Not completely. Digitise the paper that feeds the job you want AI to help with, and stop that paper being created in the first place, before you buy any AI tool. Everything else can wait. For most small firms that means job sheets, quotes, supplier invoices and receipts, done in four to eight weeks, not a full archive scan.
The reason is practical: AI can only work on what exists as a file, and it reads a typed digital form far more reliably than a photo of handwriting. The exception is a business that genuinely searches its old records, for example past quotes used to price new work, or guarantee records customers ask about years later. Then a targeted archive scan earns its place, but only after new paper has stopped arriving.
Why "scan everything" is the wrong first move
Owners who decide to go paperless often start with the filing cabinet: boxes of old quotes and invoices, fed through a scanner over several weekends. It feels productive, and it's usually the least useful part of the job, for three reasons.
- Paper keeps arriving. While you scan last year, crews are filling in new paper job sheets. The problem refills itself every week.
- Old files are rarely asked for. Most archives are opened a handful of times a year. AI adds little to documents nobody queries.
- Scans are the weakest input. A typed form arrives as clean text. A scan has to be converted by OCR (optical character recognition, the software that turns an image of text into real text), and handwriting, coffee rings and folds all cut accuracy.
So the rule for this plan: stop new paper first, digitise live paper second, and touch the archive only if you'll genuinely search it.
Step 1: Count your paper for two weeks
Time: ten minutes a day for one person. Put a tally sheet by the in-tray and another in each van, and record every piece of paper that's created or received. You're after volumes and journeys, not a perfect inventory.
| Document | Created or received by | Per week | Where it goes next | Who needs it later | Could it feed an AI job? |
|---|---|---|---|---|---|
| Job sheet | Crew leader, on site | 40 | Office, typed into accounts | Office, owner, customer | Yes: invoicing, job costing |
| Supplier delivery note | Received at yard | 20 | Stapled to invoice | Bookkeeper | Yes: matching bills to orders |
| Fuel and materials receipts | Everyone | 35 | Envelope on the office desk | Bookkeeper | Yes: expense capture |
| Signed quote acceptance | Customer | 6 | Customer folder | Owner | Rarely |
At the end of two weeks, circle the documents that feed the job you want AI to do. If you haven't picked that job yet, choose the paper that costs the most retyping time; it's almost always the right place to start.
Step 2: Sort every document into stop, convert, scan or keep
Time: one hour with whoever handles the paper most. Each document on your tally goes into one of four buckets:
| Bucket | What it means | Typical examples |
|---|---|---|
| Stop | Nobody needs it, or it duplicates something digital | Printed copies of emailed quotes; paper timesheets that repeat the job sheet |
| Convert at source | Replace the paper with a digital form or email so it's never paper at all | Job sheets, site checklists, holiday requests, supplier invoices you can ask to receive by email |
| Scan on arrival | It will keep arriving on paper; digitise it the day it lands | Receipts, delivery notes, the occasional posted letter |
| Keep on paper | Must stay in its original form, or isn't worth digitising | Anything your accountant or solicitor says to keep as an original; signed deeds; archive nobody will search |
Aim to put most of your volume into the first two buckets. Scanning is a fallback, not the goal.
Consider how the sort might go in a domestic cleaning company with twelve cleaners. Paper timesheets, filled in weekly and retyped for payroll, go to "convert at source" as a phone check-in and check-out. Client instruction cards ("no bleach on the oak worktop", "cat must stay in the utility room") are handwritten and live in a ring binder; they convert too, because they're exactly what an AI tool would need to draft a cover cleaner's briefing when someone's off sick. Supply receipts are "scan on arrival". The key register stays on paper in the locked cabinet for now: it's checked by hand every evening, nothing the AI will do needs it, and a digital version would need thinking through with the insurer first. The sort doesn't have to be all-or-nothing, and one sensible "keep" is a sign it was done honestly.
Step 3: Replace paper where it's created
Time: two to three weeks, including a trial with one crew. This is the step that makes the rest stick.
Job sheets and site forms
Build a phone form with the same fields as the paper sheet, in the same order, so crews recognise it. Microsoft Forms comes with Microsoft 365 business plans and Google Forms comes with a Google account; trade job-management apps have job sheets built in if you already use one. Keep it short: every extra field is a reason for a crew leader to go back to paper. Drop-down lists beat free text wherever the answer comes from a known set (job type, materials, crew), because AI and spreadsheets both handle consistent values far better than twelve spellings of "flashing".
Times deserve the same care. A job-sheet form with a free-text "hours on site" box will collect "7.5", "7h30", "7.30" and "half 8 till 4" within a fortnight. People read those without effort; an AI drafting invoices from them may not, and "half 8 till 4" can come out as 8.5 hours instead of 7.5. The fix is two time fields, arrival and departure, with the form working out the difference. Then nothing needs interpreting.
