Analytic Hierarchy Process-Based Decision Making for Selecting Yearly Vaccine Production Capacity Under Demand Uncertainty

Pharmaceutical industry face challenges due to high research and development costs. These costs push companies to market products globally. However, global marketing faces strict, varying regulations per country. This is because products like vaccines are widely used in government-funded healthcare systems. Several governments regulate product pricing, creating further challenges. PT. XYZ, Indonesia’s primary vaccine manufacturer, plans to increase capabilities by building a new facility to serve global Hepatitis B vaccine demand. They face challenges in determining optimal capacity due to recent changes and perceived demand volatility. A new hexavalent vaccine product entering the market in 2027 adds complexity. Capacity changes involve significant risks, such as long validation processes. Root cause analysis shows the company lacks robust forecasting methods and depends heavily on Indonesian government demand. The thesis develops a decision-making framework using Monte Carlo simulation for demand forecasts and Analytic Hierarchy Process (AHP) for capacity decisions. Demand includes national and global markets for Hepatitis B vaccines: single-dose adult, pediatric, pentavalent, and hexavalent. Forecasts span 15 years with pessimistic, forecasted, and optimistic scenarios. Six alternatives are generated, ranging from 55 million to 80 million doses per year. Metrics are benchmarked from financial feasibility reports and interviews. Alternatives are evaluated using three criteria: absorbed demand, capacity aspects, and financial feasibility. Subcriteria include utilization rate, cost per unit, net present value, internal rate of return, and payback period. AHP assesses the relative importance of each criteria to determine the best alternative. 

Introduction

The pharmaceutical industry is a highly dependent on research, where companies usually collaborate with independent research and academic facilities to develop new treatments or knowledge towards diseases. Due to its continuous research and development (R&D) nature, companies tend to have high R&D costs. To cover the high R&D cost, companies aim to market their product globally. However, the industry faces strict regulations which differs per country as the products affect the country’s healthcare systems. Some countries implement regulations regarding product pricing, as the products available affects the overall cost in healthcare expenses for the country. This hinders the company’s ability to conduct further R&D due to gaining less profit (McGuireet al., 2007).

In recent years, the growth of pharmaceutical industry is seen in Indonesia, having the highest growth gross domestic product (GDP) after COVID-19 pandemic, with yearly increase of 8.10% in the first quarter of 2024. This surpasses the growth of manufacturing industry and national GDP (Pardedeet al., 2024). The industry also records the highest yearly production index in third quarter of 2023. The production index increases 19.71% from the second quarter of 2023 and 35.24% from 2022. The growth is also reflected in export values, with an increase of 0.8% from third quarter of 2022 (Paramuditaet al., 2024).

One of the products developed by companies in the pharmaceutical industry is vaccine, which are used for preventive treatments of diseases. Vaccines, however, are harder to manufacture than over the counter or prescription products due to the risk and cost involved in the process as it is used by countries in their mandatory immunization programs, requiring tighter regulations due to its usage for large population (Baylor & Marshall, 2012).

One of the vaccines included in mandatory immunization programs is the Hepatitis B vaccine. The Market Information Access to Vaccines (MI4A) database provided by World Health Organization (WHO) states there are three types of Hepatitis B vaccines that are majorly used, which are single dose vaccine for adult, single dose vaccine for pediatrics, and 5-in-1 pentavalent vaccine. The Hepatitis B vaccine market grows tremendously from 68.8 million to 414.5 million doses worldwide, an increase of 502% from its previous year due to the introduction of pentavalent vaccine and initiatives led by WHO to eliminate the disease as a public health threat, integrating pentavalent vaccine into mandatory immunization programs across the globe (World Health Organization, 2016).

The initiative to eradicate Hepatitis B continues as several WHO tenders for Hepatitis B vaccines are announced, including the first tender procurement for 6-in-1 hexavalent vaccine tender, planned in December 2023. The procurement however is postponed into 2025–2026, and it is noted that there are several additional procurements events, such as United Nations International Children’s Emergency Fund (UNICEF) hexavalent and pentavalent procurement in 2027, and Pan American Health Organization (PAHO) procurement for Hepatitis B, pentavalent, and hexavalent in 2026–2028 (UNICEF, 2023; Pan American Health Organization, 2025).

