Article

October 08, 2026

Evaluation of the SARIMA model in forecasting input prices in a collective catering company

Evaluation of the Sarima Model in Forecasting Input Prices in a Collective Meal Company

Leandro Pereira de Lacerda; Patrícia Belfiore Fávero

DOI: 10.22167/2675-6528-202603137

Article derived from a Final Course Work (TCC), with content based on the student’s original work and adapted to the editorial format of the E&S Magazine with the support of the ResumeAI tool, an artificial intelligence solution developed by Instituto Pecege for textual synthesis and organization.

Summary

The price forecasting of food inputs in collective meal companies represents a complex challenge. The SARIMA model was evaluated for forecasting the prices of 23 food inputs in a company in the collective meal sector, and the correlation between the internal price base and public market sources was analyzed. The research adopted a quantitative approach, using monthly historical series from January 2022 to March 2025. Data standardization and consolidation were performed, followed by Pearson and Spearman correlation analysis. In the predictive stage, the Box and Jenkins methodology was applied, which included stationarity testing (ADF), identification via ACF and PACF, maximum likelihood estimation, and model selection based on the AIC criterion. Out-of-sample evaluation compared SARIMA with Naive and Moving Average models, using the Mean Absolute Percentage Error (MAPE) as a metric. The results indicated low adherence between internal prices and public indicators, suggesting the influence of contractual and operational factors on the company’s price formation. The predictive evaluation showed heterogeneous performance of SARIMA, with higher accuracy for stable items and high errors for volatile products. The SARIMA model demonstrated practical utility in some cases but did not exhibit overall superiority, with simpler models showing competitive or superior performance in the analyzed sample. The findings reinforce the complexity of food input price forecasting, which is not fully captured by traditional models, and offer important subsidies for purchasing management in the sector.

Keywords: Acquisition Cost; Data Science; Purchasing Strategy; Time Series; Decision Making.

1. Introduction

The collective meal companies segment plays a fundamental role in meeting the dietary needs of a large number of people, ranging from industries to hospitals, universities, and schools (Neves et al., 2000). In 2024, according to the Brazilian Association of Collective Meal Companies (ABERC, 2025), the sector generated R$ 33.89 billion reais in Brazil, serving approximately 39.32 million meals daily and employing about 250,000 professionals. Considering its size and role in food services, it is a sector that directly impacts the health, productivity, and well-being of workers, students, and patients.

Companies operating in this segment face complex challenges, ranging from fundamental concern with food safety (Cavalli and Salay, 2007), solid waste management (Pires et al., 2022), food waste (Miranda et al., 2022), need for qualified labor (Oliveira and Molina, 2021), high competitiveness and emergence of new business models, such as dark kitchens (Monteiro et al., 2021), and efficient cost management (Stort and Lisboa, 2022).

This work sought to analyze one of these main challenges faced by companies in this segment, efficient cost management. This subject is highly relevant, as this market is also characterized by historically narrow profit margins, with many companies operating with profit margins of less than five percent (Coldibeli, 2023). The constant pressure on profitability stems from various factors, the main ones being high competitiveness and contractual conditions that pressure price maintenance even in adverse economic scenarios (ABERC, 2025).

This is the scenario currently faced by the company under study, marked by price volatility and high inflation of food inputs. In 2024, Brazil faced a significant increase in food prices. According to the Foundation Institute of Economic Research (FIPE, 2025), the Consumer Price Index (IPC) in the food category closed the year with an evolution of 8.03% accumulated over twelve months, almost double the General IPC. The Restaurant Consumption Index (ICR) closed 2024 with an evolution of 9.2% in the average cost per transaction. The increase in food costs is motivated by several factors, such as climate disasters (Oliveira and Ribeiro, 2024), the appreciation of the dollar (Roca, 2024), and global geopolitical tensions (Nogueira, 2024).

Amidst this scenario, the decision-making process represents an especially delicate and challenging aspect for the executives of these companies, as decisions related to the acquisition of inputs, pricing, inventory management, and operational strategies often need to be made in environments of uncertainty and volatility. The quality of these decisions directly impacts not only profitability but also business continuity, further increasing the responsibility and pressure on managers. Thus, having precise and timely information, obtained through advanced analytical methods, becomes essential to support executives in assertive and well-founded strategic decisions based on data (Rautenberg and Carmo, 2019).

To meet this need, quantitative methods, including advanced techniques such as Machine Learning, have demonstrated effectiveness in analyzing and forecasting prices of commodities and food inputs, enabling more assertive and proactive decision-making (Santos, 2022; Carvalho et al., 2021). Recent studies highlight the relevance of approaches such as ARIMA (autoregressive integrated moving average) models and artificial neural networks for commodity price forecasting, offering promising results in reducing uncertainties and improving decision-making (Santos, 2022; Ferreira et al., 2011; Lima et al., 2010; Spancerski e Santos, 2021). Given the complex reality of companies in this segment, the use of Data Science and Machine Learning techniques can provide a significant competitive advantage, allowing not only for more accurate prediction of future costs but also for providing subsidies for identifying cost reduction opportunities.

