Bachelor’s/Master's thesis
Master of Business Administrator, Business Management 2021
Tiina Henttinen
AI SUPPORTED PURCHASE INVOICE POSTING
– lessons learned from piloting project and future
perspectives
BACHELOR’S / MASTER’S THESIS | ABSTRACT TURKU UNIVERSITY OF APPLIED SCIENCES Master of Business Administrator, Business Management June 2021 | 51 pages, 7 appendix pages
Tiina Henttinen
AI SUPPORTED PURCHASE INVOICE POSTING
Traditional invoice processing is often the most time-consuming procedure in an organization’s finance department. When an invoice arrives for processing by a person working in the business, he or she manually adds the dimensions used to allocate the invoice, such as account number, cost center, project and VAT code, and other allocations depending on company policies. This step could take several minutes per invoice thus the manual processing of purchase invoices is an expensive and time consuming function for any organization. However, these challenges can be addressed through automation and artificial intelligence. The opportunities to use AI for invoices are practically endless. AI can automatically extracts information and evaluate invoice against order records to ensure that the payment is a valid one, check the invoice against VAT rules and make the necessary posting to settle the invoice.
There is a clear focus by the ledger services of the University of Turku on making the invoice process more efficient. For this reason, an artificial intelligence proof-of-concept pilot was launched in late 2020. The project was done in co-operation with Certia Oy, a company that specializes in the universities financial and personnel management. AI proof of concept is often the best way to test which set-up will yield the best results thus ensuring a successful outcome.
In the best case, pilot projects can lead to a broader policy transition in an organization.
The aim of this thesis is to demonstrate the value of the AI technology, but also to show the challenges linked to its adoption. In the target organization, the aim of AI was to improve the efficiency and ultimately eliminate manual transactions from invoice posting. This kind of gradual process of automating smaller parts with AI and building up capability and learnings are shown to lead to a better outcome.
Based on the study, from a technical and application usability point of view, the AI technology worked pretty well – applications was easy to use and it was embedded within existing SAP Martti Smart invoice system. However, when viewed from a purely efficiency point of view, the improvement from AI application piloted might eventually be marginal. Of course, it is possible for an organization to focus on where else AI concept can be applied, and to expand to encompass the AI to other ledger service functions. However, the AI application in question may not have applications as such in other parts of ledger service processes thus is might be difficult to extend the pilot across wider user base. One opportunity for the target organization to reduce costs and find new ways to add value into old processes is to create more synergy across purchase to pay (P2P) process. Interviews and direct observation revealed that there is currently a lack of organization-level software that would enable real-time access to orders, good or service delivery and pricing. Upgrading organization’s P2P system needs new software to work seamlessly together and with existing SAP. Moreover, automation the different steps from procurement to payment with AI is absolutely essential to optimize this process.
KEYWORDS:
Artificial Intelligence (AI), proof-of-concept, ledger service, purchase invoice
OPINNÄYTETYÖ (AMK / YAMK) | TIIVISTELMÄ TURUN AMMATTIKORKEAKOULU
Tradenomi (YAMK), liiketoiminnan johtaminen Kesäkuu, 2021 | 51 sivua, 7 liitesivua
Tiina Henttinen
TEKOÄLYN HYÖDYNTÄMINEN OSTOLASKUPROSESSISSA
- pilottihankkeen opit ja tulevaisuuden näkökulmat
Perinteinen laskujen käsittely on usein aikaa vievin prosessi organisaation talousosastolla. Kun lasku saapuu ostoreskontraan, laskulle lisätään manuaalisesti tiliöintiedot kuten kustannuspaikka, projekti- ja ALV-koodin sekä muut kohdennukset yrityksen käytäntöjen mukaisesti. Tämä vaihe voi viedä useita minuutteja laskua kohti, joten ostolaskujen manuaalinen käsittely on kallis ja aikaa vievä toiminto mille tahansa organisaatiolle. Näihin haasteisiin voidaan kuitenkin vastata automaatiolla ja tekoälyllä (AI). Tekoälyn hyödyntämismahdollisuudet ostolaskujen käsittelyyn ovat käytännössä rajattomat. Tekoäly voi automaattisesti poimia tietoja ja arvioida laskun oikeellisuus, tarkistaa tehdyt arvonlisäverokirjaukset ja tehdä muut tarvittavat kirjaukset laskun maksamiseksi.
Turun yliopiston ostolaskupalveluissa käynnistettiin loppuvuodesta 2020 tekoälyhanke, jonka tavoitteena oli ottaa tekoäly avuksi ostolaskujen tiliöinnissä ja siten tehostaa ostolaskujen käsittelyä. Hanke toteutettiin yhteistyössä palvelukeskus Certia Oy:n kanssa, joka on keskittynyt talous- ja henkilöstöhallinnon palveluiden tuottamiseen yliopistoille ja korkeakouluille. AI- pilottihanke on usein paras tapa testata, mikä tekoälyn sovellus tuottaa parhaat tulokset ja varmistaa onnistuneen lopputuloksen. Parhaassa tapauksessa pilottihankkeet voivat johtaa laajempaan toimintatapojen muutokseen organisaatiossa.
Tämän tutkielman tarkoituksena on osoittaa tekoälyteknologian arvo, mutta myös osoittaa sen käyttöönottoon liittyvät haasteet. Kohdeorganisaatiossa tekoälyn tavoitteena oli tehostaa ostolaskujen käsittelyprosessia automatisoimalla tiliöintityö. Tavoitteena on päästä tilanteeseen, jossa tekoälyn tekemä tiliöintiennuste voidaan hyväksyä ennusteen perusteella, ja siten poistaa manuaaliset tapahtumat laskujen kirjaamisesta.
Tutkimuksen perusteella tekoälytekniikka toimi tekniikan ja sovellusten käytettävyyden näkökulmasta melko hyvin. Sovellusta oli helppo käyttää ja se integroitiin olemassa olevaan SAP Martti Smart -laskujärjestelmään. Puhtaasti tehokkuuden näkökulmasta katsottuna parannus voi kuitenkin olla melko marginaalista. Tekoälykonseptia voidaan kuitenkin edelleen kehittää ja laajentaa sen käyttöä muihinkin ostolaskuprosessin vaiheisiin. Kyseisellä tekoälysovelluksella ei kuitenkaan välttämättä ole suoria käyttökohteita ostoreskontran palveluprosessien muissa osissa. Eräs mahdollisuus vähentää kustannuksia ja löytää uusia tapoja tuottaa lisäarvoa vanhoihin prosesseihin on P2P-prosessin yhdenmukaistaminen. Haastatteluiden ja suoran havainnoinnin perusteella voidaan todeta, että tällä hetkellä organisaatiosta puuttuu työkalu, joka mahdollistaisi reaaliaikaisen ostolaskun maksukelpoisuuden tarkastamisen. Tulevaisuudessa organisaation P2P-järjestelmä olisi hyvä päivittää saumattomasti toimivaan, tekoälyä ja koneoppimista hyödyntävään P2P-ratkaisuun, joka toimisi organisaation nykyisessä SAP toiminnanohjausjärjestelmässä.