Business Issue

One of Indonesian vaccine manufacturer, PT. XYZ, mostly imports bulk Hepatitis B vaccines, then manufacture it into finished goods to be commercialized. Recently, they decided to manufacture bulk Hepatitis B vaccine, starting their research and development program with goals of being able to commercial globally starting 2028.

Feasibility study for the commercial facility have been conducted by PT. XYZ, and it was decided that the facility’s capacity will be 50–60 million doses with the basis of accommodating global demand. However, there have been discussions about the forecasted demand’s accuracy, due to UNICEF announcing the procurement of hexavalent vaccine starting in 2027. Thus, another forecast was conducted, and the decision for the capacity changes to 100 million doses. The decision for changing into 100 million doses sparks another concern, in which due to hexavalent have no prior market data for UNICEF procurement, the company also doubts the accuracy of the forecast. The volatility of the global market alongside frequent changes of managerial decision could be due to company’s perspective in facing the uncertainty of demands, as they did not know how their vaccines will do when it enters the global market.

The research questions are as follow:

1. What are the potential yearly production capacity alternatives for bulk Hepatitis B vaccine manufactured in PT. XYZ?

2. What criteria should be considered in selecting the optimal yearly production capacity for bulk Hepatitis B?

3. Which alternative is best suited to be applied in PT. XYZ?

Literature Review

As the demand is perceived as volatile, Monte Carlo method is used to determine future demand forecasts. Right now, the Monte Carlo method is described as one of the most powerful mathematical techniques due to its ability to predict the results of uncertain events. Its calculations analyze the risks involved based on historical value and probability distributions, and running large numbers of iterations of its calculations with different sets of random values. These iterations provide the possible future outcomes of the historical values (Cortés, 2023). To use Monte Carlo method in demand forecasting, (Przysuchaet al., 2024) divides the process into three steps: data preparation, testing and fitting distribution, and running a simulation.

To choose the best-suited alternative, the analytic hierarchy process (AHP) method is used. AHP refers to a problem-solving method utilized to face problems with multiple objectives that involves uncertainty (Goodwin & Wright, 2003). Further publications from Saaty (2008) and Goodwin and Wright (2003) breaks down AHP into several steps:

1. Problem defining: This step assesses the problem, and determines the decision-making objective.

2. Create decision hierarchy structure: This stage involves creating a hierarchy tree, with the decision-making goals located at the top of the hierarchy. Below the objective, criteria regarding the decision are stated, and if possible, broken down into several detail at the next level. The alternative actions are added at the lowest level of the hierarchy.

3. Make pairwise comparisons of attributes and alternatives: Each variable in the levels below the objectives are compared to variables in the level immediately below or in respect to it, to identify its importance within the levels.

4. Transform the comparisons into weights and check the consistency of the decision maker’s comparison: This stage converts the data gained from previous stages into weights which are normalized into the sum of 1. The approach uses an approach based on eigenvalues.

5. Use the weights to obtain scores for the different options and make a provisional decision: The weights in each path of the hierarchy are then multiplied together to obtain the results of each different paths, in which the scores of all the decision alternatives can be gained.

Conceptual framework in Fig. 1 aims to aid in the decision process towards the commercial facility’s bulk Hepatitis B vaccine production capacity. This research will be decomposed into three phases in determining the solution towards the issue stated, as well as utilizing several additional frameworks to aid said steps.

Fig. 1. Conceptual framework.

The first phase, inputs, starts with demand forecasting, will determine the total obtainable demand using Monte Carlo method for 15 future periods after the project is planned to start. The forecast conducted will be based on Indonesian and global demand of several Hepatitis B vaccines available. The forecasted periods will have three scenarios, which are opportunistic, forecasted, and pessimistic.

The second phase, process, will determine the possible capacity alternatives and the pairwise comparisons. Both of the aspects will be based on interview to stakeholders involved in the project. In this research, the criteria determined will be based on PT XYZ’s policy in decision making. The criteria then will be further analyzed using each demand scenarios and capacity alternatives using related frameworks to view each of their impact to the project. Furthermore, the results of the analysis will be a factor included in AHP for the pairwise comparison.

The last phase, output, will result in the best alternative to implement in the project from the AHP result. This phase will explain the details of the best alternative and provide a step-by-step framework available to be used for future project purposes.