The present study had as its main objective to develop and evaluate SARIMA (seasonal autoregressive integrated moving average) models for forecasting the costs of 23 food inputs, with monthly series from January 2022 to March 2025, aiming to improve strategic purchasing decisions in collective meal companies. Additionally, Moving Average and Naive models were used for comparative purposes, allowing for the evaluation of the relative efficiency of the SARIMA model. The research also sought to evaluate the correlation of the company’s internal base purchase price behavior with external price bases, assessing the applicability of these bases as support information for managerial decision-making.

2. Material and Methods

The research adopted a quantitative approach, structured from a case study, a methodology that enables in-depth investigation of phenomena in their real organizational context (Yin, 2010; Creswell, 2010). The study was conducted in a company in the collective catering segment located in the metropolitan region of Curitiba/PR. This design allowed for the development and evaluation of SARIMA models for forecasting input costs, as well as the analysis of the correlation between internal and external prices, according to the study’s objectives.

Data collection was carried out through a documentary survey, based on the company’s internal purchase records. Complementarily, data from recognized economic research institutions were used, such as CEPEA/USP (Center for Advanced Studies in Applied Economics at the University of São Paulo), CONAB (National Supply Company), and SEAB (Paraná State Secretariat of Agriculture and Supply), referring to historical series of agricultural commodity prices.

The monthly time series covered the period from January 2022 to March 2025. The definition of this time frame was influenced by the facilitated availability of the company’s purchase data starting from January 2022, due to previous systemic limitations. Public databases were also collected for the same period, ensuring temporal data consistency.

Due to the great distinction of data sources, both in terms of organizational structure and patterns, a systematic process of preparation and treatment of the databases was carried out, using concepts of “data wrangling”. This process involved transforming the raw data from each database into a standardized and clean structure for the subsequent application of statistical models.

Initially, 38 Microsoft Excel spreadsheets were collected, which were transformed into four databases, one for each source (Company, CEPEA, CONAB, and SEAB). The organization and standardization of the data resulted in a structure that allowed for the analysis of prices by product, type of presentation, year, and month.

The standardization and cleaning of data were crucial steps. Prices diverged between types of presentation (Kilogram, Tons, Arrobas, Liters, Dozen, or units) for the same products. Product nomenclature also varied, especially at the company’s base, where items were registered by brand or supplier.

The dates, which varied between days, years, weeks, and months, were standardized to a monthly frequency, with year and month variables. Next, product descriptions were harmonized to a unique nomenclature, removing accents and special characters, which enabled switching between bases. The presentation type was normalized to Kilogram, Liter, and Dozen (specifically for chicken eggs), allowing direct comparisons between prices.

In parallel with standardization, a consistency and cleaning analysis was performed to identify products with direct correspondence between the databases and relevance to the company. Items such as office supplies, parts, equipment, and highly industrialized products, with no correspondence to public databases, were excluded.

From the company’s purchase volume data, 23 categories of food inputs were identified that showed direct correspondence between the bases and were the most representative in terms of volume, totaling 52.3% of the food products purchased. These 23 items were selected for the correlation analysis and model application stages.

For the data analysis, time series-based models were employed, which allow identifying patterns, trends, and seasonal structures in chronologically ordered data (Nielsen, 2021). The main model adopted was SARIMA (seasonal autoregressive integrated moving average), proposed by Box and Jenkins, due to its ability to capture seasonal patterns and trends.

The SARIMA represents an extension of the ARIMA model, incorporating seasonal components to describe periodic variations observed in monthly time series. The notation SARIMA (p,d,q)(P,D,Q)s was used, where p, d, q refer to the non-seasonal components and P, D, Q, s to the seasonal ones, with s=12 for monthly periodicity.

The application of the SARIMA model followed the Box and Jenkins (2015) methodology, structured in five stages. The first stage consisted of the descriptive analysis of the series, investigating trends, seasonality, and variability to support subsequent analyses.

The second stage was the stationarity check, a fundamental condition for ARIMA/SARIMA models. The Augmented Dickey-Fuller (ADF) test (Said and Dickey, 1984) was applied to the series at level. For non-stationary series, non-seasonal (d) and seasonal (D) differencing were applied until stationarity was achieved, with a significance level of 5% (p-value < 0.05).

The third stage involved identifying the model through the analysis of the correlograms of the autocorrelation function (ACF) and partial autocorrelation function (PACF). This analysis guided the specification of the orders of the autoregressive (p, P) and moving-average (q, Q) components of the SARIMA model (Box et al., 2015).

The fourth stage was parameter estimation, using the maximum likelihood method. The final model selection was based on Akaike’s Information Criterion (AIC), which seeks a balance between goodness of fit and parsimony, penalizing models with many parameters.