ASIASANAT:
CONTENT
LIST OF ABBREVIATIONS 6
1 INTRODUCTION 7
1.1 Background 7
1.2 Purpose of the study 8
2 LITERATURE REVIEW 9
2.1 Invoice processing as part of financial management 10
2.1.1 Strategic objectives for invoicing 11
2.1.2 Improving the efficiency of the invoice process 12
2.2 Artificial Intelligence in a nutshell 14
2.3 The impacts of artificial intelligence on invoice management 16 2.4 Pilot project as a tool to introduce and test new AI-based technology 18 3 THE PURCHASE INVOICE PROCESS OF THE TARGET ORGANIZATION 20
3.1 The current purchase invoicing process 21
3.2 Automation level in the current process 23
3.3 AI assisted purchase invoicing process 23
4 RESEARCH METHODS 24
4.1 Qualitative interviews and questionnaire survey 24
4.2 Questionnaire Survey 26
4.2.1 Target group 27
4.2.2 Handling of data 27
4.3 Method of direct observation 27
4.4 Reliability and objectivity 28
4.5 Ethical concers 29
5 RESEARCH RESULTS 30
5.1 Result from interviews 30
5.1.1 Identifying the opportunities and determining the AI pilot project 30
5.1.2 Launching the AI application 33
5.1.3 Outcomes from the pilot 35
5.2 Results of direct observation 37
5.3 Results of the Questionnaire Survey 38
5.3.1 Data preparation and overview of analyses 38
5.3.2 General attitudes towards automation and AI 38
5.3.3 Getting started on AI development from the point of view of end users 39 5.3.4 The effect of artificial intelligence on work 43
6 DISCUSSION AND SUGGESTIONS 45
REFERENCES 50
APPENDICES
Appendix 1. Webrop questionnaire
Appendix 2. Associations between the number of years of employment in the
organization and general attitudes towards work automation and artificial intelligence
FIGURES
Figure 1. The Purchase to Pay process describes how to order, receive and pay for
goods or services 10
Figure 2. The different aspects of the purchase invoice processing process. 21 Figure 3. Handling of purchase eInvoices of University of Turku in the Martti system
(modified from Certia 2021) 22
Figure 4. (A) Pie chart for tenure with the target organization and frequencies of
responses to general attitudes towards automation (B) and AI (C) 39 Figure 5. Differences in perception and attitude towards AI pilot project. 40 Figure 6. The communication channels the respondents rely on 41 Figure 7. The respondents’ views on change management during the AI piloting 42 Figure 8. A)The supervisor's ability to lead change and (B) the most important
characteristics of a change leader defined by the respondents 43 Figure 9. Respondents’ assessments of the piloted AI application 43
Figure 10. The impact of AI on work 44
LIST OF ABBREVIATIONS
AI Artificial Intelligence
AP Accounts Payable
CEO Chief Executive Officer
EDI Electronic Data Interchange (standard format to exchange business information between two organizations
electronically instead of using paper documents)
ERP Enterprise Resource Planning
HTML HyperText Markup Language
JPG Joint Photographic Group
ML Machine Learning
NLP Natural Language Processing
OCR Optical Character Recognition
PEPPOL-BIS the Business Interoperability Specifications (BIS) for common eProcurement processes to standardize electronic
documents exchanged and validated through an open and secure network, between sending and receiving Access Points for public sector buyers and their suppliers across Europe and beyond.
PDF Portable Document Format
P2P Purchase to Pay
TIFF Tagged Image File Format
UBL Universal Business Language
UTU University of Turku
1 INTRODUCTION
1.1 Background
The integration of artificial intelligence (AI) into organization’s everyday operations is changing the operating methods and processes in many areas. However, applying the AI-based methods in practice is not quite simple. A whole new kind of know-how and expertise is needed to cope with the change. The best economic benefits of AI are obtained when learning is started on time. It is clear that universal, general-purpose AI solutions are not expected for the financial management processes that would perform large-scale tasks independently. It is more likely that a set of different AI techniques and applications will be exploited for different types of purposes.
At the University of Turku, tens of thousands of purchase invoices are handled per year at SAP financial management system. Automation level is still quite low, and some tasks related to purchase invoices still need to be handled manually. However, handling invoices in an intelligent manner is high on the agenda for the organization’s finance department. In late 2020, an artificial intelligence pilot project to automate organization’s purchase invoice handling process with machine learning and AI technology was launched. The aim of piloting project was to automate routing and coding of posting information of the invoices and free time to more important tasks that could provide value to the organization. Where the classic invoice process consists of reading invoices and entering the invoice data into an SAP system by hand, Machine Learning (ML) and AI should offer intelligent solutions automating these repetitive tasks and to predict how to best handle an invoice. However, due to the complexity of booking the invoices, it can be a challenge to select the most suitable automation solution to tackle and improve the specific inefficiency in University’s invoice process.
The topic of the study is to examine AI especially from the perspective of the usability.
The key theme is to describe the development stages of AI in University of Turku invoice processes - what the organization has now and what the future looks like. The theoretical core of the research is the modern understanding of AI as a whole and the technologies related to its various aspects and their development. Detailed descriptions of various digital financial management processes and systems are excluded from the thesis.
1.2 Purpose of the study
There is a clear focus by the Financial Service of the University of Turku on making the invoice process more efficient. With an increasing amount of intelligent solutions available, the question arises which solution is the best fit to the organization. The main task is not only to enable the financial team to focus on more value-adding activities, but also to gain more experience of the suitability of AI in the organization’s complex and changing environment.
The financial service of the University of Turku wants to utilize the possibilities of digitalization in invoice processing, which is why an artificial intelligence proof-of-concept pilot was launched in late 2020. The pilot project was done in co-operation with Certia Oy, a company that specializes in the universities and colleges financial and personnel management. The piloting project started at November 2020 and lasted until March 2021. The aim of this thesis is to demonstrate the value of the AI technology, but also to show the challenges linked to its adoption. Moreover, the aim is to provide a framework for how target organization should continue to build up AI based invoicing process and to build an ecosystems of external service provider in the near term.
