RPA Vendor Evaluation and Selection using AHP and Kano Model

Robotic Process Automation (RPA) as a technology has gained momentum due to the onset of Industry 4.0. RPA is considered to be the simplest forms of automation in which, typically, the human actions are mimicked without taking into account the complex judgements associated with it. Hence, RPA is deemed fit for the rule-based tasks which are highly repetitive, bulky and are error-prone. Traditionally, organizations looking to increase its operations efficiency improve workforce utilisation, turn to RPA. With the shift in working pattern enforced by COVID-19, RPA adoption is no more a luxury, but a necessity. In a short span of time, the number of RPA vendors, that is the developers and providers of RPA software have significantly increased and due to current demands of RPA, it is expected to grow further. This increase in choices makes the vendor selection part of any RPA Implementation Project highly complex and confusing. Keeping in mind the significance of the vendor selection process, various methods of RPA Vendor Selection have been previously proposed but it is still evolving along with the changing needs of the Businesses. This Research paper aims to propose a standard model using Kano Model and Analytic Hierarch Process (AHP) such that it can be customized during each new RPA implementation project without compromising on the basic structure and method for evaluation.


Introduction
RPA is a boon of Industry 4.0 and it is a "relatively new phenomenon as it started getting traction at the end of 2014" (Noppen, 2019). RPA is "driving much of the long-tail process automations that were previously impossible to achieve" (Gould, 2018). Further, the COVID-19 outbreak has made it a necessity. According to the IDC survey "Wide recognition of the value of digital transformation and information technology among all employees" (Xie, 2020) is one of the top positive impacts of the pandemic and nearly "65% of industrial users have formulated plans for working from home"(Xie, 2020) due to the outbreak. "RPA uses software to automate tasks previously performed by humans that use rules to process structured data to produce deterministic outcomes. It automates the repetitive, largely physical, clerical tasks typical of much office work." (Willcocks, 2020) Further it has been confirmed that for "tasks that are largely driven by rules, schedules, or events, a robot can take the wheel and get the job done." (Scheppler & Weber, 2020). As discussed in the report by (Gurwitz, 2020) "For companies wrestling with the challenges of COVID-19, RPA offers wideranging benefits. Every organization has clerical, time-consuming tasks that demand accuracy and speed, but don't require decision-making to accomplish." Hence RPA is the most common form of automation that companies are quickly adopting and "its (RPA) successful implementation across various sectors during COVID-19 has embarked on the upsurge in its future demands undeniably" (Srivastava, 2020).
According to (Bygstad, 2017) RPA is an example of "lightweight IT" because "it's deployment is frequently done by users or vendors, bypassing the IT departments". Thus, with the upsurge in demand, the RPA ecosystem is bound to change drastically as the main entities of this ecosystem are the "RPA tech providers, RPA implementation partners, RPA marketplaces and RPA innovators" (AIMultiple, 2020). RPA technology providers are the developers of software bots and RPA Applications. RPA implementation partners use of these RPA apps to develop process automation solutions for companies. RPA marketplace is hub of reusable RPA solutions. Reusability has the benefits like "Reduction in implementation time, reduction in programming effort, process improvement and reduced maintenance cost" thus a repository of such reusable solution are created. "These reusable solutions are provided in marketplaces built by leading RPA companies"(AIMultiple, 2020). Lastly, RPA ISSN: 00333077 6231 www.psychologyandeducation.net innovators are the companies that are currently engaged in building future breakthroughs like the no-code or self-learning RPA tools. The purpose of RPA ecosystem is enhancing the quality, variety, availability and utility of the RPA solutions keeping up with the demand and innovation. Considering that RPA adoption will create demand and RPA consultants play an important part in manipulation and determining the patterns in these demands, we can re-define the RPA ecosystem with three broad entities -(1) RPA vendors (collective group of RPA solution providers, implementation partners, marketplace and innovators), (2) RPA adopters (companies that implement RPA solutions), and (3) RPA Consultants (the facilitators of strategy, advice and expertise for the implementation projects). While implementing RPA, companies need to go through the RPA vendor selection process. The pace at which RPA vendors are increasing in numbers in remarkable. The Forrester report by (Le Clair et al., 2017) itself counted 38 RPA product vendors excluding the professional service firms that delivered RPA. The count has increased in last three years to more than 50 and is expected to plummet post COVID-19. This will add to the dilemma of the RPA adopters. Despite the dilemma, the vendor selection process should never be bypassed. In this research, traditional vendor/supplier selection methods used typically in manufacturing industry will be compared with RPA or other technology vendor selection. Existing methods and guidelines as established by research papers and articles of Consulting firms will be analysed. With the inspiration drawn from the Kano Model and the existing methodologies, a new model for RPA Vendor Evaluation and Selection will be designed which will be customer-centric, flexible and more accurate.