Methodology

Primary sources of data for this research involves interview and questionnaire. Interviews are conducted with stakeholders related to the project, to gain information about the company in the context of supply chain process, company’s policies, current markets, future plans, project progress, and past previous researches related to the project. The second method, questionnaire, is used to determine importance ranking towards the criteria determined. Both methods are subjected to stakeholders relating to this project, which are members of the project management division and business development division.

Secondary sources are gained through literature reviews, reports, verified databases, industry benchmarking, and internal company records. The details are be broken down in Table I.

Table I. Information Regarding Secondary Sources

Data obtained	Description & Source
Indonesian population	Government census forecast
Historical demand	2024 MI4A Public Vaccine Dataset and Gavi forecast
Cost assumptions	PT. XYZ’s annual report
Financial assumptions	Damodaran risk premium and synthetic ratings, IDX risk-free rate and beta, Indonesian corporate tax rate
Price benchmarking	Vaccine raw materials suppliers
Project capital expenditure, adjustment on costs	Internal company data

External data such as literature reviews and reports are analysed to understand the current global and Indonesian market situation, while verified databases are used to gain historical data such as population, total purchases, coverage rates, and publicly available financial statement. Internal company records are obtained to gain information such as financial information, previous research records, and company policies.

The data gained will be analyzed using statistical and simulation methods. In conducting the simulation, RStudio is used as a tool, utilizing several packages: tidyverse (Wickhamet al., 2019), ggplot2 for data visualization (Wickham, 2016), and fitdistrplus for distribution fitting (Delignette-Muller & Dutang, 2015). Interview data will be quantitatively analyzed by evaluating the alternatives to each of the criteria, using AHP Priority Calculator from BPMSG (Goepel, 2018).

Result and Discussion

Demand Forecast

There will be two stages of the forecast: the first stage will consist of model testing which measures the model’s accuracy with historical data, and the second stage will be forecasting values for the year of 2028 to 2042. Each stage will contain the three phases of Monte Carlo method, which are data preparation, data testing and fitting the distribution, and running a simulation. Two datasets are obtained, national demand and global demand. The coverage data is taken from UNICEF & WHO estimates, and for future values, Indonesian government’s birthrate forecast is used as the base population. For global demand, the MI4A dataset used to determine several vaccine types: HepB Adult, HepB Pediatrics, Pentavalent, and Hexavalent.

There will be two datasets for model testing, which are the national HepB birth dose coverage and third dose coverage to determine the accuracy of the model created. Using R studio, the Hepatitis B birth dose coverage data from 2013–2022 is extracted using unlist function. Using the descdist function, the data is then subjected to distribution fitting by first analyzing its density and Cullen and Frey graph. The data summary shows that the values are relatively close as seen in the range, supported by the standard deviation that shows the data have less variability due to being concentrated around the mean, shown in Fig. 2.

Fig. 2. Summary of the historical data’s distribution.

The positive skew and high kurtosis indicate that the data includes heavy-tailed values that explains the higher mean than median value. The Cullen and Frey graph shows that the data is quite far from uniform or exponential distribution, and the possible alternatives may be either be lognormal or gamma distribution which have high skewness. The next step is to fit the distribution to the alternatives by creating a model using fitdist function. Goodness of fit tests are conducted between the lognormal and gamma models, which shows that in the three statistic tests Kolmogorov-Smirnov, Anderson-Darling, and Cramer-von-Misses test indicates a good fit due to its lower value. Thus, the model is then subjected for best fit selection, in which the lowest values of AIC & BIC is chosen, as shown in Fig. 3.

Fig. 3. Goodness-of-fit statistics for gamma distribution (1) and lognormal distribution (2).

Next, the second model testing will be conducted repeating the steps stated before, using the third dose coverage data. First, the third dose coverage is subjected to distribution fitting, as pictured in Fig. 4.

Fig. 4. Statistics of third dose distribution fitting.

As the median is higher than the mean, this indicates that the model’s values are quite spread, with the lower values affecting the mean more. This is supported by the slightly positive skew, as the low skewness indicates that there are more smaller values in the data. The kurtosis, however, indicates that the data may be a good fit for normal distribution’s kurtosis score of 3, but its lower score need further analysis.

