یک روش شبکه عصبی برای پیش بینی نتایج مذاکره برای قیمت در زمینه کسب و کار به کسب و کار

Price premiums are a key profit driver for long-term business relationships. For sellers in business-to-business (B2B) relationships, it is important to have appropriate strategies to negotiate price increases without trading off the relationships with their buyers. This paper aims to understand the annual price negotiation processes of companies by predicting whether a seller’s reservation price, target price, and initial offer positively affect the price negotiation outcome between the sellers and buyers. Data from 284 B2B relationships of a chemicals supplier based in Germany was used to examine our research model. In order to capture the non-linear decisions that are involved in price negotiations and to address collinearity among negotiations’ determinants, neural network analysis was used to predict the factors that influence price negotiation outcome. The neural network model was then compared with the results from regression analysis. Compared to regression analysis, the neural network has a lower standard error, and it showed that target price played a more important role in B2B price negotiations. The neural network was also able measure non-linear, non-compensatory decisions that are involved in price negotiations. The results imply that neural networks should be more widely used by researchers to address the threats that multi-collinearity poses. For companies, the results imply that price targets should be actively managed, e.g. through clear financial aims or through seminars aiming to help sales personnel to establish more challenging negotiation aims.

Price negotiations play an important role in business as their outcomes can impact long-term business relationships’ profitability and the reputation of businesses (Carbonneau, Kersten, & Vahidov, 2008). B2B price negotiations have various challenges such as the complex business environment which usually involves multiple interactions by at least two – and often many – people (Plank, 1997), and are therefore more complex than consumer price studies (Carbonneau et al., 2008, Holden and Burton, 2008 and Kotler and Keller, 2006). Companies understandably also very rarely make their B2B pricing transparent or accessible due to its direct competitive profit relevance and often strategic character. For sales personnel who are preparing for annual price negotiations, it is difficult to know what price to demand (initial offer), what settlement to actually expect (target price), and what minimum price can be accepted before the relationship becomes unprofitable (reservation price). Price references (Mazumdar, Raj, & Sinha, 2005) influence the negotiation behavior of both sellers and buyers and ultimately the price negotiation outcome. Several conclusions can be drawn from existing studies of pricing negotiations, e.g. by Moosmayer, Schuppar, and Siems (2012) and Van Poucke and Buelens (2002). Firstly, existing studies in price negotiations are overly experimental (Krause, Terpend, & Petersen, 2006), often using student samples, and transactional in nature. However, findings based on experimental designs account neither for context factors such as negotiators’ expertise and experience, nor for the fact that the nature of industrial business is predominantly relationship-based, rather than transactional. The validity of experimental, transaction-oriented findings for price negotiations in B2B relationships thus appears questionable. Moreover, studies based on student sampling, although strong in internal validity (Bachrach and Bendoly, 2011 and Eckerd and Bendoly, 2011), may suggest inappropriate business decisions due to limited external validity (Ketchen and Hult, 2011 and Stevens, 2011). Secondly, linear regression models are often chosen to examine the relationships between the determinants of price negotiations and their outcomes. However, regression models are preference regressions which assume that price negotiation decisions are linear compensatory models (Chong, 2013). Under this assumption, the shortfall in a negotiations decision such as reservation price can be compensated for by other factors such as initial offers or target price. However, given the complexities involved in price negotiations, linear regression models may not be able to capture all the non-compensatory decision rules involved in these processes, and as a result such models are deemed unreliable. Studies in other disciplines such as information systems have found that linear models tend to oversimplify the complexities involved in decisions (Chong, 2013 and Venkatesh and Goyal, 2010). Thirdly, a limitation of regression models is the assumption of independent determinants; however, the seller’s price preferences in negotiations are often interdependent. This may result in high multi-collinearity in the data analysis thus affecting the reliability of the results. This research has several objectives. Firstly, this study aims to understand the factors that can predict the price negotiation outcomes in B2B relationships. Variables such as a seller’s reservation price, target price, and initial offer are examined to see if they predict price negotiation outcome. Secondly, this research aims to examine whether non-linear, non-compensatory decision models such as neural networks provide a better model fit and forecasting than linear regression models for predicting pricing negotiation outcomes. In order to achieve this, the results from the neural network will be compared with regression analysis. Lastly, based on the results, this research will suggest how companies can maintain profitable long term B2B relationships by managing sales personnel’s trust in pricing negotiations, and how researchers can use neural networks to help address multi-collinearity issues.

This research examined the predictors of B2B price negotiation outcomes. Variables such as reservation price, target price, initial offering, profitability and size of relationship were examined in our proposed model. Using the predictive analytic approach provided by a neural network, this research examined B2B negotiations of a major chemical firm in Germany and its buyers. The results obtained from the neural network were then compared with multiple regression analysis. The neural network was able to solve the multi-collinearity bias as well as to provide better predictions than the regression model. This research has several implications. Firstly, for companies, the results imply that it would be wise to support their sales personnel in setting ambitious targets and translating these into a specific target price for each negotiation. Concretely, companies could communicate an average target price, or a price target for each customer. Furthermore, training could explicitly aim to enable negotiators to establish ambitious price targets and to defend them in a negotiation. Wilken et al. (2010) have found that providing transparent cost information to negotiators can lead to more favorable outcomes for the company. It might similarly be effective for a company to convey a better understanding to its negotiators of how potential price negotiation outcomes are connected to the firm’s financial performance. For the individual sales representative, results generally suggest that expecting a little more might also result in a little more. In particular, this requires solid preparation with a clear target to be achieved. Secondly, the results explained variation in price negotiation outcome better that existing experimental research. Thus, relying on experimental data that overemphasizes the importance of the initial offer appears to be misleading. Using market data from business negotiations seems crucial for avoiding poor managerial decisions based on overly biased findings. Thirdly, this research showed that neural networks can be used as a means to overcome multicollinearity issues in data. In price negotiation studies where the predictors could be highly correlated, neural network analysis allows us to predict the outcomes with less error. Our research thus suggests that neural network method may be applied to broader range of business research problems where multicollinearity may be an issue.