improving the expense invoices posting process in LINK International Transport | Mindbox

Improving the expense invoices posting process in LINK International Transport

Customer: LINK International Transport Tool:  UiPath Date: 2020

Company LINK has automated the process of posting the expense invoices from its transporters thanks to the solution implemented in collaboration with Mindbox and based on OCR (Optical Character Recognition) technology from Abbyy and the UiPath software robots.


Customer:  LINK International Transport

Number of processes robotised: 6

Utilised solutions: UiPath

Tangible benefit:   75% (and increasing with time thanks to machine learning) correctly recognised and automatically posted invoices



Challenge: improving the expense invoices posting process 


Every day, LINK International Transport receives dozens and sometimes hundreds of invoices issued by its collaborating transporters. On average, this means around 50 000 invoices a year.  Their numbers peak at the beginning and at the end of every month, when the staff of the Invoicing Team was barely coping, spending all their time on manually rekeying the data from hardcopy documents into two IT systems.



Solution: invoice processing automation using a robot 


Hard copy documents first arrive at the secretariat where they are affixed barcode labels and entered into the system as scans, where the OCR technology recognises the data they contain. An additional difficulty is with the reverse-charge VAT invoices issued by foreign partners; this is also to be checked by the system. The robot also checks whether the invoice issuer is using cash basis accounting and also picks out the documents correcting already submitted invoices. The challenges the robot has to overcome include also the invoices that are expressed in euro, however, except for the VAT tax amount line that is to be expressed in PLN and has to be paid in to a separate account.


The invoices read using the OCR are routed as structured XML files to the communications interface of the forwarding system used by LINK International Transport, which stores all the transport orders. There, the XML file produced by the OCR system is supplemented with relevant fields.


“Most contractors have their proprietary invoice templates. Only a few have identical document layouts.The system has to check whether the invoice is actually intended for LINK International Transport as mistakes do happen too.Invoices are also checked for the currency of payment, contractor’s data, bank account numbers, amounts, dates, purchase order number in the system. An initial stage check is made to see if the specific contractor actually exists in the database. This is an anti-fraud measure – lists Krzysztof Kuczkowski Team Leader / RPA Developer at Mindbox.


The parameters read by the OCR include the purchase order number that matches a specific order to a contractor. Of course, the OCR cannot guarantee a 100% accuracy of recognition of data from invoices. Human involvement at this stage is necessary – the operator checks and validates the correctness of the data recorded. This ensures that the data reaching the invoice processing robot are correct.

An XML document so enhanced is then “handed over” to the robot that logs in to the SAP system and processes the day’s package of documents according to the rules. After retrieving an invoice, filling in the relevant fields and validating the bank account number against the system and the processing, the document is sent for approval. In the case of any mismatches and discrepancies between the data in the forwarding system and in the invoice, the robot signals the problem to the invoice processing department for verification and possible error correction. What is important there is that the robot collects information at every stage of the automated process, also during the OCR reading and supplementing the document in the forwarding system, and subsequently produces a report analysed by the business line employees.


“One of the challenges with automating this process was to implement a mechanism that would allow closing a fiscal month on a day other than the last calendar day of the month. For instance, this April was closed on the 5th of May. It was achieved with a configuration file where the month-closing day is defined manually. Based on this, the robot performs the postings – says Krzysztof Kuczkowski, Team Leader / RPA Developer at Mindbox



Benefits: speeding up the invoicing process, freeing up employees’ time for other tasks 


The automation process has allowed freeing the time the employees had to spend on invoice processing; now they can spend it on more creative, higher-value tasks. At present, most tasks are preformed automatically, the exception being only one-off situations exceeding the logical abilities of the robot. The current job of an employee when processing the majority of invoices is to verify whether the recognised characters correspond to the originals. In addition, the OCR notifies the operator in the case of any doubts with reading an invoice. Then the operator checks the field concerned and may easily substitute values.


Thanks to the system enhanced with the machine-learning algorithms, the robot keeps improving itself, expanding its skill set – the more invoices pass through the system, the more effective the model becomes. The error rate will be falling with time, though one can assume it will never be reduced to nil. The system is so smart that when a new supplier appears, it takes 3 to 5 documents for it to learn to correctly read all subsequent documents. This means that at the learning stage, the operator has to occasionally manually indicate a field in a document.


“Implementing an OCR (Optical Character Recognition) type solution was a breakthrough. Until recently, every scanned-in invoice was routed to the invoicing department. The employees opened the image of the document and rekeyed the data into two applications. Now their role boils down to just checking if the recognised characters match the originals and to approving the document. If they see an error, they just graphically mark up the region of the error. It’s a very user-friendly solution. Nevertheless, if the OCR has any doubts concerning any field, it will also notify the operator. Then the operator will examine the field and can easily substitute values – says Renata Kozyra, project manager at LINK.


Automating the invoice handling process is surely just a beginning of LINK International Transport’s adventure with the RPA. The success of its first robot has encouraged the company to automate more processes. Preparations are already under way to robotise two more processes at the invoicing department.