Thus, the model is compared between uniform, normal, and lognormal distribution. The goodness of fit testing shows that uniform statistic tests show the lowest value, however, its AD test shows infinite value which indicates poor fit. Lognormal distribution is chosen as it has lower values than the normal distribution, shown in the goodness of fit testing in Fig. 5.

Fig. 5. Goodness-of-fit test for third dose.

Future period forecast is then conducted by generating random values that with the respective model’s parameters using rlnorm and set.seed function. The Monte Carlo method is done by running 5000 simulations, in which the results are then averaged to gain the future period’s value. Table II shows the comparation between the empirical data and forecast. The first dose coverage’s historical value difference is 10.15% and the difference from its upper value is 1.02%. Its third dose coverage’s forecast value is considered very near to its historical value, 8.88% from the base forecast, and 0.95% from the upper value forecasted.

Table II. Forecast Results of Model Data Comparation

Product	Lower forecasted	Forecasted	Upper forecasted	Historical
First dose coverage	74.49637	82.92744	91.35851	92.3
Third dose coverage	73.97703	82.91851	91.85999	91

Indonesian forecast will be conducted by using historical data from 2013–2023. The steps in model testing are then repeated to determine the values for 15 future periods. The distribution for both first dose coverage and third dose coverage remain the same, with minimal difference in the goodness-of-fit test. Future forecasts are then generated using Monte Carlo method with 100.000 runs. The forecast results are then calculated its 95% confidence interval by using standard deviation. For the first dose forecast, its upper value ranges in 92%, base in 86%, and pessimistic in 74%. For the third dose, its upper value ranges in 91%, base in 82%, and lower in 74%.

Global forecast will use the MI4A purchase vaccine database for two products, Hepatitis B adult and Hepatitis B pediatrics vaccine. For Hepatitis B adult vaccine, data of vaccine purchases from 2016–2023 is used. First, the adult vaccine data is analyzed to identify outliers. Boxplot diagram shows that there is one data point as outlier, thus, the data is cleaned, shown in Fig. 6.

Fig. 6. Boxplot of Hepatitis B adult vaccine data.

The data is then determined its best-fit distribution by analyzing its Cullen and Frey graph alongside their statistics. The distribution fit’s statistics shows slight positive skew of 0.13 and kurtosis of 1.8. From the graph, the distribution could be fitted into uniform, normal, and lognormal distribution. When fitted into the said distribution, the best-fit is using normal distribution. The steps are then repeated for the Hepatitis B pediatrics model. The data taken consists of no outliers, so distribution fitting is done. For distribution fitting, the Cullen and Frey graph shows slight skew, and goodness-of-fit tests shows that normal distribution have the best fit.

Then, the two models are subjected to the Monte Carlo simulation. Random numbers are then generated using each respective model’s parameters using rnorm function. The simulation is then conducted for 100.000 runs for 15 future periods. The values of each period are then averaged to get the forecasted values. For the adult vaccine, the forecasted value range is in ~12 million doses yearly, and for pediatrics, it ranges in ~84 million doses yearly.

As hexavalent vaccine will only enter the market after UNICEF procurement, Gavi’s base demand forecast for 2024–2030 will be the basis of analysis. The known data for future hexavalent forecast is shown in Table III.

Table III. Hexavalent and Pentavalent Vaccine Forecast, in Millions (Gavi, 2024)

Year	Hexavalent (million)	Pentavalent (million)
2023		229
2024	1	240
2025	9	227
2026	24	211
2027	36	203
2028	58	190
2029	92	168
2030	109	154

To know the basis of the assumed relationship, linear regression is conducted between hexavalent and pentavalent vaccine doses. The pentavalent acts as a dependent variable to hexavalent, thus, scatterplot is used to see the relationship. Fig. 7 shows the inverse relationship between hexavalent and pentavalent vaccine. The R-square value is 0.98, which tells that the relationship is significant.

Fig. 7. Linear regression of pentavalent vaccine to hexavalent vaccine.

