Abstract
We address the lead time quotation problem of a manufacturer serving multiple customer classes. Customers are sensitive to the quoted lead times and the manufacturer has the flexibility to keep inventory to improve responsiveness. We model the problem as a Markov decision process and characterize the optimal lead time quotation, rationing, and production policies. We then define internal and external service level measures and analyze the impact of inventory keeping decision on these measures. Before analyzing the impact, we first derive the relation between inventory level and lead time quotes. We then show that the effect of inventory level on the service levels may not follow the intuition. We also study alternative lead time quotation and production schemes. We contrast the performance of these alternative policies with that of the optimal policy. Specifically, through a numerical study, we quantify the value of controlled arrivals and customer rejection, the value of information on customer status, and the value of a farsighted policy. Through the numerical study, we also identify the effect of revenue mix and demand mix on the inventory keeping decisions and the performance measures. Finally, we measure the impact of lead time quotation on resource pooling (i.e., inventory and capacity pooling). We show that the value of resource pooling is limited under the optimal policy, since lead time quotation is already effective in balancing the demand with capacity, and in allocating the resources among different customer classes.
This is a preview of subscription content, access via your institution.
References
AktaranKalayci T, Ayhan H (2009) Sensitivity of optimal prices to system parameters in a steadystate service facility. Eur J Oper Res 193:120–128
Arslan H, Graves SC, Roemer TA (2007) A singleproduct inventory model for multiple demand classes. Manag Sci 53:1486–1500
Benjaafar S, ElHafsi M (2006) Production and inventory control of a single product assembletoorder system with multiple customer classes. Manag Sci 52:1896–1912
Bitran G, Caldentey R (2003) An overview of pricing models for revenue management. Manuf Service Oper Manag 5:203–229
Bulut O, Fadiloglu MM (2011) Production control and stock rationing for a maketostock system with parallel production channels. IIE Trans 43:432–450
Buzacott J, Shanthikumar J (1993) Stochastic models of manufacturing systems. Prentice Hall, Prentice Hall international series in industrial and systems engineering, Englewood Cliffs
Caldentey R, Wein LM (2006) Revenue management of a maketostock queue. Oper Res 54:859–875
Cil E, Ormeci E, Karaesmen F (2009) Effects of system parameters on the optimal policy structure in a class of queueing control problems. Queueing Syst 61:273–304
Defregger F, Kuhn H (2007) Revenue management for a maketoorder company with limited inventory capacity. OR Spectrum 29:137–156
Dekker R, Kleijn M, de Rooij P (1998) A spare parts stocking policy based on equipment criticality. Int J Prod Econ 5657:69–77
Duenyas I (1995) Single facility due date setting with multiple customer classes. Manag Sci 41:608–619
Duenyas I, Hopp WJ (1995) Quoting customer lead times. Manag Sci 41:43–57
Elmaghraby W, Keskinocak P (2003) Dynamic pricing in the presence of inventory considerations research overview, current practices, and future directions. Manag Sci 49:1287–1309
Gayon JP, TalayDegirmenci I, Karaesmen F, Ormeci EL (2009) Optimal pricing and production policies of a maketostock system with fluctuating demand. Probab Eng Inf Sci 23:205–230
Gupta D, Wang L (2007) Capacity Management for Contract Manufacturing. Oper Res 55:367–377
Ha AY (1997a) Inventory rationing in a maketostock production system with several demand classes and lost sales. Manag Sci 43:1093–1103
Ha AY (1997b) Stockrationing policy for a maketostock production system with two priority classes and backordering. Naval Res Logist 44:457–472
Hopp WJ, RoofSturgis M (2001) A simple, robust leadtimequoting policy. Manuf Service Oper Manag 3:321–336
Kaminsky P, Kaya O (2009) Combined maketoorder/maketostock supply chains. IIE Trans 41:103–119
Kapuscinski R, Tayur S (2007) Reliable duedate setting in a capacitated MTO system with two customer classes. Oper Res 55:56–74
Keskinocak P, Ravi R, Tayur S (2001) Scheduling and reliable leadtime quotation for orders with availability intervals and leadtime sensitive revenues. Manag Sci 47:264–279
Li L (1992) The role of inventory in deliverytime competition. Manag Sci 38:182–197
Palaka K, Erlebacher S, Kropp DH (1998) Leadtime setting, capacity utilization, and pricing decisions under leadtime dependent demand. IIE Trans 30:151–163
Pibernik R, Yadav P (2008) Dynamic capacity reservation and due date quoting in a maketoorder system. Naval Res Logist 55:593–611
