How to Hand Off Bookkeeping Without Losing Visibility

Why this matters

The handoff that goes wrong is almost never the one where someone steals. It is the one where the books quietly become someone else's file. Statements arrive on time, questions get answered pleasantly, and eighteen months later you cannot log in without asking, cannot say how job cost is defined, and cannot tell whether the margin you price against is the one your books actually produce. The visibility did not get taken. It was never written down, so it dissolved.

This procedure is not a task list. It is one artifact, built field by field, and the fields are the handoff. Where a field is blank, the person doing the work will fill it in by default, correctly from their point of view and possibly wrong from yours.

The artifact: the handoff register

One document, kept where both of you can read it, revised when it changes. Build it in this order, because each field constrains the ones after it.

Field 1: scope, stated as two lists

Not a job description. Two explicit lists: what transfers, and what does not. The second list is the one that matters, because unlisted work drifts toward whoever is closest to it.

Transferred, typically: reconciliation, transaction coding under the rule, payables entry and scheduling, receivables aging and the reminder cadence, payroll processing, statement production, the exceptions list.

Not transferred, and say so in writing: signing authority, customer credit decisions, write-offs, the coding rule itself, changes to the chart of accounts, the definition of a completed job, and any tax position. Tax positions route to the accountant who signs the return, not to the bookkeeper and not to you.

If you skip this field: everything unlisted defaults to the bookkeeper, including the two or three items that are actually your pricing model expressed as bookkeeping.

Field 2: the coding rule, written out in full

One paragraph, in your words, that a competent stranger could apply. What counts as direct job cost, what counts as overhead, how a mixed purchase is split, how an owner draw is distinguished from a business expense.

If you skip this field: ambiguous transactions still get coded. They get coded consistently, in whatever direction looks conservative to a person who has never been on one of your jobs, and consistency will make the resulting error invisible in every trend you look at afterward.

Field 3: the access register

A table: every financial account, who holds access, at what level, and which credential the owner holds independently. Owner access is not a courtesy copy. It is a separate login, on your own device, that works without anyone else's cooperation, tested quarterly.

Include read-only bank access as its own row. A bookkeeper needs entry access to the accounting file; you need to be able to look at the bank without going through the accounting file at all. Those two views disagreeing is the single most useful signal you have, and you cannot see the disagreement if both views come from the same person.

If you skip this field: you find out on the day of a dispute, an illness, or a resignation, which is the worst possible day to discover that your history lives behind someone else's password.

Field 4: cadence and deliverables

What arrives, in what form, by which working day. Name the artifacts: the close packet, the aging, the exceptions list, the variance sheet. A cadence without a named artifact becomes "I will send you something when it is ready."

Set the date on working days, not calendar days, so a month ending on a weekend does not silently push the whole thing.

If you skip this field: the deliverable degrades to whatever is convenient, usually a login and an invitation to look around, which is not a deliverable at all because it puts the assembly work back on you.

Field 5: escalation, and what stops the work

Two rules, both stated as conditions rather than preferences.

Stop and ask when a transaction cannot be placed under the coding rule, when an account will not reconcile, when a completed job has no invoice, or when a document is missing. None of these get a provisional treatment.

Escalate immediately, not at close, for anything that looks like a duplicated payment, an unrecognized vendor, or a payroll entry for someone who does not work here. Immediate means the same day, because the recovery options on all three narrow with time.

If you skip this field: unplaceable items get placed anyway. A holding account fills up, or worse, does not, because everything got a plausible home.

Field 6: authority limits

Written as thresholds you set in your own terms, expressed against something stable. A workable starting pattern: recurring vendors already on the approved list are paid on schedule without asking, anything to a new vendor is released by you, and any single payment larger than one average invoice comes to you regardless of vendor. Refunds and write-offs are always yours.

State the limits as numbers relative to your own volume rather than leaving them to judgment, then tune them. A limit nobody can state precisely is not a control.