Thus, from the regression, the equation for pentavalent vaccine from the coefficient’s intercept will be:

(1) $$\begin{array}{r}y=233807647.058824-(-0.740588235294118)x\end{array}$$

After obtaining the pentavalent vaccine equation, the hexavalent vaccine is then fitted into the best-fitting regression by testing linear, quadratic, and exponential models in RStudio. The quadratic model proves as the best fit, with r-square value of 0.99. This is further shown in the fitted value seen in Fig. 8. As the procurement tender is shifted from 2024 to 2027, the value for hexavalent starts in 2027.

Fig. 8. Quadratic regression of hexavalent.

The equation of hexavalent’s quadratic regression is then extracted:

(2) $$\begin{array}{r}y=1,904,762x^2+3,476,190x+-5,000,000\end{array}$$

The next step is to forecast the total obtainable demand for the Hepatitis B vaccine types. For national demand, as the data available is only birth and third dose, the assumption will be that the birth dose’s coverage will be the same as second those coverage. The third dose coverage will be the same as the fourth dose, and booster dose will be the same as last year’s fourth dose. For global demand, assumption of market share will be used based on the number of players. Market share assumption is as shown in Table IV.

Table IV. Global Market Share Assumption

	Pentavalent	HepB (adult)	HepB (ped.)	Hexavalent
Shares	4.92%	5.00%	7.92%	16.67%
Players (without PT.XYZ)	Actual	19	Actual	5

The calculated obtainable demand will be based on the assumption stated. After calculating the demand, it is obtained that the total demand for each scenario is shown in Fig. 9.

Fig. 9. Total obtainable demand.

Going forward, calculations will use the demand forecast as basis for capacity alternatives. As the data ranges from 29 to 98 million, several capacity alternatives are chosen based on interview conducted: 55 million, 60 million, 65 million, 70 million, 75 million, and 80 million. The baseline scenario will be chosen for future calculations.

Absorbed Demand Analysis

The baseline scenario will be chosen for future calculations. In calculating the absorbed demand, the first step is to calculate each alternative’s total production in the next 15 forecasted period, starting from 2028 as the first year of revenue generated. The total production will be divided by each scenario’s total demand to determine the absorbed demand of each scenario. The total of each capacity’s absorbed demand is then averaged, and normalized with over-capacity as a penalization point, shown in (3):

(3) $$\begin{array}{r}Adj=100\mathrm{\%}-(Absorbed;Demand-100\mathrm{\%})\end{array}$$

The results are shown in Table V, where the normalized values are derived from the baseline scenario:

Table V. Absorbed Demand Percentage

Capacity	Absorbed demand	Normalized
55 Million	84.83%	84.83%
60 Million	92.54%	96.18%
65 Million	100.25%	99.75%
70 Million	107.96%	92.04%
75 Million	115.67%	84.33%
80 Million	123.38%	76.62%

Capacity Analysis

For capacity analysis, the first step is to determine the cost assumptions. The price and cost assumption are shown below, in Table VI:

Table VI. Price and Cost Assumption

Type	Value	Source	Notes
Price	USD 2.31/dose	Industry benchmark
Production cost	28.87%	Financial statement	Variable
Operating expense	35%	Interview	Fixed

Cost calculation is done by calculating the PT XYZ’s revenue and calculating the cost using the percentage assumptions. As shown in the cost assumptions, operating expense is 35% of the maximum capacity’s revenue, and production cost is 28.87% of the realized revenue. The variable cost and fixed costs each year are then calculated the total, and divided into the quantities produced to determine the cost per unit. Each alternative’s average cost per unit and average utilization rate is calculated. The results are seen in Table VII.

Table VII. Average Utilization Rate and Cost Per Unit

Alternative (millions)	55	60	65	70	75	80
Utilization rate (%)	91.08	87.81	84.54	81.32	78.12	74.96
Cost per unit ($)	1.57	1.62	1.66	1.72	1.70	1.69

Financial Analysis

The first step is calculating the company’s cost of capital to determine the company’s discount rate. The components are described in Table VIII.

Table VIII. Key Assumptions for Cost of Capital

Type	Value	Source
Debt proportion	12.6%	Financial statement
Equity proportion	84.7%	Financial statement
Risk-free rate	6.99%	15-year bond
Equity risk premium	6.87%	Damodaran risk premium
Beta	0.69	2024 IDXHEALTH Index
Corporate tax	22%	Indonesian tax rate

Thus, the calculation of cost of capital can be done by multiplying the debt structure with cost of debt, added with the cost of equity multiplied by equity structure. The process is shown in Table IX.

