Getting your data back out

Every system you buy will end.

The vendor is acquired and the product is folded into something else. The price changes and the new one does not suit you. The product is discontinued. Or you simply outgrow it, and the thing that fitted one site cannot hold nine.

None of these are unlikely. The only question worth asking at purchase is what you will be holding on that day.

Three things people confuse#

A report is a view, arranged for a human being to read. It answers a question you already had, in a layout somebody designed, usually as a PDF. It is not data. It cannot be loaded into anything else, it shows the current state rather than the history, and it contains only the columns the designer thought you would want.

An export is structured rows, in a machine-readable format, that another system can be made to read. This is the minimum thing you need.

A backup is everything, including the things no report shows: the audit trail, deleted and superseded records, the internal identifiers that link one record to another, and the attachments. A backup is usually taken for the vendor's benefit, not yours, and often cannot be restored anywhere except their own software. Ask which of the three you are being offered, because all three are sometimes described with the same word.

What a real export must contain#

Five things. An export missing any of them is a partial record, and a partial record loses arguments.

• Every record, not just the current state. If a rate was revised, a quantity corrected, a bill cancelled and reissued, all of it must come out. A snapshot of today tells you what you believe now, not what you did. The reason that matters is the whole argument of append-only records.
• Attachments as files. The challan photographs, the signed work orders, the site photographs, the scanned bills. As actual files, in folders, named in a way that ties them back to the record. Not as web links into a system you no longer have an account on.
• The audit trail. Who did what, when, and what the value was before. Six months after you leave, this is what answers a dispute about a quantity, and it is the part most often missing — what an audit trail is for sets out what a usable one contains.
• The links between records. Which receipt belongs to which order, which invoice to which receipt, which payment to which invoice. Two spreadsheets with no shared identifier between them are not a record of a purchase; they are two lists that happen to be about similar things.
• The code lists. Vendor masters, item masters, cost heads, site codes, units of measure, approval roles. Without them the export is full of codes that mean nothing. A column of numbers where the item names should be is technically an export and practically waste paper.

Test it during evaluation, not at exit#

Every one of these is easy to check while somebody still wants your business, and impossible to check afterwards.

Ask for an export on day one of the pilot. Not at the end. Day one, when there are ten records in the system, so you can read the file by eye and see exactly what did and did not come out. Then open it. Actually open it, in whatever your accounts team uses, and see whether the dates survived, whether the decimals survived, and whether the vendor names came out as names.

Check whether an attachment is a file or a link. Download the export, disconnect from the internet, and try to open one photograph. If it opens, it is a file. If it does not, you are holding a pointer to somebody else's server.

Check whether a deleted or superseded record appears. Cancel a document during the trial, then export. If the cancelled document is absent, the export is a picture of the present, and every correction you ever made has been quietly dropped from your history.

Time it. Ask how long an export of a full year takes, and whether it can be run by you or only by their support desk on request. An export you have to ask for is an export you will not get during a disagreement.

Contract points worth insisting on#

In plain words, no drafting language. Four points, and any reasonable vendor will agree to all four.

1. Export available at any time, without asking. A button you can press, not a favour. The moment you most need your data is the moment relations are worst.
2. A defined format. Named in the agreement — the file types and, ideally, that it includes attachments and history. "Data will be provided in a mutually agreed format" means nothing on the day.
3. A defined period after termination. A stated number of days during which you can still retrieve everything, even if the subscription has lapsed and even if there is a payment dispute. Fix the length yourself; do not accept silence.
4. No fee for your own data. An extraction charge is a hostage fee, and the time to refuse it is before signing, not after.

One more thing belongs in the same conversation, though it is not a contract point: what happens if the vendor simply goes quiet, as small companies do. Whether you can retrieve your own data without their help decides whether that is an inconvenience or a catastrophe. The cost of ignoring all of this is a large part of what a price list does not tell you.

Documents that must survive the system#

One category deserves separate attention: anything that was signed.

A work order, a measurement sheet, an approved bill, a handover certificate. A signature stored as a flag in a database — a column saying approved by — is worth nothing once the database is gone. What you need is a document that can be checked on its own, by someone with no login, using only the file itself. That property is the subject of tamper-evident documents, and it is worth asking about at evaluation, because it cannot be added afterwards to documents already signed.

Ask the vendor a direct question: if I export a signed work order today and open it in five years on a machine that has never heard of you, can anybody tell whether it has been altered. The answer separates a signature from a picture of one.

We took that question seriously enough to answer it in our own work. A sealed work order carries on every page its number, page N of M, a fingerprint of the file, and a QR code. The page that QR reaches needs no sign-in and states only that the true document has exactly that many pages, its fingerprint, the date it was sealed and who issued it — no amounts, no contents. A vendor, an auditor or a bank clerk holding the paper can check it, and it goes on working whether or not anybody still has an account with us.

All of this belongs in the evaluation itself, alongside the other axes in how to choose construction software. Exit is not a pessimistic subject. It is the only one that is guaranteed to come up.

The short version#

A report is for reading, an export is for moving, a backup is for the vendor. Insist on an export that carries history, attachments as files, the audit trail, the links between records and the code lists.

Test all of it in the first week of the trial, while somebody still wants your signature. And make sure anything signed can be verified without the system it came from.