AI can only use the records it can read, so start capturing them today
Fix how each new record is captured, starting today. The records an automation needs are the ones your business makes from now on.
Owners hear that AI can code receipts, compare job costs or chase missing paperwork. Then they look at where their records live: a truck cab, a shoebox, a phone camera roll, one person's memory. Most owners then make one of two mistakes. They decide the business is not ready and drop the idea. Or they plan a big project to scan years of old paper first. Both miss the cheap fix.
This blog has covered writing down what your team knows before you buy any tool. This post covers the everyday transaction trail: receipts, tickets and timesheets. That is a different problem with a different fix.
The records your first automation will lean on¶
Most first automations depend on six record types: receipts, job or work tickets, timesheets, quotes, supplier invoices and signed approvals.
An automation that codes receipts reads this month's receipts. A job-cost comparison reads this quarter's tickets and timesheets. Old records describe work that is finished. New records feed the work you want to automate.
So the next three months of records matter more than the last three years.
A plain test for whether a record is readable¶
A record is readable when it passes four checks:
- One known place. Everyone knows where it goes, and it goes there.
- One consistent form. Every record of that type has the same details in the same order.
- Legible or typed. A person who was not there can read it.
- Captured within a day. It is recorded within a day of the event it describes.
A record that fails one check is a record an automation cannot trust.
Capture each record where it is made¶
Change the moment of capture. Leave everything else alone.
Capture at the source. Take a clear photo or fill a short form at the job site or the desk, when the record is made.
One place per record type. Use one shared folder or one inbox for receipts, and a separate one for tickets. Do not mix them.
One named checker. One person checks each folder once a week for gaps: missing days, blurred photos, empty fields.
This does not fix a record that nobody makes. It does show you, every week, where records are missing.
Why the archive can wait¶
Scanning old paper is slow. It costs more staff time than it returns. Old records rarely follow one pattern, so an automation struggles to read them even after they are scanned.
Leave the archive in its boxes. Bring old records in only when a specific task needs them, such as year-end.
A courier firm changes one habit¶
A 10-person courier and delivery firm wants a monthly cost-per-route comparison. Drivers keep fuel and toll receipts in the cab. They hand in paper delivery sheets every Friday. Receipts are faded or missing. The delivery sheets are in three different handwritings. There is nothing consistent to compare.
The firm changes one habit. Drivers photograph each receipt into one shared folder the same day. They fill a short phone form at each drop. The office coordinator checks both folders every Friday and follows up on gaps.
After about six weeks, the records are complete enough to build the comparison from. The old paper stays in its boxes.
When your records are ready for automation¶
The signal is a few weeks of complete, consistent capture for one record type. Then look at automating the one task that uses that record type. Not before.
If the Friday check still finds gaps, keep fixing capture. An automation built on gappy records gives wrong answers with full confidence.
Start with the next receipt, not the last box.
Keep exploring¶
To find out which records in your business can support a first automation, start the AI Readiness Audit or contact FIT.