Table IX. Weighted Average Cost of Capital

Calculation	Value
Proportion of equity * Cost of equity	10.25%
Proportion of Debt * Cost of debt	0.78%
Cost of capital	11.03%

After calculating the cost of capital, free cash flow to firm model is used to determine the investment cash flow in order to evaluate each capacity alternative’s investment value. Several key assumptions are as shown in Table X.

Table X. Financial Analysis Key Assumptions

Category	Value
Product price	USD 2.31/dose
Capital expenditure	USD 80,000,000
Cost of goods sold	44.03% of Revenue
Operating expense	35% of Capacity’s Max Revenue
Depreciation	15 years, straight line method
Corporate tax rate	22%
Interest expense	1.03% of sales
Changes in working capital	USD −489.76

The investment’s net present value (NPV), internal rate of return (IRR), and payback period (PBP) is then calculated, which shows the value in Table XI.

Table XI. NPV, IRR, and PBP of Alternatives

Alternative (millions)	55	60	65	70	75	80
NPV (Thousands)	$45,744	$34,849	$21,903	$7,278	$(8,936)	$(26,510)
IRR	18.29%	16.20%	14.09%	11.99%	9.91%	7.83%
PBP	5.72	5.96	6.36	6.99	7.97	9.57

Analytic Hierarchy Process

The AHP hierarchy tree is then made with the goal of selecting yearly production capacity. The hierarchy is divided into three levels, which are goal, criteria and subcriteria, and alternatives to describe the relationship between each factor in the decision-making process. Shown in Fig. 10, the criteria are absorbed demand (AD), capacity aspects (CA) which have subcriteria of cost per unit (CU) and utilization rate (UR), and financial feasibility (FF) with the subcriteria of NPV, IRR, and PBP. The alternatives ranges are 55 million up to 80 million yearly doses, with difference of five million increase per alternatives.

Fig. 10. AHP hierarchy tree.

A questionnaire is conducted to determine the importance ranking for each criteria and subcriteria to the involved stakeholders. There are ten respondents, which are stakeholders involved in the project and related experts in the company. The questionnaire is then analysed using BPMSG website for AHP to determine the global weight. The initial questionnaires results show consistency ratio of 16.3%, thus, adjustments using the website’s tools are used to increase its consistency. The consistency ratio decreases to 9.6%, within the acceptable rate of below 0.10 consistency ratio with details as seen in Fig. 11.

Fig. 11. Pairwise comparison of criterion and subcriterion of initial values (left) and adjusted values (right).

Pairwise comparison matrix is then conducted for each main criteria, which can be seen in Table XII.

Table XII. Pairwise Comparison Matrix for Main Criteria

Pairwise matrix				Normalized matrix				λmax	CR	CI
	AD	CA	FF		AD	CA	FF
AD	1.00	1.43	0.23	AD	0.17	0.17	0.17	3.00	0	0.00
CA	0.70	1.00	0.17	CA	0.12	0.12	0.12
FF	4.28	5.75	1.00	FF	0.72	0.70	0.71
Sum	5.98	8.18	1.41	Sum	1	1	1

After conducting pairwise for the main criteria, pairwise matrix for each subcriteria are conducted. Table XIII shows the pairwise comparison included in CA.

Table XIII. Pairwise Comparison Matrix for CA Subcriteria

Pairwise matrix (CA)			Normalized matrix			λmax	CR	CI
	CU	UR		CU	UR
CU	1	0.858	CU	0.462	0.462	2.00	0	0.00
UR	1.164	1	UR	0.538	0.538
Sum	2.164	1.859	Sum	0.462	0.462

Then, pairwise comparison of subcriterion for FF are calculated, shown in Table XIV.

Table XIV. Pairwise Comparison Matrix for FF Subcriteria

Pairwise matrix (FF)				Normalized matrix (FF)				λmax	CR	CI
	NPV	IRR	PBP		NPV	IRR	PBP
NPV	1.00	1.24	1.58	NPV	0.41	0.49	0.28	3.08	0.04	0.07
IRR	0.80	1.00	3.14	IRR	0.33	0.39	0.55
PBP	0.63	0.32	1.00	PBP	0.26	0.12	0.17
Sum	2.44	2.56	5.72	Sum	0.41	0.49	0.28

Thus, the global score for each criteria and subcriteria are determined by multiplying the main criteria score with the subcriterion’s local score, as detailed in Table XV.