Two field tests before rolling it out: fill it in wearing work gloves, and fill it in somewhere with no mobile signal. If either fails, fix it before crews see it. If your crews prefer talking to typing, turning on-site voice notes into job sheets is a good alternative.
Supplier invoices and statements
Email each regular supplier asking for invoices as PDF to a single address, such as an accounts inbox. Most will switch the same week. That one email often removes a third of the paper in a trade business.
Subject: Please send our invoices by email from [date]
Hello [name],
From [date] we're moving to digital invoices. Please send all
invoices and statements for account [account number] as PDF to
accounts@[yourdomain], one invoice per PDF, with your invoice number
in the file name or subject line if your system allows.
Please stop posting paper copies once the first emailed one arrives.
Thanks for your help.Expect a holdout or two: a small local merchant whose invoices print from an old till system, say. Don't chase them for months. Move that supplier to "scan on arrival", with the scan done the day the invoice lands, and ask again when they change systems. "One invoice per PDF" matters more than it looks, because a supplier who emails a month of invoices as one file has recreated the 80-page-scan problem described below.
Receipts
Photograph them on the spot, before they fade or get lost in a van footwell. A receipt-capture app or your accounting software's own app does this; the rule that matters is "photo at the till, bin it later".
Step 4: Set up scanning for what still arrives on paper
Time: half a day to set up. For most small firms a phone is enough:
- OneDrive app: the plus button, then Scan. Microsoft retired its separate Lens scanning app in early 2026 (scanning in it stopped working after 9 March) and now points people to OneDrive instead (see Microsoft's retirement notice).
- Google Drive app: the Add button, then Scan, on both Android and iPhone. It crops, straightens and can clean marks off the page.
- Your phone's built-in document scanner works for the odd letter, but saving straight into a shared cloud folder is easier to control.
If you're scanning more than about 50 pages a day, a desktop sheet-fed scanner pays for itself in staff time. The ScanSnap iX1600, a common small-office choice, was listed at roughly $410 to $600 on the maker's store in September 2026, depending on edition.
Whichever you use, set these once and leave them: save as searchable PDF (so the text inside can be found and read), 300 dpi for normal documents, greyscale unless colour carries meaning, and one document per file. A single 80-page PDF of a month's receipts is almost useless to an AI tool that needs to tell one receipt from the next. For forms that will always be handwritten, turning handwritten forms into spreadsheet data with AI covers the extraction side.
Step 5: Name and file so software can find things
Time: an hour to agree the rules, then seconds per document. AI tools that search your files, such as Copilot in Microsoft 365 or Gemini in Google Workspace, lean heavily on file names and folder structure. A folder called "Scans" full of files named "IMG_4471.pdf" defeats them. Agree one pattern and pin it above the scanner:
Pattern: YYYY-MM-DD_customer-or-supplier_document-type_reference.pdf
2026-09-14_barn-conversion_job-sheet_J1043.pdf
2026-09-15_timber-supplies_invoice_INV88213.pdf
2026-09-15_fuel_receipt_van2.pdf
Folders:
/Jobs/2026/J1043 barn-conversion/
/Purchases/2026/09/
/Receipts/2026/09/Dates first means files sort in order. Lower case with hyphens avoids problems between systems. The full approach to folders and permissions is in organising shared files so AI tools can use them.
Step 6: Decide what happens to the originals
Time: a 20-minute conversation with your accountant. Record-keeping rules for business documents vary, and some originals must be kept for a set number of years or in their original form. Ask your accountant two questions: which documents can we keep as scans only, and how long must each type be kept? Write the answers into a one-line policy per document type.
For anything you're allowed to destroy, wait until the scan has been checked and the backup has run at least once. Going paperless moves your risk from a filing cabinet to a cloud account, so confirm your files are backed up somewhere separate before you shred anything.
The archive: when scanning old files is worth it
Scan the backlog only when you can name the question you'll ask of it. Good reasons in a trade business:
- Past quotes and final invoices, if you want AI to compare a new job against similar past jobs to sanity-check your price.
- Guarantee and warranty records, if customers call years later and you need to find the job fast.
- Anything a buyer would ask to see if you plan to sell the business in the next few years.
Everything else can stay boxed. If the backlog is large, scanning bureaus quote per page or per box; get two quotes and ask whether the price includes OCR and your naming pattern.