The fifth stage consisted of the analysis of the residuals, evaluated through the correlogram of the autocorrelation function (ACF) and the Ljung-Box test. The objective was to verify the absence of remaining autocorrelation, indicating that the residuals behaved as white noise, with p-values greater than 0.05.

The available historical series comprised 39 monthly observations, of which 33 were used in the estimation stage and six were reserved for out-of-sample validation. The training set, containing less than three full annual seasonal cycles (s = 12), limited the amount of information available to estimate seasonal structures and models with a larger number of parameters.

Additionally, for comparative purposes, two other forecasting models were used: the Naive model and the Moving Average model. The Naive model assumes that the predicted value for the next period is equal to the last observed value, serving as a basic reference in time series forecasting. The Moving Average model estimates future values from the average of the last observed periods, smoothing short-term variations and capturing recent trends. These simpler models were included to evaluate their competitive performance in relation to SARIMA.

The error measure adopted for evaluating the predictive performance of the models was the Mean Absolute Percentage Error (MAPE). MAPE expresses the average error in percentage terms, facilitating the comparison of accuracy between products with different price levels. MAPE < 10% was adopted as an indicator of high predictive accuracy.

The analyses were implemented in Python language, using the pandas, matplotlib, statsmodels, and scikit-learn libraries.

3. Results and Discussion

This section details the results obtained in the study, interpreting them in light of the proposed objectives and the applied methodology. Both the findings regarding the comparison between the company’s internal price base and external market databases, and the performance of the SARIMA model and its comparison with the Naive and Moving Average models in forecasting food input prices were analyzed. The discussion aims to provide an in-depth understanding of pricing dynamics and the predictive capabilities of the models, offering subsidies for strategic decision-making in the collective catering sector, according to the research problem and the objectives established in the introduction.

Internal base adherence compared to external price bases

In order to evaluate the adherence of the purchase price behavior practiced by the company in relation to public market indicators, a correlation analysis was performed between the internal historical price series and the CONAB, SEAB, and CEPEA databases. This procedure is of great relevance, as, in the company’s routine, consulting these external sources is a common procedure in the planning and negotiation processes of purchases, serving as a reference for defining commercial and input acquisition strategies.

Correlation analysis, using Pearson and Spearman coefficients, revealed considerable variation in the relationship between the company’s internal prices and public indicators for the 23 products evaluated. It was observed that most items showed low adherence to external databases, with correlation coefficients frequently close to zero or negative. For example, products such as Iceberg Lettuce, English Potato, and Arabica Coffee exhibited very low correlations, indicating little or no similarity between their internal pricing dynamics and those of the general market.

In contrast, some products showed higher correlations with external bases, suggesting greater sensitivity to market fluctuations. Pork, Pêra Orange, Watermelon, Melon, and Wheat, for example, presented more substantial correlation coefficients, indicating that their domestic prices may be more influenced by market trends. However, even for these items, adherence was not complete, which reinforces the complexity of price formation in the specific environment of the studied company.

The low adherence observed for most products suggests that the company’s internal price formation is significantly influenced by organization-specific factors, which transcend widely publicized market dynamics. In discussions with the company, several factors were identified, such as local supply and demand conditions, regional supply characteristics, the company’s bargaining power with its suppliers, and specific contractual agreements. A notable example is the contractual requirement to acquire part of the inputs from family farmers, which naturally distances internal price behavior from market averages.

These findings are crucial for purchasing management, as they indicate that external price bases, although useful as market intelligence and benchmarking tools, should not be considered a single and deterministic parameter for internal pricing. The company needs to develop a deep understanding of its own cost and negotiation drivers, recognizing that its operational and contractual reality uniquely shapes its acquisition prices, as pointed out by Rautenberg and Carmo (2019) regarding the importance of data for strategic decisions.

The correlation analysis also had a direct impact on the methodological decision for the application of the SARIMA model. Given the predominance of products with their own pricing dynamics and low adherence to external bases, the decision was made to exclusively use the company’s internal historical series for predictive modeling. This choice aimed to ensure that the estimates and forecasts reflected, more faithfully, the real negotiation, operation, and acquisition conditions of the organization under study, guaranteeing the practical relevance of the developed models.

Application of the SARIMA Model

The application of the SARIMA model strictly followed the five steps of the Box and Jenkins (2015) methodology, an approach widely recognized in time series literature. These steps included descriptive analysis of the series, stationarity verification, model structure identification through correlograms, parameter estimation, and finally, residual analysis for validation. Each phase was fundamental to building robust models suitable for the specific characteristics of each food input price series.