Table XV. Global Rank for Each Criteria and Subcriteria

Criterion	Criterion score	Sub criterion	Local score	Global
AD	0.169			0.169
CA	0.121	CU	0.462	0.056
CA	0.121	UR	0.538	0.065
FF	0.710	NPV	0.391	0.278
FF	0.710	IRR	0.426	0.302
FF	0.710	PBP	0.183	0.130

The next step is evaluate each alternatives. Each alternative are compared to each other, categorized by criteria. The inverse score is calculated by dividing the alternative’s results with the other’s alternative results. Then, the score is normalized to determine its priority ranking and local weight. In comparing each alternative, assumptions were made where the higher values of absorbed demand, utilization rate, NPV, and IRR are preferred. In cost per unit and payback period subcriteria, lesser values are preferred. Thus, in criteria that have lesser values preferred, the values are divided by 1 before pairwise is conducted.

As the NPV subcriterion include negative values, min-max normalization method is utilized to be able to conduct the pairwise. Then, the normalization values are transformed into a 1-9 scale by using linear equation, with the highest value of NPV scores 9, and lowest scores 1, shown in Table XVI.

Table XVI. NPV Subcriterion Normalization

Alternative (millions)	55	60	65	70	75	80
NPV Subcriterion values	$45,744	$34,849	$21,903	$7,278	$(8,936)	$(26,510)
Min-Max normalization	1.00	0.85	0.67	0.47	0.24	0.00
Transformed (1–9 Scale)	9.00	7.79	6.36	4.74	2.95	1.00

Each alternative’s pairwise weights are then analysed to determine the best-fit alternative. Global score of each criteria and subcriteria are multiplied with the scores of each alternative in their respective aspects. The results of each multiplication are added to determine the score of each alternative, seen in Table XVII.

Table XVII. Score of Each Alternative

	AD	CU	UR	NPV	IRR	PBP	Score
55 Mill	0.027	0.010	0.012	0.079	0.071	0.026	0.22
60 Mill	0.031	0.010	0.011	0.068	0.062	0.025	0.21
65 Mill	0.032	0.009	0.011	0.055	0.054	0.023	0.19
70 Mill	0.029	0.009	0.011	0.041	0.046	0.021	0.16
75 Mill	0.027	0.009	0.010	0.026	0.038	0.019	0.13
80 Mill	0.024	0.009	0.010	0.009	0.030	0.016	0.10

Based on the analysis, the best fit alternative capacity for PT. XYZ is 55 million doses, with the highest score of 0.22. The second rank is 60 million doses, followed with 65 million doses. While the demand forecast increases as the years go on, overcapacity became a substantial factor in determining the best fit alternative. When the capacity is increased to 80 million doses, its utilization is low for several initial years, while the fixed costs for operations stays the same, which affect the financial feasibility of the investment.

Conclusions

The main conclusions of this research are:

1. The viable yearly production capacity alternatives are identified within 55 million doses to 80 million doses, with increments of five million doses. This is benchmarked from the demand forecast and stakeholder interview.

2. Criteria in choosing the yearly production capacity includes three aspects, which are demand absorption, capacity aspects, and financial feasibility. Capacity aspects include utilization rate and cost per unit, while financial feasibility includes NPV, IRR, and PBP.

3. The best-fit alternative for the commercial facility is ~55 million doses, with AHP score of 0.22. The capacity absorbs 84.83% of demand, with 91.08% utilization rate and $1.57 cost per unit. Its NPV is ~$45 million, and IRR of 18.29%, with 5.72 years of payback period.

Recommendations

Based on the conclusion, the recommendations drafted are:

1. That PT. XYZ can choose 55 million doses as the commercial facility’s capacity, as it is the best-fit alternative conducted by AHP.

2. Develop a robust demand model that will allow forecasting in volatile markets.

3. Conduct stakeholder meetings to align perspectives of stakeholder involved to ensure the project’s compliance with regulatory standards.

Conflict of Interest

Conflict of Interest: The authors declare that they do not have any conflict of interest.