The research questions are as follows:
1. Lessons learned from AI pilot project. Whether the pilot project demonstrate AI’s validity and usefulness.
2. The effect of organizational factors on the possibilities of AI platform and the choice of the platform. How difficult would it be to implement the proposed AI solution—both technically and organizationally?
3. What are the incentives for organization to start adopting an AI powered invoicing system? Would the benefits from launching the application be worth the effort?
4. What to focus on in the near future?
2 LITERATURE REVIEW
The higher education system of Finland has gone through a broad reform and as a part of reform, university sector’s role has changed enormously from the beginning of 2010 when universities autonomous status was strengthened. The Universities Act (558/2009) changed the status of universities by giving them more procedural autonomy and separated universities from the state budget. The dominant ideas about the organization and governance of universities have reshaped with the change. This change has extended to the way in which organizational and decision-making structures within universities are built. Currently Finnish university can be considered as a stakeholder organization, where strong management structure and centralized services take care of day-to-day management instead of collegial decision-making. Government by academics that used to be based on collegial decision-making bodies have become integrated into the administrative line of the organization and thus become part of top- down decision-making structures. From the organizational perspective, organizing activities in a way that each of the departments performs a specialized function while constantly collaborating with each other to achieve organizations goals and values, leads to an efficient decision-making and processes.
Highly effective organizations exhibit strengths across five areas: leadership, decision making and structure, people, work processes and systems, and culture (McAfee &
Brynjolfsson 2012). Today’s leading organizations are using machine learning based tools and data-driven decision making, and they’re starting to experiment with more advanced uses of artificial intelligence (AI) for digital transformation. However, introduction and implementation of the new technologies into the organization’s existing practices is not simple. A whole new kind of expertise is needed to cope with the change and to know which problems to tackle. Moreover, when starting to implement AI into processes that need automating, it is important to realize that not all the new solutions are implantable into organization all at once.
In the following chapters, I will focus more on leveraging AI technology to handle invoices in an intelligent manner. In the first part, I will go through AI especially from the perspective of the invoice management process and the opportunities that AI brings to the invoicing process are reviewed. In the last paragraph, I will go through the
purchasing-to-pay process of the target organization and what elements the process consists of.
2.1 Invoice processing as part of financial management
Financial management is one of the important parts of the organization. Today’s financial management includes multidimensional functions, which are directly related with various departments like purchase, personnel, marketing and production. The deep embedding of technology in business in the late 20th century facilitated a rapid evolution in financial management practices and just like other finance processes, the Accounts Payable (AP) function was no exception. The AP can be defined as the part of the accounting cycle that consists of receiving, recording and paying suppliers invoices in an accurate and timely manner (Schaeffer 2013).
From a financial management perspective, the purchase invoice process starts when the purchase invoice is received by the company and ends when the invoice is paid, recorded in the accounts and archived. On the other hand, if the company's acquisition process is handled in its entirety, the process will start long before the purchase invoice is received. The entire process of ordering products, processing invoices, and making payments, also known as the P2P process (purchase to pay process), combines the processes of the procurement and entire supply chain within a company through the process of receiving the goods, and finally to the payment issued to the vendor (Figure 1). (Lahti & Salminen 2014, pp. 52-57.)
Figure 1.The Purchase to pay process describes how to order, receive and pay for goods or services.
Invoice processing is the method by which companies track and pay supplier invoices.
At its most simple, the process involves receiving an invoice from a third party, validating it as legitimate, paying the supplier, and noting the payment in company records (Lahti
& Salminen 2014, p. 53.) The process sound quite simply however, there are usually multiple people involved, and thus plenty of opportunities for error. In recent years, the invoice processing has become increasingly complex, as invoices are received through a multitude of different platforms, in both hard copy and electronic formats, via diverse channels including post, email and The Electronic Data Interchange (EDI) (in Finnish organisaatioiden välinen tiedonsiirto, OVT). EDI is an electronic transfer of data using an agreed standard to design an EDI message, which is compatible between sender and receiver (European Commission (EC), 94/820). This message is defined as a set of information which is structured in agreed formats and capable of being read by a computer and able to be automatically and unambiguously processed (Eurostat 2021).
Although invoices are often processed in specific invoice processing platforms that are integrated into the ERP system of the company or organization, the vast majority of platforms are ‘semi-structured’ and there are still lot of things that have to be entered manually. Such a procedure lead to high staff costs and furthermore it can be error-prone thus leading to issues such as invoice duplication, incorrect or repeated payments, paying the wrong sum and a host of other mistakes. In worst case, it can damage organization’s reputation and may result in late-payment fees.
2.1.1 Strategic objectives for invoicing
The Accounts Payable, which refers here to the business department that is responsible for making payments owed by the company to suppliers, has three main objectives: it generates the information needed to complete purchase transactions, it ensures that only the items received and listed in the purchase order are paid, and it provides information for decision-making (Sedevich-Fons 2020). °°All of these objectives help guide the overall accounts payable process. According to Beretta et al. (2002), the invoice process involves several external and internal users of the organizations, and the strategic goal of invoice processing can therefore be thought of as creating value for these actors.
One of the most significant objectives of the accounts payable process is the timely processing of vendor invoices. Time is one of three dimensions of so called “Iron triangle”
that are used to evaluate and balance the competing demands of cost, time and quality (Stojcetovic et al. 2014). The quality means that the financial data is accurately recorded
and timeliness of data flow of cash flow and cost refers to invoice-specific processing costs. These should all be linked to the strategic objectives of invoice processing. As the three dimensions are tightly connected to each other, increasing quality will increase the amount of time needed, which also will lead to an increase in cost. Therefore, it is important to clearly define the value creation strategies of the invoice process thus helping to identify useful areas for development and balance these three dimensions.
2.1.2 Improving the efficiency of the invoice process
Process efficiency is one of the non-financial metrics of performance. In the context of a company's processes, efficiency refers to how well the process uses the available resources to achieve the desired results (Laamanen & Tinnilä 2002, p.47). In today’s rapidly changing business environment, organizations need to closely monitor their operating costs to ensure that inefficient and resource-intensive operations are continually improved. According to Pastinen (1998, p. 48) process improvement should be seen as an approach to satisfy customers, employees and organization´s own needs.
The efficiency of the purchase invoicing process can be affected in several different ways; adding automation, using ERP systems' invoice processing modules, introducing uniform processes and accounting principles at the organization level and optimizing organization functions (Lahti & Salminen 2014, p. 58). Furthermore, purchase invoicing and purchasing processes should be set specific goals that guide operations and the achievement of which is monitored by process metrics. For example, the duration of the invoice cycle or the proportion of invoices paid late can be used to measure the overall performance of the process (Lahti & Salminen 2014, p. 59).