If you skip this field: authority is inferred from what has been tolerated, which grows one exception at a time and is never reviewed.

Field 7: definitions

Three, minimum, because these move your statements more than any other wording in the register: what makes a job complete for revenue purposes, what makes a cost direct rather than overhead, and what distinguishes an owner draw from a business expense. Write them once, and treat a change to any of them as a decision with a date on it, because it breaks comparability with every prior month.

Field 8: the reversal clause

What happens on the first day of an exit, whether the exit is friendly or not. Where the files live, in what format, who transfers what, and how quickly. Write this while everyone likes each other. It costs ten minutes now and it is the field you will be most grateful for.

The first 90 days: parallel run and sampling

Do not go from doing it yourself to reading a statement. Run a sample check every month, and let the sample tell you when to stop.

Pull 20 transactions at random from the closed month and re-code them yourself against your own written rule, without looking at how they were coded. Then compare.

• Mismatch rate above 10% of the sample: rewrite the coding rule. Do not correct entries one at a time. At that rate the rule is ambiguous, and correcting entries treats a definition problem as a diligence problem.
• Mismatch rate at or below 5% of the sample for two consecutive months: move to quarterly sampling.
• Anything between: stay monthly and keep sampling.

The unit of analysis is the sample, not the month's whole transaction count, and the comparison is against your rule as written, not against what you meant.

Worked example: four months of sampling

A shop hands off in January to an outside bookkeeper with all eight fields of the register filled in, but field 2 is weak: the coding rule runs to two loose sentences, both of which use the word "job" without defining it.

Month 1. Sample of 20, mismatches 4, which is 20%. Above the 10% gate, so the rule gets rewritten rather than the entries corrected. Reading the 4, three of them are the same shape: consumables bought on a day with several tickets. The rewritten rule adds the split method for a mixed purchase.

Month 2. Sample of 20, mismatches 2, which is 10%. That is at the gate, not above it, so no second rewrite, and the register records that the tie was resolved by not rewriting. It stays on monthly sampling because 10% is well outside the at-or-below-5% band. The two mismatches are unrelated to each other, which is the signal that the systematic problem is gone.

Month 3. Sample of 20, mismatches 1, which is 5%. That is at or below the 5% band, but the band requires two consecutive months, so sampling stays monthly.

Month 4. Sample of 20, mismatches 0. Months 3 and 4 are the two consecutive months, so sampling moves to quarterly.

Total owner time across the four months: 20 transactions re-coded at roughly 90 seconds each is about 30 minutes per sample, so about 2 hours over four months, plus perhaps 40 minutes rewriting the rule after month 1. Under 3 hours to establish that the coding you price against matches the coding you intended.

Notice what the sequence proves and what it does not. It proves the rule is now unambiguous enough that two people apply it the same way. It does not prove the rule is right. A rule that sends the wrong things to overhead, applied perfectly, produces a 0% mismatch rate and a wrong gross margin, which is why the register keeps the rule on your side of the line permanently rather than declaring it settled after month 4.

What losing visibility actually looks like

It is gradual and each step is reasonable. The exceptions list stops arriving because there were no exceptions this month, and nobody notices that it also did not arrive the month after. A question gets answered with a number instead of a report, so you stop opening the report. A password gets reset during a software migration and only one person writes down the new one. The close packet becomes a link to a live report, so the version you signed no longer exists.

The tell is not a missing document. It is the day you want to check something and your first instinct is to ask rather than to look. When you notice that instinct, go back to the access register, log in to all of it yourself, and pull a sample. Visibility is a habit with a maintenance interval, not a state you reach once.

References

• U.S. Small Business Administration (SBA), small business internal control and recordkeeping practices
• Generally Accepted Accounting Principles (GAAP), consistency of accounting treatment across periods
• See related: What the Owner Must Own and What the Bookkeeper Owns; The Monthly Close SOP; When the Bank Balance Lies: Reconcile Discipline