Eleven Branches, One Missing Field Log. Five Checks on the Records You Actually Keep

A mechanical contractor with eleven branches across two states went into a payment dispute in 2022 over a hospital renovation finished three years earlier. The argument was narrow: whether four changes to the chilled water piping had been directed verbally by the general contractor's superintendent before the written change orders caught up. The company's own foremen had documented every one of them, in a field app, with photographs, GPS stamps and the superintendent's name typed into a notes field. None of that reached the attorney. What reached the attorney was a set of invoice PDFs with no attachments, exported during a software migration in 2021 by a project coordinator who had been told to bring over the billing history.

What counts as the record, and what merely displays it

In 2014 the same company kept job records in cardboard boxes in the Toledo warehouse, and the record was unambiguous: a carbon-copy daily log, a signed ticket, a marked-up drawing. The thing you kept was the thing itself. Field software changed that quietly, because what a supervisor sees on a screen is a rendering assembled from several database tables, and what a vendor exports is usually a flattened version of it. The photographs sat in one place, the timestamps in another, the note text in a third. Ask, before you ever need it, what an export actually contains and what it silently leaves behind.

Who holds it, and what happens the day you leave

The decision that made the loss likely was made in 2019, when the company signed a field management contract without anyone reading the termination and data provisions. The agreement gave them a full export on request during the term and a thirty-day window after cancellation, in the vendor's native format. That is a common arrangement and not an unreasonable one. It is only dangerous when nobody at the customer has decided, in advance, who runs the export, what they check it against, and where the result is stored so it survives the next system too. A larger provider has this problem several times over, because branches adopt tools locally.

Which clock applies, and why the longest one wins

Retention schedules written a decade ago tended to borrow a single number from tax practice and apply it to everything, which is how a company ends up destroying construction records at year seven while the statute of repose on the building still has years to run. The Internal Revenue Service is responsible for the recordkeeping expectations behind a return, and those expectations are real, but they are one clock among several. Contract limitation periods, lien and warranty windows, payer audit lookbacks, employment and benefits obligations, and state licensing rules all run on their own schedules. Write the schedule by category of record, and let the longest applicable clock govern each one.

Whether it will still open, and still be believable

Paper had one advantage that took a while to appreciate: it needed no software to be read, and its alteration was visible. Digital records need a plan for both readability and integrity. Readability means storing in formats that are not tied to a single vendor, checking a sample every year or two, and treating a proprietary database backup as a liability rather than an archive. Integrity means keeping the metadata that makes a record credible when someone disputes it, the creation date, the author, the version history, and the audit trail showing who changed what. A file that opens but cannot be authenticated has lost most of its value.

Whether anyone can find it in an afternoon

The contractor's records were not, strictly speaking, gone. Two of the four disputed changes were eventually reconstructed from a former project manager's email, a subcontractor's own daily reports, and a text message thread on a phone that had been kept in a drawer. That reconstruction cost roughly six weeks and a meaningful amount of billable attorney time, and it worked. The lesson the company drew was about indexing rather than volume: records get organized by job number, by legal entity, and by retention class at the moment they are created, because nobody has ever successfully organized them later. They also put one named person, not a department, in charge of the schedule.

What changed in the decision itself

Ten years ago the retention question was a storage question, and the honest answer for most multi-site organizations was to keep everything, because a warehouse bay was cheap and the risk of discarding something ran one direction. Now the question is a control question. Storage is close to free, so the failure mode is no longer destruction but dispersion: fifteen systems, four of them owned by vendors, each holding a slice of the record and none holding the whole. The fix is not more retention. It is deciding which system is authoritative for each class of record, and getting a complete, readable copy out of it on a schedule you set rather than one a cancellation notice sets for you.

The contractor now runs a quarterly export from every operational system into a single archive keyed to job number, and reviews the data provisions of any software contract before signature. It took a dispute to get there, which is the usual route, and the second dispute cost them almost nothing to document.