Descriptive Analysis of the Series

The descriptive analysis of the company’s internal monthly price series, covering the 23 selected products, provided an initial overview of the volatility and patterns of each input. Measures of central tendency, such as mean and median, and dispersion, including minimum, maximum, standard deviation, and coefficient of variation, were calculated to characterize the distribution of prices over the period from January 2022 to March 2025. These indicators are essential for contextualizing the SARIMA modeling, as series with greater dispersion generally exhibit higher predictive uncertainty.

The results of the descriptive statistics revealed that the vast majority of product categories presented medium and high volatility, a common behavior in agricultural markets. Of the 23 items analyzed, only three showed low variability, with a coefficient of variation less than or equal to 0.10: sugar, chicken meat, and tilapia. Ten other categories exhibited medium variability, with coefficients between 0.10 and 0.20, while the remaining ten categories presented high variability, with coefficients greater than 0.20.

This heterogeneity in input price volatility already indicated a significant challenge for the application of adherent forecasting models, suggesting that a standardized approach for all products might not be effective. The observation of the original time series for a sample of products, such as sugar, wheat, tilapia, and whole long-life milk, showed relevant differences in amplitude and variability over the analyzed period. These distinctions reinforced the need to estimate SARIMA models individually for each product, considering that the price formation dynamics and the degree of seasonality have distinct behaviors among categories.

Stationarity Test

The verification of the stationarity of time series is a fundamental condition for the application of ARIMA/SARIMA models, as it assumes that the mean and variance of the series are approximately constant over time. For this purpose, the Augmented Dickey-Fuller (ADF) test, proposed by Said and Dickey (1984), was applied to the series in level. The decision criterion adopted was a significance level of 5% (p-value < 0.05), where p-values lower than 0.05 indicate the rejection of the null hypothesis of a unit root, meaning the series is stationary.

The initial ADF test revealed that six of the 23 products were already stationary at level, showing p-values below 0.05. These included Iceberg Lettuce, Curly Lettuce, Cavendish Banana, Chicken Meat, Watermelon, and Tomato. For the remaining 17 series, transformations were necessary to achieve stationarity. Thirteen products required non-seasonal differencing (d = 1), three products needed two non-seasonal differences (d = 2), and Potato required one seasonal difference (D = 1).

After applying the appropriate differencing, all series presented p-values below 0.05 in the ADF test, indicating that the null hypothesis of a unit root was rejected and stationarity was achieved. This step was crucial to ensure the validity and robustness of the subsequent SARIMA models, allowing the models to capture temporal patterns without the influence of non-stationary trends or seasonalities, which could compromise the accuracy of the forecasts.

Autocorrelation Functions Analysis

After the series stabilization stage, the correlograms of the autocorrelation (ACF) and partial autocorrelation (PACF) functions were analyzed. The ACF measures the correlation between the values of a time series and its own lagged values, allowing the identification of temporal dependence and recurring patterns. The PACF, in turn, measures the correlation between the series and a given lag, removing the effect of intermediate lags, which is fundamental for isolating the direct influence of each lag.

The joint analysis of the ACF and PACF correlograms is an essential tool in the Box and Jenkins (2015) methodology to guide the identification of the orders of the autoregressive (p, P) and moving average (q, Q) components of the SARIMA model. An abrupt cut-off in the PACF generally indicates the AR order, while a cut-off in the ACF points to the MA order. Although the final model selection was guided by the Akaike Information Criterion (AIC), the inspection of the correlograms provided statistical support to justify the presence or absence of short-term temporal dependence and annual recurring patterns in each price series.

Model Estimation

The estimation of the SARIMA models was performed using the maximum likelihood method, implemented through the SARIMAX function of the “statsmodels” package in the Python language. For each of the 23 products, several combinations of model parameters were tested, seeking to identify the structure that best represented the behavior of the price series. The choice of the final model for each product was based on the Akaike Information Criterion (AIC), which is widely used in the selection of statistical models for promoting a balance between goodness of fit and parsimony.

AIC penalizes models that use an excessive number of parameters, favoring those that can explain the data well with greater simplicity. Thus, between two models with similar fit, the AIC criterion prefers the simpler one, with models with a lower AIC value being considered preferable. In this study, AIC was used to select, among the different combinations of non-seasonal (p,d,q) and seasonal (P,D,Q,s) parameters tested, the most adequate specification for each product, ensuring the efficiency and interpretability of the fitted models.

Evaluation of the Adjusted Model

The quality of the adjusted SARIMA models was evaluated through two complementary approaches: residual diagnostics and out-of-sample forecast performance. Residual diagnostics are crucial to verify if the model has captured the entire temporal dependence structure of the series, leaving only random “white noise”. For this purpose, the Ljung-Box test was used, which assesses the presence of joint autocorrelation in the residuals up to a certain lag.

The Ljung-Box test indicated that 19 out of 23 models showed no evidence of residual autocorrelation, with p-values equal to or greater than 0.05. This suggests that, for most products, the selected SARIMA models were adequate for capturing the temporal dependence of the price series. However, in four specific cases — Iceberg Lettuce, Curly Lettuce, Potato, and Pork — the p-values were below 0.05, indicating the presence of remaining autocorrelation in the residuals. This result suggests that the selected specification for these products did not fully capture the temporal dependence, which may impact the accuracy of their forecasts.