Creating synergy across purchase to pay (P2P) process is key when an organization wants to reduce costs and find new ways to add value into old processes. The time- consuming and complex steps in the end of the P2P process is the validation, booking and matching of an invoice with the correct order. Organizations financial department might spend valuable time and money in manually managing invoices, validating and reconciling invoices. With a large list of active suppliers, the complexity of creating the correct booking increases. Furthermore, the whole P2P process becomes more complicated when software used form procurement and invoicing processes are siloed and fragmented. The more complicated the structure of the organization, the harder it becomes to manage it efficiently. However, most organizations continue to use their
outdated solutions and still about 80 % of organizations use manual or semi-digital tools to manage their P2P cycle. (PayStream Advisors 2018.)
The goal of effective invoicing at any organization is to modernize and fully automate P2P processes as far as they can. The main problem for example with an ineffective invoice approval procedure is the hidden costs incurred with numerous manual steps.
This leads to a longer processing cycle and higher labor costs. The paper-based invoicing done in the earlier days has been steadily declining as the digital invoice processing; especially electronic invoicing (e-invoicing) has taken over the market. An electronic invoice (e-invoice) is an invoice that has been issued, transmitted and received in a structured data format, which allows for its automatic and electronic processing, as defined in Directive 2014/55/EU (EUR-Lex 2021). Europe and the Nordic countries in particular, have been at the forefront of developing e-invoicing to meet the needs of businesses and consumers. Various e-invoicing standards are used in Europe, such as the international UBL 2.1 and PEPPOL-BIS, but many other standards are used nationally, such as the domestic Finvoice and TEAPPSXML (Imposia 2021).
A structured e-invoice contains data from the supplier in a machine-readable format that can be automatically imported into the buyer's ERP system without requiring manual entering. e-invoice do not include a visual presentation of the invoice data although they can be temporarily rendered during processing or transposed into visual formats. For eInvoices, the visual format is secondary and the objective in automation is not to view the invoice. Unstructured invoice data issued in PDF or Word formats, images of invoices such as JPG or TIFF, unstructured HTML invoices on a web page or in an email, and OCR scanned paper invoices are not considered as an electronic invoice. (Koch 2019.) Because the benefits of the automation of invoicing are easily verifiable, organizations want to take advantage of automation throughout the entire process of ordering products, processing invoices, and making payments. Moreover, many organizations benefit from examining their processes, identifying and defining those that are ‘sub-optimal’ and then deploying business process automation to make operations more efficient and effective.
The goal is not to see AP just about invoice processing, rather to bring everything together and seamlessly combine the entire process of ordering products, processing invoices, and making payments (Koch 2019). Already now, AI and ML are used to deliver many benefits across the procure-to-pay cycle, for example capturing invoice and extracting data to coding and matching. AI and ML technologies add value by
accelerating processing times, reducing the need for manual effort, and eliminating errors.
2.2 Artificial Intelligence in a nutshell
As a concept, artificial intelligence is broad and multidimensional. Artificial intelligence is not a single entity, but a set of different artificial intelligence techniques, methods and applications that are suitable for different types of purposes. In computer science, artificial intelligence is defined as a component that focuses on the creation of intelligent computers and computer programs. On the other hand, even AI researchers have not very precisely defined the concept of artificial intelligence, as the definition of a science lives, and changes as new AI topics and applications emerge. (University of Helsinki 2019.)
The term "Artificial Intelligence" encompasses a set of applications that can act
"intelligently" such as Machine Learning (ML) and Natural Language Processing (NLP).
This implies that machines can perform tasks that require capabilities specific to the human brain, including reasoning, autonomy, searching for information and learning.
While the definition of artificial intelligence is reshaping over time, the two factors remain fairly constant and unite all artificial intelligence; autonomy and adaptability. Autonomy means the ability to do things without the constant assistance of the user and adaptability means the ability to develop functional ability through learning. One can also speak of weak (or narrow) and strong artificial intelligence. When a machine can solve only one problem, such as a chess program, we speak of weak artificial intelligence. The execution of operations is based on pre-entered instructions, according to which, for example, a chess program makes its move. Weak artificial intelligence is not able to independently assess whether the transfer was successful or failed - i.e. good or bad, but the program analyzes the situation based on the programming logic. Strong artificial intelligence, on the other hand, is capable of independent thinking. So far, strong artificial intelligence has not been developed. (Kumar 2018.) Adaptability refers to the ability of artificial intelligence to develop functional capacity through learning so that the system can function in a sensible way even if the task or situation changes unforeseen. In only a few applications of artificial intelligence is the situation so narrowly limited that all situations can be identified and programmed into the system in advance. (Ailisto et al.
2018, 47-48.)
Levels of AI autonomy can be divided into assisted, augmented, added and autonomous intelligence based on the “intelligence” of the system (Rice 2020). In the most basic level of AI, assisted intelligence, routine tasks are automated, and there is little autonomy in the system. This kind of AI often performs tasks that are more mundane; thus, it frees people up to perform more in-depth tasks. Assisted intelligence works only with clearly defined inputs and outputs as it needs constant human input and intervention (Rice 2020). At the next level of AI is augmented intelligence. This AI focuses on the technology’s assistive role, thus augmented AI is designed to enhance, rather than replace, human intelligence. The system assists in the performance of tasks, but the level of autonomy is still quite low. When the system helps to perform tasks and make decisions, it is a system of added intelligence. If the system makes decisions automatically without human intervention, i.e. a high level of autonomy, we speak of autonomous intelligence. This is the most advanced form of AI as machines, bots and systems act on their own, independent of human intervention. The fact that AI needs this kind of autonomy – independency of human interventions, can result in organizations never being willing to hand total control over to machines. Additionally, when considering autonomous intelligence, it is important to realize that it is not a good fit for all applications. Sometimes it is better to consider AI only as an automated advisor and let people to keep the responsibility of accepting and implementing decisions made by technology. This is particularly relevant when more qualitative and intangible factors such as positive work environment must be considered. (De Cremer & Kasparov 2021.) Technologies based on AI, or efficient algorithms, are already present in many industries to make business more efficient and enable new business. Applications include intelligent manufacturing robots, automated quality control, networked sensors for proactive maintenance, and real-time supply chain and its monitoring and optimization.
