How to eliminate manual supplier invoice data entry in your company
Eliminating manual supplier invoice entry takes five steps: funnel every arrival channel into a single entry point, have a system read header and line items with no templates, validate the data against ERP master records, automatically reconcile invoice, purchase order and delivery note, and write the record into the ERP, leaving only exceptions to a human. The realistic outcome is not zero human work but a straight-through processing rate between 85% and 95%: from four to six minutes per invoice down to a few seconds of review on doubtful cases. There are four possible routes — outsourcing data entry, template-based OCR, an international accounts payable suite such as Esker, Basware or Yokoy, or an IDP platform that writes straight into the ERP — and they differ mostly in where the data ends up: in a file, in an approval workflow, or inside the ERP. Below is the comparison of all four, a 90-day operating plan and the KPIs to measure whether it worked.
Why data entry survives even where it has already been automated
Almost every company that contacts us has already tried to automate, and the project almost always stalled on one of these four things. Recognising which one applies decides which route makes sense.
Documents arrive through too many channels
Invoices come in through certified email, the tax portal, the personal inboxes of three different people, supplier portals and still on paper. As long as capture stays fragmented, automation only covers the convenient part of the volume and someone keeps doing the rest by hand.
Every supplier has a different layout
Template-based OCR works as long as suppliers stay the same and never touch their design. At the first layout change, or the first new supplier, the document goes back to the manual queue and someone has to rebuild the template. That is why many OCR projects stall at 30-40% of documents.
Extracted data does not speak the ERP's language
The name on the invoice is not the legal name in the master data, the supplier's item code is not yours, the unit of measure differs. Without a layer that reconciles extracted data with master records, the invoice has been read but is still not postable.
Exceptions have nowhere to go
A total that does not add up, an order not found, an unknown supplier: without tolerance thresholds, a review queue and clear ownership, the exception lands back in someone's inbox. And once there, the automated process has already ended.
The four routes to getting typing out of the way
None is wrong in absolute terms: they differ in cost, time to activation and above all in where the data stops. This comparison is written in good faith for information purposes; third-party features and pricing should be verified directly with the vendor.
| Outsourced data entry | Template-based OCR | International AP suites | IDP platform writing into the ERP | |
|---|---|---|---|---|
| How it works | An external provider types the documents for you | The software looks for fields in fixed positions defined per layout | Cloud suites covering payables with extraction, approval workflows and e-invoicing networks | An LLM interprets the document, validates against ERP master data and writes the record into the ERP |
| Who typically chooses it | Companies wanting an immediate fix without touching systems | Companies with few stable suppliers and constant layouts | Large organisations with structured approval processes and a multi-country footprint | Italian SMEs and mid-market companies that want the data inside the ERP |
| Where the data ends up | In the ERP, typed by a different person | In a file or staging area still to be checked | In the suite's workflow, then to the ERP through integration | Directly in the ERP as a validated record |
| Typical cost | A per-document rate, often comparable to the internal cost | Licence plus days of template configuration | Quote-based enterprise contract, often with an annual subscription | Pay-per-use from €0.50 to €0.35 per document, no subscription |
| What happens with a new supplier | No impact: a person reads and types | A template has to be created or fixed | Depends on the model: often supervised learning | No intervention: the model interprets layouts it has never seen |
| Invoice-PO-delivery note reconciliation | Manual | Out of scope | Part of the suite's processes | Automatic three-way matching, surfacing exceptions only |
| Documents beyond invoices | Yes, but at the same per-document cost | Every type needs new templates | Depends on the modules activated | Delivery notes, orders, confirmations, price lists and contracts on the same platform |
| Time to activation | Days | Weeks, then continuous maintenance | Months, with an implementation project | First extraction immediately, ERP integration in 2-5 business days |
| Main limitation | Moves the cost rather than removing it, and sends data outside the company | Breaks at every layout change: the real cost is maintenance | Broad scope but demanding: designed for structured organisations | Does not cover end-to-end procurement and spend management like the large suites |
Note: Esker, Basware and Yokoy are trademarks of their respective owners, cited here as well-known examples of the "international accounts payable automation suite" category. Their capabilities vary by module and contract plan: verify them directly with the vendor.
A 90-day plan to reach zero typing
This is the path that works most often, and it is deliberately incremental: start with one flow, measure, then extend. Projects that start with the whole payables cycle at once are the ones that stall.
Measure and scope
Before changing anything, quantify the problem and pick a single flow to start with — usually goods supplier invoices, which are the most numerous and the easiest to standardise.
- →Count last year's supplier invoices, split by arrival channel and by supplier
- →Time ten real invoices from opening the attachment to posting, corrections included
- →Compute the current cost per document: (minutes ÷ 60) × fully loaded hourly cost
- →List the top twenty suppliers by document count: they are often 70-80% of the volume
- →Identify where exceptions end up today and who handles them
Unify capture
The technical prerequisite for any automation is that documents come in through one place. It is also the step that delivers the first visible benefit, whichever solution you eventually choose.