To evaluate the predictive capacity of the models, a temporal out-of-sample validation was performed. The last six months of the historical series for each product were reserved as the test period, and the predictions generated by the SARIMA models were compared with the actual values observed during this period. This approach allows for a more realistic evaluation of the model’s performance in scenarios not observed during the estimation phase, providing a robust measure of its practical utility for decision-making.

Additionally, projections were generated for the six months following the validation period, with 95% confidence intervals, for each product. These future projections are of great value to the company, as they offer an estimate of expected input prices, aiding in purchasing planning and strategy formulation. Visualizing these projections, along with their confidence intervals, allows managers to assess the level of uncertainty associated with each forecast, facilitating risk management.

The error metric used to compare the predictive performance of the models was the Mean Absolute Percentage Error (MAPE), which expresses the average error in percentage terms, facilitating the comparison of accuracy between products with different price levels. For the interpretation of the results, it was adopted that a MAPE below 10% would indicate high predictive accuracy. MAPE values were calculated for the SARIMA, Naive, and Moving Average models in the out-of-sample validation period, allowing for a direct comparison of the relative efficiency of each approach.

The MAPE results revealed heterogeneous performance of the SARIMA model among different products. Only four products showed high predictive accuracy (MAPE < 10%) with SARIMA, three of them being industrialized items: Wheat (3.28%), Sugar (4.39%), and Whole Long-Life Milk (7.17%). In contrast, products such as English Potato (151.89%), Yellow Onion (151.84%), and Salad Tomato (119.67%) showed MAPEs above 100%, indicating a low predictive capacity of SARIMA for these highly volatile items.

In comparison with simpler methods, the Naive model presented high accuracy (MAPE < 10%) for six products, while the Moving Average model achieved this level of accuracy for seven products, outperforming SARIMA, which reached this accuracy for only four. To mitigate the influence of extreme values and obtain an aggregated view, the comparison was performed by the median of MAPE. The median MAPE for SARIMA was 23.69%, while for Naive it was 16.46% and for Moving Average it was 15.39%.

These results indicate that, in the specific sample analyzed, simpler models, such as Naive and Moving Average, demonstrated greater robustness for a relevant portion of the products. However, this finding should not be interpreted as an intrinsic inferiority of SARIMA, but rather as an indication that its performance is sensitive to the characteristics of the series and the available sample design. The short length of the historical series (33 observations for estimation) and the limited number of complete seasonal cycles (less than three annual cycles for a 12-month seasonality) restricted the statistical support for estimating more complex seasonal structures and parameterizations with stability.

In this context, more parsimonious models, such as Naive and Moving Average, can present greater robustness, especially when the series combines a limited historical extension with high volatility. The high errors observed in the most volatile series and in some more complex specifications reinforce the need to interpret the results by simultaneously considering price volatility, model parsimony, and the amount of available observations. The prediction of food input prices, in the analyzed context, is a problem of high complexity, which is not fully captured by traditional models, whether simple or more sophisticated, highlighting the importance of comparative evaluation between different approaches and the need for methodological evolution.

In summary, the research results highlight the complexity of forecasting food input prices in the collective catering sector. The low correlation between the company’s internal prices and market indicators underscores the influence of specific organizational factors, while the heterogeneous performance of the SARIMA model, with simpler models showing competitiveness in many cases, emphasizes the importance of the time series length and product volatility. These findings provide valuable insights for purchasing management, indicating the need for differentiated approaches and continuous improvement of predictive strategies.

4. Conclusion

This study aimed to develop and evaluate SARIMA models for forecasting the prices of 23 food inputs in a collective catering company, as well as to analyze the correlation between internal prices and public market indicators, in order to improve strategic purchasing decisions. It was found that the formation of the company’s internal prices showed low adherence to most public indicators, suggesting the predominant influence of contractual and operational factors specific to the organization. This finding directed the predictive modeling towards the exclusive use of the internal historical series. In the application of the SARIMA model, heterogeneous performance was observed among products, with greater precision for items of lower volatility, such as sugar, wheat, and whole long-life milk. However, for highly volatile products, the predictive capacity of SARIMA was limited, with high percentage errors. Comparison with simpler models, such as Naive and Moving Average, indicated that these, in many cases, showed competitive or superior performance, especially due to the short length of the historical series available for estimation. The findings reinforce the inherent complexity of forecasting food input prices in the sector, offering important subsidies for purchasing management, which should consider the individuality of each input and the influence of internal factors on price formation.