In the financial sector, the use of artificial intelligence will increase in the future, and banks in particular will increasingly use it, for example, to anticipate fraudulent attempts (Maruti Techlabs 2020). Several large banks have said they are already making major investments in AI technology and its applications. The goal of companies is to develop their operations as a whole and automate certain work steps. In the future, human creativity will only be needed to resolve exceptional and special cases. In the future, the implementation of artificial intelligence in financial sector processes and strategies will also use the so-called an extended approach (Intelligence Augmentation, IA) designed to use machine learning techniques to help - rather than replace, people (Pratt 2016).
With automated processes, data related to the company's operations is collected and utilized in real time. Increasingly, CEOs and company boards want to utilize real-time key figures in their own decision-making and monitor the effects of decisions made in near real time. In the future, the importance of mass data analysis (Big Data) will increase. For example, public data can be combined with up-to-date data produced by financial administration, and this information can be utilized in making estimates and forecasts (Aunimo 2017). Overall, advanced analysis and real-time financial data visualization help make decisions and improve organizations performance.
Although an ever-growing number of AI-applications are available on the market, most organizations are not yet actually using AI. For those using AI, they are mostly using built in-house machine learning systems.
2.3 The impacts of artificial intelligence on invoice management
As mentioned earlier, invoice management is often the most time-consuming procedure in an organization’s finance department. When an invoice arrives for processing by a person working in the business, he or she manually adds the dimensions used to allocate the invoice, such as account number, cost center, project and VAT code, and other allocations depending on company policies. This step could take several minutes per invoice. The time spent in processing manual purchase invoices is an expensive expense for companies. This expense is often not recognized. Most companies are used to processing purchase invoices and have accepted it as a manual process. In addition to this, the time spent processing purchase invoices is directly out of the work that creates value for the business. Thus, there are many incentives for organizations to start adopting an AI-powered invoicing system. The specifics of how AI will change the way AP functions are still under dispute. However, it is safe to say that the benefits of higher efficiency, faster processing times and more powerful data insights are inarguable effects from the oncoming AI revolution.
Already now, there are advanced accounts payable automation solution available and it is unlikely that AI will replace them. Rather, as AI becomes more sophisticated and widely available, the leading automation tools will remain on the cutting edge of technology and integrate AI into their current technology to become even more powerful.
A good example is the development work that has taken place in optical character recognition (ORC). During the 1980’s, the rise of computers in the enterprise sector
marked a big change in data capture when OCR was introduced for extracting data from invoices. The early versions of OCR had to be trained with images of each character and were limited to recognizing one font at a time. In the early 2000s, OCR became available online as a cloud-based service, accessible via desktop and mobile applications. (Krumdieck & Southon 2015.)
Today’s OCR technology is capable of recognizing most characters and fonts to a high level of accuracy. It recognizes text from image-based invoices, extracts and converts the images and the text included in them into machine-readable text data (Somani 2019).
The digital invoices are stored in an electronic archive where they can be retrieved, for example based on vendor or billing information. In this case, purchase invoice inspectors and approvers have an electronic archive of their invoices, so their own paper archives or invoice copies are not required. (Lahti & Salminen 2014, 54.) Even though the technology continues to improve, there is always scope for errors thus human intervention is needed. Moreover, although OCR technology has ability to “learn” and develop to reading specific types of invoices, its features are not enough in the operating environment where the types of invoices received are constantly changing (Larsson &
Segerås 2016).
In the case of intelligent invoice processing, the OCR should be fine-tuned to identify not only the letters and numbers but also the structure of the invoice. Therefore, the software must be able to mimic the human skills of recognizing mandatory fields (invoice number and series, net/gross amounts, tax, etc.) correctly and any other information which might be relevant (Somani 2019). In order to increase automation and digitization while also contributing to the quality, quantity, and availability of their data, more sophisticated invoice processing and document data extraction tools are needed. AI, specifically cognitive data capture, bring document data extraction to a new level of efficiency and efficacy. Cognitive solutions combine human capabilities in a single system. They identify patterns and handle massive amounts of data. This means that cognitive data capture systems process documents much the same way as humans do. The data are classified and extracted without seeing the document format, rather the layout is examined, clues are found and a document type is ranked. Cognitive data capture is capable of processing not only huge amount of structured data but also unstructured data. The cognitive data capture systems are adaptive and interactive and they simulate human thought processes to find solutions to complex problems. Cognitive data capture is best
fit the big companies, which have many vendors and invoice format changes frequently as the neural networks, learn and get adapted to different invoices. (Zheng et al.2017.) OCR has long been the only way to turn printouts into data that could be processed by computers. However, e-invoice submission now offers a far superior approach. AI and ML can take this a step even further and accelerate extraction and validation of the invoices. In processing purchase invoices, AI is a self-learning and constantly evolving probability calculation engine (Heath 2020). Initially, it is trained for its task by giving it purchase invoice and billing history data. Based on the historical data, the artificial intelligence builds rules for the posting and routing of purchase invoices in its own model and it predicts the accounting dimensions and the circulation completely automatically.
The AI assisted invoicing provides all the key elements needed to reach high level of automation. In an ideal setting, there would be no need for human intervention, which frees employees from repetitive low-level work and let them concentrate on added-value activities. The one benefit, which is not fully attainable yet, is eliminating human errors due to repetitive work. Furthermore, when using automated tools, invoicing becomes more accurate, as the machine takes only seconds to fill in all the fields compared to humans who usually only fill in the mandatory ones. (Heath 2020.)
As AI for invoice processing advances, it will have more features such as fraud detection possibilities, predictive spending patterns, and auditing expenditure. The opportunities to use AI for invoices are practically endless: from clearing invoice payments and prioritizing them based on a set of internal rules to assessing financial risks by looking at the balance. As with other major technologies such as RPA, Internet of Things, chatbots, and Blockchain, AI gives us a glimpse into a new future for invoicing services, with a strategic change and high value-added tasks such as influence and creativity that rely on purely human knowledge.(Sammalkorpi & Teppala 2019.)
2.4 Pilot project as a tool to introduce and test new AI-based technology
Pilot projects are seen as means to test innovations in a real-world situation and to develop knowledge about the interactions of the innovation and the context (Lee 1999;
Raven 2007). In the best case, pilot projects can lead to a broader policy transition in an organization. In addition to the common features mentioned above, six main characteristics can be identified that vary in their presence and nature across pilot projects. These characteristics include scale, innovation (level, driver, type), knowledge
orientation (monitoring type and intensity, knowledge and learning types, knowledge stance), special status (attitude, flexibility, resource allocation), relation to policy and local context (local dependency, connection to policy, incidence of occurrence), and actor network (initiators, participants, governance style). Every pilot has different ‘values’
for each of these characteristics, thereby making every pilot project a custom-made design for the policy domain to which it is applied. (Vreugdenhil et al. 2010.)