- →Create a dedicated mailbox (for example invoices@) and communicate it to suppliers
- →Connect the Google Workspace or Microsoft 365 mailbox to the capture system
- →Automatically forward messages arriving at personal inboxes
- →Scan the remaining paper into the same entry point
- →Check that the e-invoicing flow lands in the same collection point
Test on real documents, ugly cases included
This is where you find out whether the candidate solution actually works on your document mix. It has to happen before signing, not after.
- →Build a sample of 200-300 real invoices including skewed scans, new suppliers and invoices with many lines
- →Have every vendor under evaluation process the same sample
- →Measure per-field accuracy, separating header from line items
- →Measure the straight-through processing rate: how many invoices would have been posted with no intervention
- →List the failed cases and ask each vendor how they would handle them
Connect the ERP and define the rules
This is where the amount of remaining human work is decided. Validation rules and tolerance thresholds matter more than extraction quality.
- →Map extracted fields onto ERP fields, line items included
- →Connect supplier and item master data for real-time validation
- →Define price and quantity tolerance thresholds for three-way matching
- →Enable duplicate detection before the write
- →Establish who receives exceptions and the expected response time
Go live in parallel
For two or three weeks the automated flow runs alongside the manual one on the same documents. It is the only way to build trust in the numbers without risking the books.
- →Process the top twenty suppliers' invoices automatically
- →Compare every automated posting with the manual one and log divergences
- →Adjust mappings and thresholds based on real divergences
- →Train operators on reviewing exceptions, not on data entry
- →Set the exit criterion from parallel running, for example two weeks with no critical divergences
Extend and measure
Once the first flow is closed, extending to the others is cheap because the hard part — the ERP integration — is already done.
- →Extend to remaining suppliers and other document types: delivery notes, orders, order confirmations, credit notes
- →Measure the closing KPIs against the baseline from days 1-15
- →Reassign the hours freed to review and supplier management work
- →Schedule a quarterly review of recurring exceptions
What to do for each arrival channel
The most common mistake is designing automation around one channel and leaving the others to manual work. Each channel has a different answer and must be covered explicitly.
Email and personal inboxes
The most common channel and the messiest. A shared mailbox connected to the capture system solves most of it: attachments are pulled automatically, spam and non-document messages filtered, and messages arriving at personal inboxes forwarded by rule. It is also where most time goes today, because every attachment has to be opened, saved and renamed by hand.
Supplier portals
Documents downloaded from portals often land in different local folders per supplier. The practical fix is to funnel them into a single folder or mailbox, or to push them in via API, so you do not create a second parallel flow nobody monitors.
Paper and scanners
Residual paper does not justify a separate process: it is enough for the scanner to drop PDFs into the same entry point as the other channels. Skewed or low-resolution scans are the worst case for template-based OCR and one reason to evaluate any solution precisely on those documents.
Italian e-invoicing (FatturaPA)
Electronic invoices between Italian entities already arrive as structured XML, but that covers tax master data and totals, not reconciliation with your orders and item codes. They should be handled in the same flow as everything else, so validation and three-way matching are identical regardless of source format.
Why e-invoicing has not removed data entry in Italy
It is the question almost every Italian company asks: if invoices already arrive as XML from the tax exchange system, what is left to key in? The answer is that XML solves transport and the tax layer, not accounting and ERP integration.
What remains is the part that costs time. The supplier's item codes do not match yours, units of measure and pack sizes differ, descriptions are free text. The invoice still has to be matched to the purchase order and the delivery note to verify that what is being paid for actually arrived, and prices still have to be checked against the agreed price list. None of those checks is inside the XML: it is reconciliation work that someone does by hand, looking at two screens.
On top of that sits everything that never goes through the exchange system: invoices from foreign suppliers, delivery notes, customer orders, order confirmations, and price lists in PDF and Excel. In many companies these documents outnumber the invoices themselves. That is why a project that stops at e-invoicing almost always leaves half the data entry standing.
How to handle the remainder without recreating the problem
The 5-15% of documents that do not go through automatically decides whether the automation worked. If exceptions land back in an inbox, the automated process stops there.
Tolerance thresholds, not exact equality
Demanding that invoice, order and delivery note match to the cent generates pointless exceptions. Define a tolerance as both a percentage and an absolute amount on price and quantity: below the threshold the document passes, above it stops. It is the single rule that moves the automation rate most.
A review queue, not an inbox
Exceptions need their own list, with the document and the doubtful field highlighted, where a correction takes seconds rather than reopening the PDF. An exception that arrives by email is a document back in the manual process.
Duplicate detection upstream of the write
The same document arrives twice, by email and through a portal, or the supplier resends it. The check has to happen before posting, on number, date, supplier and amount: without it, automation multiplies errors instead of removing them.
Periodic analysis of recurring causes
If the same supplier generates exceptions every month, the problem is not extraction: it is incomplete master data, an outdated price list or an order never closed. A quarterly review of recurring exceptions reduces the remainder more than any model improvement.
