SMACC - Automatic Bookkeeping with AI
2 likes523 views
Wojciech Barczyński, head of development at Smacc, discusses the automation of finance department tasks using AI and deep learning, focusing on invoice extraction and accounting processes. Smacc has significantly improved processing times, error rates, and costs for companies, showcasing a considerable market opportunity in automating finance for SMEs in Germany. A case study highlights the transformation of a customer's invoice processing from 20 days and high costs to less than 12 hours and €1 per invoice using Smacc's technology.
SMACC - Automatic Bookkeeping with AI
- 1. November 8, 2017 Wojciech Barczyński Lead Software Developer www.smacc.io AUTOMATE BOOKKEEPING WITH AI
- 2. Wojciech Barczyński • Lead Software Developer, Head of Development Office - SMACC • Before: System Engineer – Lyke • Before: SAP Research (& Development)
- 3. SMACC automates tasks of the finance department end-to-end applying AI Deep Learning application DATA EXTRACTION FINANCIAL REPORTS ACCOUNTING PAYMENTSWORK- FLOWS CON- TROLLIN G Automated Financial Management Data entry Manual accounting and missed tax filling deadlines Outdated numbers and gut feeling Messy reconciliation and closing Missed payment targets and IBAN typing Receipt stamps and signature folders
- 4. SMACC‘s AI Extractors offers extraction of 70 data fields and all invoiced items Scalable Seamless integration High Availability & Secure Modern, RESTful API
- 5. FLEXIBLE INPUT PLUG & PLAY IMPLEMENTATION CONTINUOUS IMPROVEMENTS FULL SCALABILITY GOOD GENERALIZATION Challenge: • > 300 000 documents in system • > 10 000 different layouts Template & Rule based approaches: • Limited generalization • Maintenance and updating is time consuming & error prone Why Deep Learning for Extracting Invoices?
- 6. PDF / scanned document OCR Characters + Regions + Segments Bidirectional GRU RNN Labels Bidirectional GRU RNN for extracting data
- 7. Automatic accounting using NLP & Neural Networks
- 8. Challenge: • Extract domain specific context from position descriptions • position descriptions are short, keywords sometimes help, often not Bag of words, n-grams (word2vec – exp.) Additional input: Client ID, customer name, tax rate etc. Client specific approach Feed to fully connected Neural Network NLP & Neural Networks for accounting
- 9. Source: AP automation study 2014 of Institute of Finance Operations, Kofax & SSON 2016, Techvalidate 2017, University of Mannheim, Thomson Financials, client case studies Processing time Processing cost Error rates 12 days 0.5 days 5 € 1 € 3 % 1 % w/o SMACC with SMACC w/o SMACC with SMACC w/o SMACC with SMACC - 95 % - 80 % - 65 % Automation with AI results in drastic improvements within finance departments
- 10. Market value SME finance software in Germany 3bn € Spending for manual processing of SMEs in Germany 43Bn € x14 Source: Deutsche Bank, PWC, destatis, own assumptions Automating the finance department is a huge market opportunity
- 11. Movinga case study: things are moving Starting point mid 2016 • ~ 1‘000 paper based invoices per month • Processing time of ~ 20 days • Error rates ... (high) • No transparency in payment process • Cost per invoice > EUR 10 • Accounting team of 8 FTE Smacc impact today • > 3‘000 invoices per month all processed with SMACC • Processing time < 12 hours • Error rate < 1% • Cost per invoice EUR 1 • Accounting team of 3 FTE COMPANY PROFILE • Leading online moving marketplace • Headquartered in Berlin • 250 employees • Active in 6 countries • Hyper growth • Troubledwatersandrestructuringoverlastmonths
- 13. • Integrations • Refactoring platform for the growing number of customers • Support new use cases, more more APIs Next Steps
- 14. Thank you! December 8, 2017 Wojciech Barczyński wojciech.barczynski@smacc.io www.smacc.io