When an organization wants to increase the utilization of artificial intelligence, it is important to select the right AI pilot project. It is a vital to choose a company-specific project, so that internal stakeholders can directly understand the value. Furthermore, for the initial project, it might be good to start small and do the pilot project with external partners to bring in AI expertise quickly. The pilot project does not have to be the most valuable AI application as long as it delivers a quick win and creates value. It is important to develop a pilot project upfront about how specifically an AI system will create value thus convincing internal stakeholders to invest in building up organization’s AI capabilities. For example, in purchase invoices, the success of AI pilot project can add value by reducing costs.
Before executing on an AI pilot, one should examine the specific tasks that people are doing and identify if any of these can be automated. For example, the tasks involved in a financial assistant job may include creating, sending, and following up on invoices and collecting and reviewing data for reports. Rather than trying to automate their entire job, it is wise to consider if just one of the tasks could be automated or made faster through partial automation. It is important to realize, that despite the popularity and promise of AI, there is a good chance that the problem to be solved does not require a complex AI solution. For this reason, it is crucial to ensure the problem truly requires AI and ML before starting an AI pilot project.
If an organization outsources AI piloting to third parties, for example the third party builds the machine learning models to solve the problem in question, it is crucial to have someone who can work cross functionally by bridging both AI and organization’s experts.
This ensures that when the project is successful, it will influence the rest of the organization. It is good to realize that that AI piloting is not about building an AI startup, rather the goal is to build a successful project that will help organizations to gain firsthand knowledge about AI and what it takes to build an AI product.
3 THE PURCHASE INVOICE PROCESS OF THE TARGET ORGANIZATION
The financial services of the University of Turku, as well as 11 other Finnish universities are produced by Certia Oy. Certia Oy is a shared services organization owned by universities, and it is specialized to producing financial and personnel administration services as well as expert services regarding the introduction and maintenance of modern IT systems to universities. Certia Oy was founded in 2008 in response to the decree on the Service Center of Universities issued by Finnish Government (Valtioneuvoston asetus Valtion talous- ja henkilöstöhallinnon palvelukeskuksesta 229/2009). Once The Universities Act (558/2009) came to have legal force and effect in 2010, Certia became a limited company and began providing services to the universities and polytechnics. Certia's financial management system includes accounting, ledger, sales invoicing, fixed assets, budgeting, cash management and reporting functions (Figure 2). It is one of the largest SAP systems in Finland and has more than 20 000 end-users. The purchase invoice processing process of the target organization involves SAP-based function (Figure 2; the grey box) and non-SAP functions (Figure 2; green and orange boxes). Smart invoice Martti is an integral part of SAP enterprise resource planning Central Component (ECC). Martti directly uses basic SAP information such as the charts of accounts and the vendor register. Certia Oy tailors a service agreement - in co-operation with the customer, which fits customer’s specific needs. This service agreement can be adjusted and updated to correspond to the needs of the customer.
(Certia 2021.)
The concept of shared services differs from outsourcing and centralization. Shared services operates as an internal customer service business and it is used by multiple parts of the customer’s organization. Typically, shared services charges for services provided based on service agreement. The work or part of it remains within the organization. The concept of shared services demands that the systems used are fully integrated with the organization’s systems and processes. Shared services is not a new thing as organizations have been implementing it since mid-1980s. Outsourcing is the consolidation of services to external third party companies for execution. Centralization on the other hand is the process by which activities involving planning and decision- making within an organization are concentrated to a specific leader or location. (Miskon et al. 2009.)
Figure 2. The different aspects of the purchase invoice processing process of the target organization.
The P2P process of the target organization involves a number of sequential stages needed to be executed in a strict order. Although each of steps are important in the overall process, only the purchase invoice processing process is reviewed here in detail.
3.1 The current purchase invoicing process
At the University of Turku, all purchase invoices are processed at the University's centralized ledger services. The ledger services are divided into four service teams (UTU2-5) with a total of 11 financial professionals (Interviewee 1, oral communication).
Purchase invoice services include accounting and factual verification of domestic and foreign purchase invoices based on order information. As a whole, the purchase invoice process includes multiple functions that presented in Figure 3.
The domestic purchase invoices arrive mainly as eInvoices and the operator transmits the invoice material to the Martti, which is an integral part of SAP system. The invoice material contains the university's Electronic Data Interchange (EDI), which directs the
invoices to the correct university (Certia Oy 2021). At the University of Turku, Certia’s responsibilities in the purchase ledger include completing and checking the basic invoice information, maintaining a vendor register, checking payment reminders and credit memos, clarifying incomplete reference and banking information from the supplier, routing invoices for processing and payment (Figure 3). At the invoice check-in stage, the basic invoice information is automatically supplemented on the invoice if the supplier remains identified. The SAP system compares the VAT ID and bank account information from the supplier register with the invoice presentation. Any automatic posting in use are stored in the invoice data during the check-in phase.
Figure 3. Handling of purchase eInvoices of University of Turku in the Martti system.
(Modified from Certia 2021)
The invoice is relayed from the service center Certia to the reviewer at the University ledger service via Martti system. The process is scheduled to run automatically. The reviewer checks that the invoice match against the good or service ordered and sends the invoice to the approver (Figure 3). The reviewer and the approver of the invoice cannot be the same person. The invoice is allocated to a cost center, internal order or a project number. In addition, general ledger account number, the VAT code and the approver of the invoice is marked on the invoice. When goods have been bought from another EU country, an intrastat form is also filled in the Martti system. In order to comply with VAT laws, purchase invoice must contain mandatory notes, for example information concerning invoice date, invoice sequencing number, rate of VAT applied to each item etc. Furthermore, the universities' own purchase invoice processing processes also set requirements for the content of purchase invoices. (Turku University 2021.)
3.2 Automation level in the current process
In the current purchase invoicing process, invoices requiring accounting information are routed to the users for entering the distribution information. In order to manage the routing of invoices, predefined assignment rules are generated with Business Rule Framework Plus (BRF+). BRF+ provides a comprehensive application programming interface (API) and user interface for defining business rules (to create calculations, validations, and decision logic). The primary purpose of a rule is to separate the business logic from the system logic. This allows changing the system behavior without making significant changes to the code. The generated business rules are incorporated into SAP program. Moreover, in the current invoice processing, the suppliers’ information is automatically collected into the supplier portal.
3.3 AI assisted purchase invoicing process
Today, most domestic invoices addressed to the University of Turku are eInvoices.
