Invoice processing · Production

Nanonets automates construction invoice processing, delivering 10x speed increase and 7,200 work hours reprioritized

The problem

A Minnesota-based construction company processed a large volume of invoices from over 40 suppliers with different formats, spending significant time and resources on manual data entry and verification before data could be entered into accounting software.

First attempt

The company first tried Textract and then Abbyy; both failed to handle multi-page invoices, multi-line fields, and unstructured documents with imperfections. Accuracy was poor requiring excessive rework, the verification UI took up to ~5 minutes per invoice, and it proved impossible to retrain either algorithm, forcing the company to abandon both tools.

Workflow diagram · grounded in source

Invoice submission via API

trigger

“The AP team at the client site sends invoices to Nanonets using a simple API integration”

Document type classification

ai_action

“The Nanonets algorithm can filter, recognise and parse different types of documents such as invoices, material lists, packing lists, purchase orders etc. all at the same time! The Nanonets AI can segregate critical documents from non cri…”

Real-time invoice data extraction

ai_action

“Nanonets AI automatically processes these invoices in real time and populates a csv that gets stored in the AP team's Drive”

Data validation and formatting

validation

“Custom validation rules allow you to reorganize data into convenient output layouts and formats that are easier for further processing”

CSV upload to accounting software

integration

“The AP team, at the end of the day, downloads the csv and uploads it to their accounting software”

Continuous model retraining

feedback_loop

“the AI retrains itself with the data you process. This ensures that the algorithm functions accurately even if you onboard new suppliers each month”

Reported outcome

After switching to Nanonets, the company automated the most labor-intensive steps of document processing, achieving a 10x increase in processing speed and reprioritizing 7200 work hours.
Invoice verification that would earlier take over 5 minutes for batches of documents could be blazed through in under 30 seconds.

Reported metrics

Invoice processing speed10x increase in processing speed

Work hours reprioritized7200

verification time before Nanonets (per invoice)~5 minutes per invoice

batch verification time before Nanonetsover 5 minutes

Show all 8 reported metrics

invoice processing speed10x increase in processing speed

work hours reprioritized7200

verification time before Nanonets (per invoice)~5 minutes per invoice

batch verification time before Nanonetsover 5 minutes

verification time with Nanonetsunder 30 seconds

Nanonets setup timejust about 1 day

prior tool struggle durationabout 3 months

number of supplier invoice formats handledover 40 different suppliers

Reported stack

NanonetsTextractAbbyyNanonets OCR API

Source

https://nanonets.com/customer-success-story/construction-invoice-processing

Read source ↗

Frequently asked questions

What did this team achieve with this AI workflow?

After switching to Nanonets, the company automated the most labor-intensive steps of document processing, achieving a 10x increase in processing speed and reprioritizing 7200 work hours.

What tools did this team use?

Nanonets, Textract, Abbyy, Nanonets OCR API.

What results were reported?

Invoice processing speed: 10x increase in processing speed; Work hours reprioritized: 7200; verification time before Nanonets (per invoice): ~5 minutes per invoice; batch verification time before Nanonets: over 5 minutes (source-reported, not independently verified).

What failed first in this deployment?

The company first tried Textract and then Abbyy; both failed to handle multi-page invoices, multi-line fields, and unstructured documents with imperfections.

How is this invoice processing AI workflow structured?

Invoice submission via API → Document type classification → Real-time invoice data extraction → Data validation and formatting → CSV upload to accounting software → Continuous model retraining.