Literature Review
Vendor Evaluation, also known as "Supplier Evaluation", includes activities to "identify, evaluate and contract" suppliers (Taherdoost & Brard, 2019). For conventional industries like manufacturing, the objective is "to reduce purchase risk, maximize overall value to the purchaser, and develop closeness and long-term relationships between buyers and suppliers" (Taherdoost & Brard, 2019). It is also necessary for staying relevant in the market and attaining supply chain superiority. The two major issues that are to be taken care of while developing an effective vendor/supplier evaluation and selection model are -"evaluation criteria and the method to be used". (Fırat et al., 2017). Both of these have been discussed in detail.

Evaluation Criteria
A list of 23 criteria by(Dickson, 1966) known as the Dickson's vendor selection criteria were laid down. These criteria were reviewed during another research and it was found that "47 of the 74 articles (that is 64% of the articles that were reviewed) discussed more than one criteria and that in one article, 18 criteria are discussed." (Weber et al., 1991). Ideally, "the nature of the item to be purchased has a major influence on the factors that are considered when selecting a supplier" (Dickson, 1966 , 2017) pointed out that "One size never fits all" and that an Organization must avoid "blindly selecting an RPA tool" following their close competitor, most of the papers and corporate reports that were studied have either pre-mapped the existing RPA vendors into certain categories that create a bias. Others have provided a generic approach to select the vendors which are helpful but may not be holistic for a particular Organization.
In reality, the criteria can vary according to situation, industry, business type, business goals and many other factors. Thus, here we have focused more on what to go with the selected evaluation criteria rather than stating any hard and fast criteria. For this we have taken reference of the Kano Model.

Kano Model
Named after its pioneer Noriaki Kano, the Kano Model is a Quality Function Deployment (QFD) tool which "helps in setting direction and priorities for addressing customer needs" (Asif, 2015). According to the Kano model, there are three types of customer needs that can be mapped out on a coordinate system of y-axis representing 'customer satisfaction' and x-axis representing 'provision of product/service attributes' respectively. Based on that, the products or services can be said to have most and the least impact on customers. Hence Kano Model can be used to identify the "spoken" and "unspoken" customer expectations. (Fırat et al., 2017). "Spoken" expectations are the ones which the customer openly and directly states as wants.
Level of presence of these spoken expectations proportionally affects the level of fulfilment. "Unspoken" wants are classified as "Attractive" and "Must-be". "Must-be requirements are untold expectations, and if they are not delivered, this will lead to extreme dissatisfaction of customers and can result in complaints. However, if they are delivered, customers will only be in a state of 'not dissatisfied. Attractive requirements are the ones that the customers do not expect from the design essentially. If they are not delivered, the customers will not be dissatisfied. However, if they are delivered, customer satisfaction will be increased substantially" (Matzler et al., 1996). Precisely, there are the following five categories as given below in which the "Kano Quality model classifies the attributes of a product" ( Model could "generate different classification of quality attributes for the pharmaceutical logistics industry".(Chen et al., 2020). Using Kano Model could give insights to "Hong Kong Express to identify their service areas that needed to be improved and paid attention to increase customers' satisfaction in future" (Wong & Ho, 2019) "Kano model can be deployed to identify a wide range of complex patient needs and convey its potential usefulness in the continuous improvement of the healthcare sector." (Materla et al., 2019). It has also proven effective to identify "how different after-sales services quality elements affect customer satisfaction"(Shokouhyar et al., 2020). with a large number of growing vendors offering a variety of services in their RPA software package, setting the priorities and making the right decision is important. AHP being a mathematical model "provides a hierarchical representation that enables analytic decision-making" and is quite versatile in the fields of both qualitative and quantitative analysis. (Fırat et al., 2017). "The AHP method is flexible and allows development stakeholders to assign a priority (relative weight) to each factor through pairwise comparison" (Pesonen et al., 2001) "In AHP analysis, participatory consultation with stakeholders is an initial step for constructing indicators critical for attaining the overall goal and deciding on their corresponding weights"(Baffoe, 2019). Hence, in our case it is necessary that the stakeholders, that is the RPA adopters participate by giving preference scores to the different available options so that the goal of selecting the best fitting RPA vendor can be attained.