Despite this, most of this invoice processing is done semi-automatically. A need to raise the level of automation of purchase invoice processing is evident. As the handling of purchase invoices is an ideal task for AI, an AI pilot project was launched to find out whether AI could be applied to somewhat automate the purchase invoice processing process.
In the AI solution being piloted, AI modifies the invoice before the invoice material is send to the Martti for approval and verification. The original idea is to use data from historical records for AI to predict how to match future invoices more relevantly. With the help of historical invoice data, the AI piloted should be able to predict the accounting dimensions automatically. Even though certain AI assisted functions do not require 100% accurate model to be successful in the operating environment, in the purchase invoices a significantly lower accuracy is not desirable. It is clear that designing for 100 % accuracy is not perhaps even possible as AI uses approximate algorithms, which by definition have some error rate. When automating the purchase invoice handling process, AI solution that can predict invoice routing and other dimensions with ~ 90% accuracy, is very good tool when considered in light of business objective.
4 RESEARCH METHODS
A case study presents a type of study, which examines individual, group, institution, community to answer specific research question. Case studies have been found to be especially valuable in practice-oriented fields such as education, management and public administration (Yin 2003). The key characteristic of case study tends to be its basis on multiple sources of evidence that concentrates on reveal of reasons, opinions as well as underlying facts. The data collected enable diversified and robust view on the problem. Qualitative research in a case study could be done via qualitative interviews, documentation and archive materials as well as observations with focus on quality data collection (Swanson & Holton 2005, 339-41). According to Gillham (2000), qualitative research in a case study enables to investigate subject despite the lack of information available, examine subject in detail and get underneath of it and investigate the case from the perspective of involved people.
This thesis aims to elaborate the usability of digitalization, especially artificial intelligence in invoice processing of the target organization by analyzing the proof-of-concept project.
The aim is to reveal the required strategy that identifies concrete obstacles and opportunities where AI can create value in the invoice processing. In order to achieve the goals, qualitative case study research is implemented. The main sources of information for this thesis are the interviews with the staff of the case organization, the questionnaire sent to the financial secretaries, who are working with the purchase invoice system, observation of the actual invoice handling work and using the documentation available at the organization’s intranet.
4.1 Qualitative interviews and questionnaire survey
Qualitative interview, which involves asking questions to converse with respondents and collect elicit data about a subject, is one of the most common qualitative research methods. Edwards and Holland (2013) organize qualitative interviews into three categories; structured, semi-structured and unstructured. While a structured interview has rigorous questions, which are asked in the same order and the same way, with little flexibility available, a semi-structured interview is open, providing much more space for interviewees to bring up new ideas and own insight. Generally, in semi-structured type
of interview interviewer’s interest is to allow the interviewee to talk from his/her own perspective thus revealing how the interviewee understands the topic under discussion.
The third type of interview, unstructured interview gives to interviewee even more flexibility and freedom to express thoughts than semi-structured one. Whatever the type used, the major objective is to obtain comparable data from all subjects of the research.
(Edwards & Holland 2013.)
As the study is qualitative and requires deeper knowledge of how AI has been utilized so far, interviews and questionnaire survey are chosen as the most proper method to evaluate the stage of invoicing process and its automatization in case organization.
Concerning the interview part of primary data collection, its main purpose is to receive deeper knowledge about the artificial intelligence project implemented at autumn 2020.
The project was implemented as a pilot project in cooperation with Certia, and the piloting ended at March 2021.
To gather data, two financial experts of the target organization were individually interviewed in February 2021. The experts were selected by purposive sampling, i.e. the interviewees were selected and contacted because they were considered to have the necessary information for the research topic. Interviewee 1 presents herself as a Finance Manager of the Financial Services, which is responsible for producing the University’s centralized procurement and invoicing services. She holds many years of experience in the field of finance, accounting and SAP enterprise resource planning tool. In her current position as a Head of procurement and invoice service, she has overall responsibility for the smooth running of the organization’s procurement and invoicing services.
Interviewee 2 is a Financial Planner, and he is responsible for the organization’s ledger services. Current position includes managing of purchase invoice process and its development, especially the possibilities of digitalization of invoice processing. In addition, he serves as the supervisor of UTU service teams with a total of 11 financial professionals.
The interviews were conducted remotely via ZOOM platform due to the coronavirus situation. At the beginning of the interview, the host granted permission to record and the interviewee was told the purpose and confidentiality of the interview. Interviews were done in Finnish. The interviews lasted about one and a half hour (1:34 and 1:39).
Recordings were qualitative research material for the thesis. Recordings were transcribed into a typed-out word document. Clean verbatim transcription type was used
to capture everything that was said in an audio recording, but filler words (e.g. true, such as), repetitions, intermittent syllables and single pronunciations were omitted from the text. In addition to speech, relevant emotional expressions (e.g. laughter, movement, etc.) were spelled out. The spelling of all interviews followed the same spelling accuracy and logic. Clean verbatim transcription is used especially when one wants to analyze mainly the substance of the speech.
The typed-out word documents were carefully studied and at the same time, the aim was to reflect on what has been read. Reflection seeks to gain new perspectives and insights.
The aim was to get acquainted with the material and understand the actual content of the material. The analysis was performed by inductive data-driven content analysis, i.e.
without any theoretical presuppositions. This kind of analysis has its limitations such as it is a time-consuming process, and it requires in-depth reading and rereading of the typed-out word documents. The analysis also takes skill and practice to effectively analyze without bringing in persona biases, while producing useful data that can lead to a hypothesis about the phenomenon.
4.2 Questionnaire Survey
Testing new operation models and complicated unconventional setups before large- scale deployment is absolutely critical to the AI pilot project’s success. In order to identify successful implementation and potential biases of AI, information from respondents were gathered using a questionnaire survey. The aim of the questionnaire was to find out the trends the new AI driven purchase invoice system has brought to the organization. The main goal was to answer the questions; what will change in purchase invoice processing? How the change appears in practice? The questionnaire was designed so that straightforward gains and challenges that AI might bring in real settings were found out.
The query (appendix 1) was created using Webropol application. Webropol web service enables easy creation of queries from planning to result reporting. It is used for instance in collection of research, course feedback and different registration responses. The University of Turku has given independent access to all Webropol users to create and manage their own queries. Each survey owner has an independent responsibility to ensure the legality of their survey and compliance with data protection obligations. This
includes the information of data subjects (respondents) and the processing and storage of the personal data register.
