden. A freight term was ignored in favor of the contract. Each one of those gaps is a rule somebody applied silently, and unlike a memory, the gap is sitting in your systems right now.
So the raw material is pairs.
- Twelve months of order lines as entered. SKU, quantity, price, ship-to, freight terms, order date, customer.
- The incoming POs those orders came from. PDFs, email bodies, scans, whatever the mix is. Start with your top 25 accounts by line count, not by revenue, because line count is where the rules live.
- Anything already written down: ERP customer notes, comment fields, the sticky note taped to a monitor, that one spreadsheet from 2019.
The third input is optional and usually thin. Get the first two and the sweep works.
You are analyzing order-entry practice at a manufacturer or
distributor. You have two sets of files that describe the
same orders: the customer purchase orders as received, and
the order lines as they were actually entered into the ERP.
There may also be a file of existing customer notes.
Your job is to find the undocumented rules. A rule is any
place where what was entered differs systematically from
what the customer sent.
STEP 1 - PAIR THE RECORDS:
Match each incoming PO to the order lines entered from it,
using PO number, customer, date, and value. List any PO you
could not match and any order with no matching PO. Do not
guess at a pairing to make the numbers work.
STEP 2 - FIND EVERY DELTA:
For each matched pair, compare field by field and record
every difference:
- Part number on the PO vs. SKU entered
- Quantity or unit of measure on the PO vs. entered
- Line count on the PO vs. line count entered, including
any line that was split or merged
- Ship-to on the PO vs. ship-to entered
- Freight terms, payment terms, requested date
- Price on the PO, where stated, vs. price entered
- Anything added at entry that appears nowhere on the PO,
such as a charge line, a note, or a substituted item
STEP 3 - SEPARATE RULES FROM NOISE:
A delta that occurs once is probably a correction. A delta
that recurs is a rule. For each customer, group the deltas
by type and report:
- The rule, stated as an instruction a new hire could
follow on their first day
- How many times it fired in the twelve months
- Consistency: what share of eligible orders it was
applied to. Flag anything under 100 percent, because an
inconsistently applied rule is either two rules or an
error rate.
- The dollars that ran through it
- Whether the rule appears anywhere in the customer notes
file, if one was provided
STEP 4 - RANK BY WHAT BREAKS:
Sort every rule you found by exposure if it were missed
once, using the actual dollar value of the orders it
touched. Then flag separately:
- Rules that fired 3 times or fewer in twelve months.
These are the dangerous ones. They are real, they are
invisible, and they will not come up while somebody is
covering a two week absence.
- Rules that apply to only one customer contact rather
than the whole account
- Any rule that appears to contradict a term in the
customer notes file
RULES:
- Report only patterns you can point to in the files.
Quote the PO line and the entered line side by side for
every rule you claim.
- Never infer a business reason. State what happens, not
why. If the intent is unclear, say the intent is unclear.
- Do not merge two similar rules into one general rule.
Specific beats tidy here.
- If a customer's PO format changed partway through the
twelve months, say so and give the date. Rules often
changed with it.
What comes back is the document that was supposed to exist. Per account, the rules somebody has been applying for nine years, each one written as an instruction, with a count, a consistency rate, and a dollar figure next to it.
Two things to do with it in the first hour. Hand the top accounts to Denise and ask what is wrong, because she will find errors and her corrections are worth more than the original output. Then look hard at the rules that fired three times or fewer. That list is the one nobody has ever seen, and it is the reason a well run coverage plan still produces a mess in week two.
Whatever survives that review goes on the worksheet, one page per account. That is your rulebook, and it did not exist on Monday.
First run takes an afternoon. Most of that is pulling the POs.
An Honest Take
The standard prescription here is to document your processes, and I think it mostly fails, for a reason that has nothing to do with discipline. A process document is a project with no customer waiting on it, assigned to the busiest person on the team, and it goes stale the first time an account changes its PO format. I have watched companies spend a quarter building a beautiful order-entry manual that nobody opened in year two. A document nobody reads is tribal knowledge with a filename.
What I actually believe: the rules have to live in whatever reads the incoming order, and for most companies that is currently a person. Your ERP is not the answer here, and this is the part I think gets confused. The ERP never sees the PO. It sees what somebody typed after they applied the rule. Which means the ERP cannot hold the rule, it can only hold the result, and the judgment stays in the person's head where it started.
The thing that can hold the rule is a system that sits where Denise sits, between the inbox and the ERP, reading the document before it becomes an order. That is a specific and fairly recent category of software, and it is what we build, so discount me accordingly. But the logic holds no matter who you buy it from. Write the rules down as extraction, not as the destination, then put them somewhere that fires them on every order whether Denise is at her desk or in Michigan.
And one smaller thing. Your coverage plan for a two week vacation is the most honest risk assessment anybody at your company produces all year, and it gets thrown away the day she comes back. Keep it. Compare next year's to this one. The rules that show up on both lists have been undocumented for two years running, and now you know that on purpose instead of by accident.
The Bottom Line
Denise coming back in two weeks does not close this. The vacation just showed you the shape of it, at low stakes, with the person still available to answer questions. That combination will not be true the next time.
The rules exist whether or not anybody wrote them down. Right now their only backup copy is a person's memory and a pattern in your order history that nobody has read. One of those two you can do something about this month.
The rules do not have to live in somebody's head.
Halvorsen's part number mapping, Corbett's split lines, the freight term that gets read off the contract instead of the PO. Those are business rules. They are just business rules stored in the wrong place.
Y Meadows runs order entry with the rules encoded: inbox pickup to ERP posting, matching customers and SKUs, applying your pricing, freight, and credit logic on every order regardless of who is working that day. Want to see it run on your messiest account?