The main methodological limitation of the study lay in the sample horizon of 39 months, with only 33 observations for estimation. This restricted scope, combined with the need to capture annual seasonalities, limited the ability to stably estimate seasonal structures and more complex parameterizations, impacting the SARIMA model’s performance compared to more parsimonious approaches. For future studies, it is suggested to expand the historical data base, incorporate exogenous variables such as climatic and macroeconomic factors, and explore more advanced or hybrid predictive models. Such initiatives can improve the estimation of seasonal patterns and predictive capacity in contexts of greater complexity, contributing to the continuous development of more effective cost management strategies in the collective meal sector.

Bibliographic References

Associação Brasileira das Empresas de Refeições Coletivas [ABERC]. 2025. Alimentação do trabalhador em risco: o impacto da inflação no setor de Refeições Coletivas. Disponível em: https://www.aberc.com.br/alimentacao-trabalhador-inflacao-alimentos. Acesso em: 15 mar. 2025.

Carvalho, R.; Biston, J.V.; Favan, R. Deolindo, D. 2021. Avaliação de Algoritmos de Machine Learning na Cotação do Preço do Contrato futuro de Milho. Revista E-Fatec 11 (1), 30-46.

Cavalli, S.B.; Salay, E. 2007. Gestão de pessoas em unidades produtoras de refeições comerciais e a segurança alimentar. Revista de Nutrição 20 (6), 657-667.

Coldibeli, L. 2023. Empresas lucram com venda de marmitas e administração de refeitórios. Disponível em: https://economia.uol.com.br/noticias/redacao/2013/07/16/empresas-lucram-com-venda-de-marmitas-e-administracao-de-refeitorios.htm. Acesso em: 15 mar. 2025.

Ferreira, L.; Moura, G.L. de; Borenstein, D.; Fischmann, A.A. 2011. Utilização de redes neurais artificiais como estratégia de previsão de preços no contexto de agronegócio. Revista de Administração e Inovação 8 (4), 6-26.

Fundação Instituto de Pesquisas Econômicas [FIPE]. 2025. IPC – Índice de Preços ao Consumidor. Disponível em: https://www.fipe.org.br/pt-br/indices/ipc. Acesso em: 10 mar. 2025.

Lima, F.G.; Kimura, H.; Neto, A.A; Perera, L.C.J. 2010. Previsão de preços de commodities com modelos ARIMA-GARCH e redes neurais com ondaletas: velhas tecnologias – novos resultados. Revista de Administração (RAUSP) 45 (2), 188-202.

Miranda, M.B.; Santos, J.M.; Bezerra, G.B.; Oliveira, J.M. 2022. Efetividade de uma campanha para redução de desperdício em uma unidade de alimentação e nutrição. Cadernos Saúde Coletiva 30 (3), 424-430.

Monteiro, M.A.M.; Henriques, G.S.; Garcia, M.A.V.T. 2021. Organização e Gestão do Setor de Alimentação Coletiva no Pós-Covid-19: Desafios e Perspectivas. Brazilian Journal of Development, Curitiba 7 (10), 97134-97158.

Neves, M.F.; Chaddad, F.R.; Lazzarini, S.G. 2000. Alimentos: novos tempos e conceitos na gestão de negócios. 1ed. Pioneira, São Paulo, SP, Brasil.

Nogueira, I.L. 2024. Escassez à Mesa: correlação do risco geopolítico e as mudanças climáticas na inflação dos alimentos. Trabalho de Conclusão de Curso em Administração. Universidade Federal do Amazonas (UFAM), Manaus, AM, 2024.

Oliveira, B.V de; Molina, V.B.C. 2021. Gestão de pessoas em unidade de alimentação e nutrição. Revista Multidisciplinar da Saúde (RMS) 3 (2), 24-36.

Oliveira, F. A. P. de; Ribeiro, M. E. C. 2024. Logística no agronegócio e desafios climáticos: impactos das enchentes no Rio Grande do Sul. Trabalho de Conclusão de Curso em Tecnologia em Agronegócios. Faculdade de Tecnologia de Presidente Prudente, Presidente Prudente, SP, 2024.

Pires, V.C. da; Seabra, L. M. J.; Rolim, P. M. 2022. A relação entre alimentação coletiva e gestão de resíduos: o papel da Educação Ambiental na redução do desperdício. Revista Brasileira de Educação Ambiental 17 (5), 341–360.

Rautenberg, S.; Carmo, P.R.V. do. 2019. Big Data e Ciência de Dados: complementariedade conceitual no processo de tomada de decisão. Brazilian Journal of Information Studies: Research Trends 13 (1), 56-67.

Roca, G. 2024. Dólar sobe 27% ante o real em 2024, maior valorização desde 2020. Disponível em: https://valor.globo.com/financas/noticia/2024/12/30/dolar-sobe-27percent-ante-o-real-em-2024-maior-valorizacao-desde-2020.ghtml. Acesso em: 10 mar. 2025.