Evaluation Method
Analytic Hierarchy Process is carried out using following steps (Baffoe, 2019) -1. Determining the goal and the associated options (criteria, alternatives, etc.) on which decisions need to be made.
2. Give preference score to each options based on the Saatys 9-points scale (Refer Table 1) and construct a Pairwise Comparison Matrix. 3. Determine the consistency index (CI) as follows: CI = (λmaxn)/ (n -1), where n refers to the size of matrix that depends on the number of options you are comparing. λmax is the average of the weighted sum/priority ratio of each of the alternatives. 4. Next the consistency is checked.
Consistency ratio (CR), is referred to the ratio CI/RI, in which RI is the Random Index (see Table 2). The RI essentially depends on the number of alternatives being compared. The CR is the measure of the "consistency of judgments" and is expected to be less than or equal to 0.10. A greater value than this indicates inconsistencies which will prompt us to reassign the scores and repeat the steps. 5. Finally, the relative weights of the individual indicators are aggregated to generate a "vector of composite weights" for each of the alternatives and rank them accordingly.

Model Creation
Our proposed model has been inspired from the Kano Model and the AHP method, the former is for selecting, segregating, and deciding priorities qualitatively, and the latter is to do the quantitative weight assignment and relative scoring to build the model for calculation of RPA vendor scores and solve the underlying purpose of RPA vendor selection based on the score. The creation of equivalent model has been logically explained under the following 3 subheadings -

Model Overview
For both, Kano Model and AHP, the common ground is the criteria prioritization. Using the reference from the pre-defined categories of Kano Model as discussed in section 2.3, we will be creating customised categories for the requirements of the RPA adopters. Then, analysing the offerings of shortlisted RPA vendors for this evaluation process, we will categorize their offerings into these customdefined categories. Although the name and essence of the categories will remain same, every new RPA adopter will be given a chance to adjust the priority scores of these categories. This is how the model will be standard, yet customizable according to specific business needs. Apart from this, the features under each category will be given priority scores by the RPA adopters. Also, the number of features in each category will be taken into account during the scoring along with the feature and category priority scores. It is important to understand that we are trying to standardize the system of RPA vendor evaluation without compromising on the customisability of the model based on the diverse and unique needs of the business. The steps to be followed according to this model is given in Figure 2 which will be further explained in the upcoming sections. However, before explaining the new categories and detailed working, it is important to state the assumptions.

Model Assumptions
Following assumptions are to be considered:  The RPA adopters are sufficiently aware of the features they need either on their own or via consultation.  They can easily segregate the identified features into categories suggested in the model  An initial shortlisting of RPA vendors already done  RPA adopters can categorise the unconsidered features that are discovered only during Vendor analysis as per our model

Model Development for Criteria Categorisation
The new attributes have been designed based on the scenarios as explained below and also illustrated in the mapping diagram (figure 3  Persuaders (P): These are the features that the RPA adopters did not consider, but they will be delighted if they get these. These features are perceived as an added advantage and can compel the RPA adopter to consider the vendor. This category has a positive weightage depending upon the level of positive impact it creates.  Dissuaders (D): The presence of these features is perceived as a disadvantage. So, its presence in an RPA offering dissuades the RPA adopter from selecting that particular vendor. The weightage of this category is directly converted to its negative equivalent consider the dissuading effect.  Figure 3 shows the steps to be carried out for calculation of final vendor scores of the selected RPA vendors. In figure 3, step 2 to 8 gives a detailed process for the calculation based on the category priority score, feature priority score, as well as the number of features in a given category. Once the Step 1 of feature identification is complete,

Proposed Model for Calculation
Step 2: Categorize the features and assign score to the categories. Cg, Ci, Cp and Cd are the category priority scores generated using AHP based on the preference scores entered by the RPA adopter. The subscripts g, i, p and d denotes the categories Game Changer, Influencer, Persuader and Dissuader respectively.
Step 3: RPA adopter gives preference score to each of the features under a category to generate a priority score of the features using AHP. Example: If category 'G' has 5 features, preference scores will be given to each one of them. Then, based on AHP calculations, the model generates priority scores Gj for j=1,2,3,4,5 where j is the number of features in the category.
Step 4: Analyse each of the shortlisted RPA vendor and count the number of features corresponding to each category. Note that for category P and D matching does not apply as they consist of previously unconsidered features. Hence, these features can only be categorised into P and D. Also, post categorisation, step 3 needs to be repeated for category P and D features to find their priority scores. Example: Ngk -Number of features matching into G category for the kth RPA vendor Npk -Number of features categorized into P category for the kth RPA vendor Step 5: Calculate the Revised Category Priority Score (Rgk,Rik,Rpk,Rdk) by multiplying the category priority scores with the number of matching features of that category of every RPA vendor. Example: Let Ngk = 3, then Rgk = Ngk x Cg Step 6: For each RPA vendor, category-wise find the aggregate feature priority score (Agk, Aik, Apk, Adk) based only on the features present. This calculation is based on the type of category. It has been discussed more in detail in section 3.2.3.
Step 7: For each RPA vendor, now multiply the Revised Category Priority Score with the Aggregate feature priority score to get categorywise final priority score for each RPA vendor. Example: Final category G priority score for vendor 'k' is Fgk= Rgk x Agk Step 8: For each RPA vendor calculate the final score by taking sum total of all the final category scores Example: Final score for vendor 'k' is Vk = Fgk + Fik + Fpk + Fdk Based on the highest score, the RPA vendor can be selected for implementing the project of RPA adoption in the Organization.