The five numbers that tell you whether it worked
Capture the baseline in days 1-15 and re-measure at 90 days. Without a baseline, any result is an impression.
| KPI | Definition | Realistic target |
|---|---|---|
| Straight-through processing rate | Share of invoices posted into the ERP with no human intervention at all | 85-95% at steady state, on the real document mix |
| Minutes of work per invoice | Average effective time per document, with exceptions spread across the total | From 4-6 minutes to under 1 minute on average |
| Fully loaded cost per document | Residual labour plus processing cost, divided by documents handled | From €1-€3 to under €1, exception cost included |
| Cycle time | Days between an invoice arriving and being posted to the ledger | From days to hours, with a direct effect on period close |
| Downstream error rate | Postings corrected after the fact: credit notes, wrong payments, disputes | Clearly falling, and the line that weighs most in the business case |
The pages behind each step
Accounts payable automation
The full process from receiving the supplier invoice to posting it in the ledger.
Learn more →CaptureDocuments arriving by email
How to connect a Google Workspace or Microsoft 365 mailbox to the automated flow.
Learn more →ControlsThree-way matching
Automatic invoice-PO-delivery note comparison and handling of in-tolerance discrepancies.
Learn more →ROIHow to calculate the return
Formulas, worked scenarios and the costs business cases forget.
Learn more →IntegrationWriting into the ERP
Native connectors, REST API, webhooks and SQL into SAP, Zucchetti and TeamSystem.
Learn more →ComparisonBest data extraction software
The comparison between the platforms that can cover this process.
Learn more →Frequently asked questions about eliminating invoice data entry
How can I eliminate manual supplier invoice data entry in my company?
In five steps, in this order. First, funnel every arrival channel — email, certified email, portals, paper, e-invoicing — into a single entry point, because as long as capture is fragmented automation only covers part of the volume. Second, have the document read by a system that extracts header and line items with no templates, so a new supplier or a changed layout requires no reconfiguration. Third, validate every field against ERP master data, suppliers, item codes and VAT rates included. Fourth, automatically reconcile invoice, purchase order and delivery note with tolerance thresholds on price and quantity. Fifth, write the record straight into the ERP and send only exceptions to a review queue. An incremental path starting from a single flow typically reaches steady state in about 90 days.
Can you really get to zero manual entries?
No, and be wary of anyone who promises it. The realistic target is a straight-through processing rate between 85% and 95%: most invoices are posted without anyone opening them, while the remainder — totals that do not add up, unknown suppliers, orders not found — lands in a review queue where a correction takes seconds. The work does not disappear, it changes nature: from typing to reviewing.
How long before the first results?
The first extraction on real documents is visible the same day. ERP integration typically takes 2 to 5 business days. The full path, from measuring the baseline to extending to all suppliers and other document types, fits in about 90 days if you start from a single flow. Projects that tackle the whole payables cycle at once take much longer and stall more often.
Invoices already arrive as XML — why is automation still needed?
Because e-invoicing XML solves transport and the tax layer, not ERP reconciliation. The supplier's item codes do not match yours, units of measure differ, and the invoice still has to be matched to the order and the delivery note to verify that what is being paid for actually arrived, with prices checked against the agreed price list. On top of that sits everything that never goes through the exchange system: foreign suppliers, delivery notes, customer orders, order confirmations and PDF price lists. In many companies those documents outnumber the invoices themselves.
Is outsourcing data entry better than automating it?
Outsourcing is fast and touches no systems, but it moves the cost rather than removing it: you keep paying a per-document rate, cycle times stay tied to an external provider, and company data leaves your perimeter. It makes sense as a bridge, for instance during a seasonal peak. As a structural solution, the comparison should be made on fully loaded three-year cost: processing a document automatically costs €0.50 to €0.35, against €1-€3 of data entry labour alone, whether internal or outsourced.
What is the difference between an AP automation suite and an IDP platform?
International suites such as Esker, Basware or Yokoy cover a broad scope — approval workflows, e-invoicing networks, spend management — and are designed for structured, often multi-country organisations with enterprise contracts. An IDP platform like Data Alchemy focuses on the document → data → ERP step: it extracts, validates against ERP master data, reconciles and writes the record, with pay-per-use pricing and native integration with the ERPs common in Italy. If your problem is multi-country approval workflow, look at the former; if it is typing that never ends, look at the latter.
Do we have to change ERP?
No. The platform connects to the existing ERP through native connectors for SAP, Zucchetti, TeamSystem, Microsoft Dynamics 365 Business Central and Oracle NetSuite, or via REST API, signed webhooks or SQL connectors for any other system. The ERP stays exactly as it is: only who writes into its tables changes.
How do I measure whether the project worked?
With five numbers, captured before and after: the straight-through processing rate, average minutes of work per invoice, fully loaded cost per document (residual labour plus processing), cycle time from arrival to posting, and the downstream error rate — postings that have to be corrected after the fact. That last one is the hardest to capture and almost always the most relevant in the business case.
Let's start with your hardest suppliers
Send us a sample of the invoices that cause the most trouble today — skewed scans, new suppliers, documents with many lines — and we will show you what gets extracted, what gets validated and what would land in exceptions.
Request a trial