Santos, J.A.A. dos S. 2022. Previsão do preço do café arábica: uma aplicação de redes neurais CNN-BLSTM. Research, Society and Development 11 (3), e3511326101.

Spancerski, J.S; Santos, J.A.A

Stort, L.G.; Lisboa, F.C. 2022. Quality tools application to improve cost management in food and nutrition unit: A case study. Research, Society and Development 11 (13), e128111335254.

Article originating from the Final Course Work of the Specialization in Data Science and Analytics of the MBA USP/Esalq

To learn more about the course, click here and access the MBX Academy platform

You may also like

November 17, 2026

Detection and Diagnosis of Diseases in Corn and Rice in Mozambique with Deep Learning

Mozambican agriculture depends on corn and rice, fundamental crops for food security, but their productivity is strongly affected by phytosanitary diseases, whose early and accurate identification remains a challenge in resource-limited contexts. The study aimed to develop and compare Deep Learning-based models for the automatic detection and diagnosis of diseases in corn and rice leaves in Mozambique. Three approaches were used: a classic Convolutional Neural Network (CNN) developed from scratch, and two transfer learning models, ResNet50 and MobileNetV2. The dataset consisted of images from public databases and collected in the field, which underwent cleaning, balancing, and artificial augmentation steps. The models were trained and evaluated with standard performance metrics, including accuracy, precision, recall, F1-score, Log Loss, and Cohen’s Kappa. The classic CNN presented the best overall performance, with 94% precision, recall, and F1-score, and 93% accuracy and Cohen’s Kappa. MobileNetV2 achieved 91% F1-score and 90% accuracy, standing out for its lower Log Loss (0.30) and training time (6h32min59s). ResNet50 reached 80% F1-score and 79% accuracy. It was concluded that Deep Learning techniques have high potential to support phytopathological diagnosis in real agricultural contexts, contributing to faster and more accurate decisions in the management of corn and rice diseases.

Keywords: Deep learning; Rice; Plant diseases; Corn; Computer vision.

November 17, 2026

Artificial intelligence in accounting operational routines and financial efficiency

The incorporation of Artificial Intelligence (AI) and automation tools has reconfigured the routines of Brazilian accounting offices, impacting productivity, error mitigation, and financial efficiency. These impacts were analyzed in a small accounting firm, specialized in serving service providers in the medical field in São Caetano do Sul, São Paulo. The research adopted a qualitative approach, with a single case study design and incorporated units of analysis. Data were collected through a structured questionnaire, applied via Google Forms between January and February 2026, to five professionals in the operational structure of the firm in a census survey. The treatment followed content analysis with a priori and emergent categories. All respondents declared the use of automated tools or tools with AI resources. The perception of reduced execution time, decreased operational errors, rework, and better use of the team’s time converged in 80% of affirmative responses and 20% partial responses. The most benefited activities were bank reconciliation, import of fiscal documents, and tax assessment, with visible gains on typing errors, duplicate entries, and incorrect classification. Human intervention remained indispensable in tax planning, scenario analysis, and verification of ancillary obligations. The overall evaluation of AI’s contribution was predominantly favorable, with moderation linked to the complexity of fiscal regulations.

Keywords: Content analysis; Process automation; Accounting 4.0; Small business; Digital transformation.

November 17, 2026

Factors influencing schedule management in sewage treatment plant “retrofit” projects

Projects involving “retrofit” in Sewage Treatment Plants (STP) are characterized by high complexity, operational restrictions, and multiple technical interfaces. The study analyzed the perception of the professionals involved regarding the management and adherence to deadlines in “retrofit” projects for sewage treatment plants, considering technical, operational, and organizational factors. The research was characterized as an applied study, with a quali-quantitative approach, designed as a field survey. Data analysis was performed using descriptive statistics and qualitative categorization, employing SWOT, risk matrix (Probability x Impact), and 5W2H action plan tools. The results indicated a high perception of clarity (77.8%) and realism of deadlines (100% agreement), evidencing consistent initial planning. However, 77.7% of respondents pointed out that rework and scope changes directly impacted deadline adherence. The integration between engineering, manufacturing, and Production Planning and Control (PPC) showed an intermediate evaluation (77.7% partial/total agreement), indicating fragility in interfunctional coordination. The risk matrix highlighted as critical: scope changes, operational limitations of the STP, and technical complexity, all with a high level. As a central improvement, the standardization of scope change control, analysis, and approval was proposed, aiming to reduce rework and increase schedule predictability.

Keywords: Performance; Integration; Planning; Rework; Risk.

