Boxes Nobody Opens, Files Nobody Deletes. How to Decide What Your Retention Schedule Keeps

In most organizations past about thirty people, retention decisions get made twice: once when someone writes a policy, and continuously after that by whoever is standing in front of the cabinet or the shared drive with a deadline. The second set of decisions is the one that governs. So the useful work is not drafting a longer policy. It is building enough shared judgement that the person deciding in the moment reaches the same answer the policy would have, without reading it.

That judgement rests on two questions per record, asked in order. What question does this record answer, and who would ask it? And how long does anyone retain the ability to ask?

Sort by the question the record answers, not by the department that made it

Filing by department is why organizations end up with four copies of the same invoice on three retention clocks. Sort instead by evidentiary purpose. A billing and admin function typically produces five or six distinct classes, no more.

• Proof that money moved. Invoices, remittances, bank statements, merchant settlement reports. These support what you reported to a taxing authority.
• Proof of what was agreed. Signed contracts, purchase orders, change orders, rate schedules, the email where the client approved the overage.
• Proof of what was done. Timesheets, delivery confirmations, service tickets, call logs.
• Proof of who was told what. Dunning notices, dispute correspondence, cancellation confirmations.
• Operational scaffolding. Draft reconciliations, working spreadsheets, screenshots someone took to explain a variance in a meeting three quarters ago.

The first four classes are evidence. The fifth is not, and it is usually eighty percent of your shared drive by volume. Naming it as a class is what gives a staffer permission to delete it, which no general policy statement ever quite does.

Three clocks, and the one that runs longest wins

Every record in the evidence classes sits under at least three clocks, and the retention period is simply the longest of them. The tax clock runs from when a return was filed or due; the IRS is the authority that sets how long records supporting a return must be available for examination, and that period varies with the situation rather than being one flat number. The liability clock is your state's statute of limitations on the underlying claim, which for a written contract commonly runs years longer than the tax clock. The contractual clock is whatever your own customer agreements, grant terms, or payer contracts obligate you to produce on request, and it is the one organizations most often forget they signed.

Do the comparison once per record class and write down the winner with its reason. "Signed service agreements: retained through the contract term plus the state limitations period on written contracts, because that is the window in which a dispute can still be filed." A staffer who reads that sentence can reason about a record type nobody anticipated. A staffer who reads "seven years" cannot.

Cost the storage out, because that is what breaks the tie

When the three clocks are silent, cost decides, and cost is knowable in a way most retention debates pretend it isn't. Take a hypothetical accounts receivable function with 140 standard record boxes in an offsite vendor's facility. Suppose the vendor bills $0.40 per box per month plus $18 per retrieval, and your team pulls nine boxes a year. That is $672 a year in storage against $162 in retrievals. Annual carrying cost per box: about $4.80. Over a nine-year retention period, roughly $43 per box, before anyone's time.

Now the retrieval side. Nine pulls a year, each costing a specialist perhaps ninety minutes of locating, requesting, waiting, scanning and refiling. At a fully loaded $52 an hour, that is around $700 a year in labor, more than the storage line. The lesson is not that storage is cheap. It is that retrieval difficulty is the expensive variable, so indexing quality earns more than deletion does.

What cannot be costed honestly is the value of a record you did not keep. Nobody can tell you the probability that a specific 2019 change order becomes the pivot of a dispute in 2026. So do not let anyone assign that a number. Treat it as a reason to favor the longer clock in the evidence classes and be ruthless in the scaffolding class, where the downside genuinely is zero.

Assign an owner per class, and one exception that overrides everything

Each record class needs a named role, not a person, accountable for its schedule: accounts receivable manager, controller, HR director. Review the schedule annually against a short agenda: contracts signed this year that added obligations, systems retired, storage invoices, and any retrieval that took more than a day.

The override is legal hold. The moment litigation, an audit, or a government inquiry is reasonably anticipated, routine destruction stops for anything arguably relevant, regardless of what the schedule says. Everyone who can delete anything needs to know that sentence and know who issues the hold. Build that reflex and the rest of the schedule can be as aggressive as you like, because you have a brake.

Start with one class. Take invoices, write the three clocks and the winner, put a name against it, and see how quickly the next four classes reveal that they were mostly copies of the first.