4.2.1 Target group
The invitation messages and the link were shared via email to 10 financial secretaries working in the financial department. In the email it was stated where the information collected in the survey is to be used and in what way. The survey was completely anonymous and no implication of the identities of respondents could be found within the answers of the questionnaire. The respondents were informed of anonymity among other things as required by the Privacy Regulation (Webropol 2020). The questions and information provided in email were in Finnish.
4.2.2 Handling of data
The survey was open for 10 days and during that time, 10 responses (100 %) were collected. The data report was created and the data was analyzed and visualized with the Webropol 3.0 survey tool. The response material collected by the survey was kept in the Webropol service only until the completion of the thesis.
4.3 Method of direct observation
Observation could be described as a method of primary data collection, when researcher examines the subject of research by physically observing it. Although the author of this thesis works on a permanent basis at the case organization, the invoicing process is more or less unknown to her. As the author is not a member of an investigated environment, the method of question is not participant observation, rather it is direct observation.
As any method of data collection, direct observation has advantages and disadvantages.
Concerning advantages, direct observation method enables to observe factual acts and situations and obtain data in the most direct way possible. The disadvantage is that here, this part does not create much information, however it enables to study the actual invoice
process in its normal setting thereby providing a richer understanding of the process and its bottlenecks. (Gillham 2000.)
This thesis observation data was based on observation of organization’s employee invoice handling workflow, observation of movements of the invoice within organization, and observation of invoicing database available in the organization’s SAP systems. Due to the coronavirus situation in Finland in spring 2021, the observation was conducted online using Zoom platform. The author collected the data while the financial secretary, who worked remotely, demonstrated how the purchase invoices and other payment transaction documents are handled in the Martti system. Martti is an SAP environment Smart Invoice product of Bilot Oy and it is used via SAP Portal and it is owned by Certia.
During the observation, the author asked questions in order to understand the contextual details. At the beginning of the observation, the author granted permission to record the observation session. The session lasted approximately 2 hours (1:50) and it was done in Finnish.
4.4 Reliability and objectivity
As pointed out by Scholz and Tietje (2002, pp 2-4), case study should provide research process information in details with a purpose to increase reliability and objectivity of the work. Also, in order to achieve high level of reliability and objectivity, it is important to make questionnaires in a proper way. During the questionnaires’ creation, guidelines by Magnusson and Marecek (2015) are used. According to authors, interview questions should be clear and understandable and any difficult words, foreign languages and jargon should be avoided. Furthermore, it is important that questions are related directly to the topic of the interview and if possible, questions should be phrased as open-ended invitations for further discussion. In this way, interviewees have a chance to develop discussion by adding what they may want to say. It also encourages participants to speak in their own words. Primary data from interviews and questionnaires should be gathered consistently and analyzed in a systematic way.
4.5 Ethical concers
While collecting data by questionnaire survey, anonymity, confidentiality and consent of the respondents should be taken into account. For example when investigating sensitive issues such as employee’s engagement and motivation, the researcher must seek to minimize the possibility of intrusion into the autonomy of study participants. Firstly, the researcher’s responsibility is to completely inform participants of different aspects of the research; the nature of the study, the participants’ role, and how the results will be published and used must be expressed to participants in comprehensible language.
Second, the researcher should clarify how the research will benefits or otherwise affects respondents. (Warusznski 2002.) Many employees consider it necessary to participate in research that their working community may benefit from; however, employee should not feel any coercion to participate in a study. Thus, it is important to provide sufficient information in order the respondents to make decision on their participation. When answering questionnaires, the uncertainty felt by respondents for any reason, can affect the process of responding to questions and thus on the quality of the data.
5 RESEARCH RESULTS
In this chapter, the empirical findings collected from the two expert interviews, direct observation and questionnaire survey are presented.
5.1 Result from interviews
The aim of the expert’s findings was to gather opinion from professionals who are actively involved in the daily life of the target organization’s ledger services. This section summarizes the findings of the expert interviews through three different sub-topics.
Direct quotes are used to illustrate the views of the experts. Filling words have been omitted from the quotations, and then translated into English, however, so that the substance of the matter remains unchanged. The results observed in the study are reviewed based on the main themes of the interviews.
5.1.1 Identifying the opportunities and determining the AI pilot project
While AI is already transforming many areas within the private sector, public sector is cautiously employing AI and its segments. For instance, the impact of AI on the Finnish universities’ financial management is still small. However, there is reason to believe that the economic innovations created by AI technologies will soon have a very real presence also in the public sector.
Certia Oy, the service center for the 11 Finnish universities, provides financial and personnel administration services as well as system services to Turku University.
Certia's SAP system includes accounting, ledger, sales invoicing, fixed assets and furniture, budgeting, cash management and reporting functions. As a service provider, Certia Oy has a strong interest to positively transform the public service workforce by tackling labor-intensive tasks with digitalization and robotization. However, the use of AI in automating the financial processes is a relatively new thing and it involves many external and internal challenges. The external challenges relate to the policies and procedures of the university, and the internal challenges relate to the complex organizational structure of the target organization such as amount of sub-entities.
Despite the challenges, Turku University has own drivers of efficiency and university
management and central services aims to provide more services with less spending. In fact, Turku University seems to be a pioneer in the digitalization of invoice processes.
Turku University is innovative and is participating in these digital reforms. AI is not the only development target, but other process development projects are underway. (Interviewee 2)
In order to automate their invoice process, the finance department of Turku launched a custom-made AI piloting together with Certia Oy. It greatly increases the chances of success of this kind of piloting if top management in the customer organization (Turku University) understand that the development work is an important aspect of the organization’s operations. First of all, these people have authority and power to influence these matters and second, these people can communicate further and motivate others.
Interviews revealed that in the target organization the development work is supported by the top management and that the actual driving force is the CFO responsible for managing the financial actions of the Turku University.
“The CFO is the person who kind of started it by making it clear to people in Certia Oy, that we want to adapt an artificial intelligence into our processes” (Interviewee 1)
“Our CFO is open to all innovations. Whenever something new comes up she will give the green light as long as you give sufficient justification that the thing is worth doing” (Interviewee 2)
Next, Certia Oy characterized the core objectives of a pilot project and the problem that AI might be able to help solve. The AI assisted invoicing of purchase invoices was selected as the target of the pilot. It was important that the core objective of a pilot project was not to solve any major issues, rather it was meant to serve as a reference point for later implementations. Furthermore, since this was the first venture with AI, the duration time was set to 28 weeks. The first 10 weeks was reserved for planning and testing of AI application and the next 18 weeks for actual piloting. (Interviewee 1, oral communication).
“The university informed of its willingness to act as a pilot customer, however, the process to piloted and the technical implementation of AI all came from Certia, who chose external partner with the resources to effectively implement the AI