Model Demonstration
To demonstrate the designed model properly, we need to first establish a premise or scenario and then proceed. So, we are first going to establish the context by taking up a pseudo RPA adopting organization, then do feature identification and categorization based on some corporate literature since RPA implementation is very much a concept discussed more in businesses rather than in academia.

Premise for Demonstration
Let us consider a pseudo RPA adopting organisation AB Pvt. Ltd. The management of this organisation had decided to create a Centre of Excellence (COE) for the diagnostic study and research of this implementation. After the complete analysis was done, RPA implementation work was to be started. The RPA vendor had to be selected for the same. The COE had done a market study of the existing research and had shortlisted 3 RPA vendors who could be considered for the evaluation. These are RPA Vendor 1, RPA Vendor 2, RPA Vendor 3, and RPA Vendor 4. Now, the evaluation process was to begin.  Table 3) Table 3 -Categorization of identified features Now that the categorization is done, we shall move on to the quantitative calculation part of the model and demonstrate the vendor score calculation.

Demonstration of the calculation of vendor score
Let us recall the steps as explained in Section 3.1.4. The Step 1 is already done, that is the identification of features and its categorization.
Step 2 Categorization is already done. Preference score to be assigned for the categories to get their priority scores. Step 3 Preference score to be assigned for the features of each categories to get their priority scores. Step 4 Analyse each of the shortlisted RPA vendor and count the number of features corresponding to each category. The additional features that were not considered are categorized into P (Persuader) and D (Dissuader).
Step 3 has to be repeated for these new features to get their priority scores.  Step 5 Calculate Revised Priority Score for every category based on the count of features matching for the category per vendor.  = 1-( G3+G4); Since the absence of G feature disappoints the RPA adopter  Ai1 = (I1+I2+I3+I4); As the vendor score should improve with presence of these features  Ap1= P1; Simple Sum to contribute a positive score  Ad1= (D1+D2+D3); Simple Sum to contribute a negative score, sign taken care of in step 5 itself Similar approach followed for all the vendors across all categories and the results are as below.

Table 12-Calculated Aggregate Feature Score
Step 7 For each RPA vendor, now multiplied the Revised Category Priority Score with the Aggregate feature priority score to get category-wise final priority score for each RPA vendor. Step 8 For each RPA vendor, calculated net score by taking sum of final scores across categories.

Results and Discussions
We can observe in section 3.2.3 that by step 8, the final vendor scores are available to us. So, we can select the vendor accordingly.

www.psychologyandeducation.net
It is recommended that the RPA adopter selects the highest scoring vendor because the scoring mechanism is transparent and extensively based on the initial preferences of the RPA adopter. However, the Organization may choose to go with the second best in case of any special scenario which could not fit into this model. As mentioned earlier, it is highly essential that the preferences are clearly identified and stated by the RPA adopter. Also, the scoring has to follow the Saatys scale and be consistent. This will be ensured through the conventional AHP calculation, that is by checking the Consistency Ratio (CR) which needs to be less than or equal to 0.1 in order to be consistent.

Conclusion
In this paper an evaluation model for RPA vendors has been developed by taking references from the Kano Model and Analytic Hierarchy Process. Both the models being versatile and usable in diverse areas of manufacturing and service industry, provided the perfect blend to our model. While Kano Model was used to develop the qualitative part of the model, like categorization of requirements into appropriate brackets and assessing RPA vendors for the presence or absence of the requirements, AHP proved to be a powerful tool is assigning preference scores to each of the categories and requirements so that we get a ranking of the RPA vendors based on the aggregate weightage of each requirement across all the categories. The flexibility of this model lies in the fact that, each new RPA adopter can assign different priorities to the categories and requirements to take the decision accordingly. This make the model suitable for any industry opting for RPA adoption. Another advantage of this paper is the fact that it could be extended to be used for any kind of supplier evaluation. Keeping in mind that there are many established methodologies to evaluate suppliers of physical goods, it is best left to the discretion of manufacturing organizations whether or not they wish to adopt this model. However, for any organization primarily targeting best fit RPA vendor for their well-planned RPA implementation project, can definitely use this standard yet tailored model.