Before accepting "we'll do it ourselves over a few weekends", do the sum. Twelve archive boxes at roughly 2,500 pages each is 30,000 pages. Even a fast desktop scanner spends most of its time waiting for someone to pull staples, unfold delivery notes and name the file; at a realistic 300 to 400 pages an hour including that preparation, 30,000 pages is 75 to 100 hours of staff time (illustrative). Put your hourly staff cost against that and compare it with the bureau quotes. Often the answer is neither: scan only the two boxes of past quotes you actually want to search, and leave the other ten alone.
Worked example: a roofing contractor's eight-week switch
An illustration: a nine-person roofing contractor with three crews. The two-week count showed about 45 paper job sheets, 20 delivery notes and 35 receipts a week. The office manager spent roughly six hours a week typing job sheets into the accounts system and chasing missing ones. The owner wanted AI to draft invoices from completed job sheets and flag jobs running over budget.
- Weeks 1 and 2: paper count and sorting. Job sheets and supplier invoices marked "convert at source"; receipts "scan on arrival".
- Weeks 3 and 4: a 14-field job-sheet form in Microsoft Forms, already included in their Microsoft 365 plan. One crew trialled it; the materials field became a drop-down after the first week produced 11 spellings of the same membrane.
- Week 5: all three crews on the form. Suppliers asked for PDF invoices to the accounts inbox; six of eight switched within ten days.
- Weeks 6 and 7: receipts photographed with the OneDrive app into dated folders using the naming pattern. The accountant confirmed which originals to keep.
- Week 8: readiness test (below), then the first AI trial: summarising a week of job-sheet responses into invoice drafts for the office manager to check.
Cost: no new software, about 25 hours of the office manager's and owner's time, and one crew leader's patience. Typing time for job sheets fell from about six hours a week to under one, before any AI had been added. That saving alone justified the project.
Five tests that say your records are ready for AI
- Find any document from last month in under 30 seconds, by searching, without asking the person who filed it.
- Search a customer's surname and get their job sheet and invoice, which proves the text inside your files is searchable, not just the file names.
- Four clear weeks with no paper in the process you care about.
- Spot-check 20 files: at least 19 follow the naming pattern and sit in the right folder.
- Give an AI tool ten documents and check its extraction (dates, amounts, names) against the originals. If it gets fewer than nine right, fix scan quality or the form design before building anything on top.
Test 5 is quick to run in any business chat assistant. Upload ten scanned receipts and ask: "For each file, give the file name, supplier, date as DD/MM/YYYY, and total paid. If any value is unclear, write UNCLEAR rather than guessing." At the roofing contractor, the first run might come back with eight clean rows and two problems (illustrative):
File Supplier Date Total
2026-09-03_fuel_receipt_van2 [fuel station] 09/03/2026 $86.40
2026-09-16_merchant_receipt [merchant] 16/09/2026 $212.00The fuel receipt was from 3 September. Its till printed the month first, and the model read the date the other way round, as 9 March; the file name, which the office had dated correctly, gave it away. The merchant total was the subtotal before tax, because the faded thermal paper had lost the last line. Neither was flagged UNCLEAR, which is the real lesson: the model guessed confidently. The fixes were to rescan faded receipts in colour, which picked up the last line, and to tell the extraction step which suppliers print dates month first. The rerun scored ten out of ten, and the invoice-drafting trial went ahead.
Where paperless projects stall
- Running paper and digital side by side "for now". Set a switch-off date for the paper version and remove the old pads from the vans.
- Forms designed in the office for people on roofs. Too many fields, no offline use, tiny buttons. Watch someone fill it in on site before you finalise it.
- Scanning without searchable text. Image-only PDFs look digital but can't be searched or read reliably. Check the setting.
- A dumping-ground folder. "To file" folders become the new in-tray. Name and file on the day, or the scan hasn't really happened.
Once the paper flow you care about is digital and tidy, the AI step is much smaller than it looked. AI document processing for PDFs and scans is the natural next read.
Further reads
- AI Document Extraction vs Traditional OCR: Which Should You Use? — Which text-reading approach suits your scanned documents.
- Can AI Sort and File Incoming Documents for Your Business? — The next step once documents arrive digitally.
- How to Prepare Your Business Data for AI, Step by Step — Cleaning the records you've just digitised.
- How to Document Your Processes Before Adding AI — Write down the process the paper belonged to.
- AI Invoice Processing: Stop Typing Supplier Bills by Hand — Put your newly digital supplier invoices to work.
- Dext vs Hubdoc: Which Receipt Capture Tool Saves More Time? — Two receipt-capture tools compared for small firms.
- Automating a Broken Process: Why It Backfires and What to Fix — Why automating a broken process backfires, the five-pass fix to do first, and how to split the cleaned-up process between rules, AI and people.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: Microsoft Support (Microsoft Lens retirement); Google Drive Help (scanning documents); PFU Ricoh store listing for the ScanSnap iX1600 (checked September 2026).