November 17, 2026

B2C Acquisition in diagnostic medicine: strategic partnerships with specialized medical offices

The diagnostic medicine sector in Brazil operates with high dependence on health plan providers, which limits the financial autonomy of service providers. This study proposed and analyzed a direct patient acquisition model (B2C) for a highly complex diagnostic medicine institution, structured through partnerships with Oncology and Cardiology practices and operationalized by a personalized customer service called concierge. The investigation adopted a mixed-methods approach in three phases: a field survey with a structured questionnaire applied to 87 patients, mapping of the operational flow using “Service Blueprint”, and analysis of the conversion rate from 312 medical referrals registered in the pilot project. The results indicated that trust in medical referral achieved 92.0% agreement, establishing itself as the main factor for adherence. The convenience of performing the exam on the same day reached 78.2%, and the willingness to pay directly was 55.2%, with polarization associated with the predominance of health plan users. The “Service Blueprint” identified three critical points in the transitions between stages of the patient journey. The overall conversion rate was 51.0%, with superior performance in Oncology (58.2%) compared to Cardiology (45.0%), and the greatest loss was concentrated in the acceptance and scheduling stage (28.3%). It was concluded that the model demonstrated possible operational viability and indications of patient acceptance, conditioned by price transparency and a reduction in the interval between consultation and approach by the concierge.

Keywords: Direct payment; Patient journey; Supplementary health; Concierge service; Conversion rate.

November 17, 2026

The role of the school manager in promoting an inclusive culture: leadership and educational policies

The research analyzed the school manager’s performance in promoting an inclusive culture, considering the need to transform legal guidelines and pedagogical foundations into effective institutional practices. It was assumed that school inclusion requires planning, organization, and strategic leadership in the school’s daily operations. A qualitative approach of an applied nature was adopted, developed in three complementary stages: documentary analysis of Brazilian legislation on inclusive education, bibliographic review on school leadership and management, and data systematization using management tools such as SWOT analysis and the PDCA cycle. The results indicated that the school management’s performance directly influenced the consolidation of inclusive environments, especially in the organization of processes, continuous team training, and strengthening of a collaborative institutional culture respectful of diversity. Structural, pedagogical, and relational challenges that demanded planned interventions were also identified. Finally, a Practical Guide for School Managers was developed, with guidelines, protocols, and operational orientations for decision-making and the organization of inclusive processes. It was concluded that the construction of an inclusive culture depended on the management’s ability to articulate public policies, pedagogical practices, and institutional processes, consolidating inclusion as a permanent commitment in the school context.

Keywords: educational accessibility; educational equity; participatory management; strategic planning; institutional processes.

November 17, 2026

An AI artifact based on Design Thinking to support the elicitation of software requirements from user needs

The development of software oriented to user needs frequently faces challenges in requirements elicitation, stemming from ambiguities in natural language and communication failures. Given this, an intelligent artifact integrating Design Thinking and Artificial Intelligence was proposed and validated, focusing on the Empathy and Definition stages, to support the collection, interpretation, and structuring of user needs. The research, of an applied, qualitative, and exploratory nature, conducted a Proof of Concept in a controlled environment, employing hypothetical scenarios. The artifact operated through guided natural language interviews, organizing information into elements such as pain points, context, and impact, and generating traceable problem syntheses. The results indicated that the system organized unstructured reports, reduced ambiguities, and preserved the meaning of the original information in most cases. The ability to reorganize logically in complex scenarios was also observed, although with limitations in separating multiple problems and dependence on instructions. It was concluded that the artifact has potential as a support for requirements elicitation, improving the clarity and structuring of information, but requires supervision and contextual control.

Keywords: Design Thinking; Requirements Elicitation; Requirements Engineering; Artificial Intelligence; Natural Language Processing.

November 17, 2026

The Extinction of the Monophasic Tax System in the Pharmaceutical Industry in Light of the Tax Reform

The consumption Tax Reform, established by Constitutional Amendment No. 132/2023, introduced structural changes to the Brazilian tax system, with the replacement of the Social Integration Program (PIS) and the Social Security Financing Contribution (COFINS) by the Contribution on Goods and Services (CBS), in a dual Value Added Tax (VAT) model. The objective was to analyze the impacts of the extinction of the single-stage system on the pharmaceutical industry, in light of the new regulatory design and its economic and legal effects. A qualitative, exploratory research was conducted through a case study and documentary analysis of legislation and specialized doctrine. It was examined whether the end of the single-stage system would imply a relevant increase in the tax burden for the sector. The results indicated that, although the formal overcoming of the single-stage system has occurred, the material exemption of medicines has been preserved, mainly due to the application of reduced rates of 60% and, in certain cases, a zero rate of the Tax on Goods and Services (IBS) and CBS. The change in model may cause occasional variations in revenue collection, without characterizing a structural increase in the tax burden, tending towards neutrality over time. It was preliminarily concluded that the Tax Reform reorganized the taxation technique in the pharmaceutical industry, replacing a concentrated regime with more transparent exemption mechanisms. This approach proved compatible with the principles of neutrality, full non-cumulativeness, and fiscality, maintaining tax protection for essential medicines and aligning fiscal policy with the fundamental right to health.

Keywords: CBS; COFINS; IBS; Dual VAT; PIS.