Pibernik R, Yadav P (2009) Inventory reservation and realtime order promising in a maketostock system. OR Spectrum 31:281–307
Puterman ML (1994) Markov decision processes: discrete stochastic dynamic programming. Wiley, New York
Savasaneril S, Griffin PM, Keskinocak P (2010) Dynamic leadtime quotation for an M/M/1 basestock inventory queue. Oper Res 58:383–395
Slotnick SA, Sobel MJ (2005) Manufacturing leadtime rules: customer retention versus tardiness costs. Eur J Oper Res 163:825–856
Topkis DM (1968) Optimal ordering and rationing policies in a nonstationary dynamic inventory model with n demand classes. Manag Sci 15:160–176
Upasani A, Uzsoy R (2008) Incorporating manufacturing lead times in joint productionmarketing models: a review and some future directions. Ann Oper Res 161:171–188
Vericourt F, Karaesmen F, Dallery Y (2002) Optimal stock allocation for a capacitated supply system. Manag Sci 48:1486–1501
Weber RR, Stidham S (1987) Optimal control of service rates in networks of queues. Adv Appl Probab 19:202–218
Webster S (2002) Dynamic pricing and leadtime policies for maketoorder systems. Decis Sci 33:579–600
Wein LM (1991) Duedate setting and priority sequencing in a multiclass M/G/1 queue. Manag Sci 37:834–850
Weng ZK (1999) Strategies for integrating lead time and customerorder decisions. IIE Trans 31:161–171
Yano CA, Gilbert SM (2005) Coordinated pricing and production procurement decisions: a review. Chakravarty AK, Eliashberg J (eds) International series in quantitative marketing. Springer US. Managing Business Interfaces
Author information
Authors and Affiliations
Corresponding author
Appendices
Appendix 1: Proof of Theorem 1
We first discuss the existence of an optimal policy under the average reward criteria. Establishing the existence under average reward criteria is not straightforward when state space is countable and onestep rewards are not bounded. Weber and Stidham (1987) provides the sufficient conditions for existence of an average reward optimal policy. Our Markov decision process satisfies the conditions; therefore, an average reward optimal policy exists. Also the Markov decision process under consideration is a communicating multichain which implies that the optimal policy has a constant gain (Puterman 1994). This implies that under the optimal policy there exists a single recurrent class and possibly a set of transient states.
In our problem, we can assume without loss of optimality that production is stopped under a sufficiently large stock level (say, \(i=L_B<0\)), and that all customers are rejected when the cost due to the expected lateness is sufficiently high (say, \(i=L_U>0\)). This assumption can be made since the holding cost is convex increasing in stock level and lateness cost is convex increasing in number of customers in the system.
In the following we show that for \(i\in \mathbb {Z}\), \(v(i)v(i+1)\) is increasing in i. We analyze the cases \(i<0\) and \(i\ge 0\) separately. Let \(\Delta _i=v(i)v(i+1)\), \(\bar{\Delta _i}=v(i)v(i1)\), \(\Lambda =\sum _j\lambda _j\), \(\gamma =\Lambda +\mu \).
\(\underline{\hbox {Case }1\;(i<0)}\) At \(i=L_B\) production stops, whereas for \(i>L_B\) a decision as to whether to stop or continue production is taken. For \(i\le L_B\), the functional equation is expressed as follows:
For \(L_B<i<0\),
Note that for \(i\le L_B\), \(\Delta _i\) and \(\Delta _{i1}\) have almost the same expression in (7) except the term \(\frac{(h)i^}{\Lambda }\). Since \(\frac{(h)i^}{\Lambda }\) is increasing in i, \(\Delta _i>\Delta _{i1}\) for \(i\le L_B\). For \(i=L_B+1\), \(\Delta _{i}\) and \(\Delta _{i1}\) have almost the same expressions in (7) and (8) except that \(\Delta _i\) has the additional term \(\frac{\mu }{\Lambda }\max \{\Delta _{i1},0\}\ge 0\). Thus \(\Delta _i\ge \Delta _{i1}\). For \(i>L_B+1\), the additional term is increasing in i which implies \(\Delta _i\ge \Delta _{i1}\). Therefore, for \(i<0\), \(\Delta _i\) is increasing in i.
\(\underline{\hbox {Case }2\;(i\ge 0)}\) Note that \(\bar{\Delta }_i=v(i)v(i1)=\Delta _{i1}\). To show that \(\Delta _i\) is increasing in i, we show \(\bar{\Delta }_i\) is decreasing in i.
Note that in Case 1 we have shown \(\Delta _{1}>\Delta _{2}\), which is equivalent to \(v(1)v(0)>v(2)v(1)\), or \(\bar{\Delta }_0<\bar{\Delta }_{1}\). We show \(\bar{\Delta }_i<\bar{\Delta }_{i1}\) for \(0<i\le L_U\).
For states \( 0<i\) and \(i=0\), the functional equation in (2) can be expressed, respectively, as follows:
Rearranging the terms in (10) and (11), we obtain
By definition of \(L_U\) rejecting all arriving customers is the optimal decision for \(i\ge L_U\). Then \(\bar{\Delta }_i=\frac{g}{\mu }\) for \(i\ge L_U\). We use induction in the proof. For \(i\ge n+1\) suppose \(\bar{\Delta }_{i1}\ge \bar{\Delta }_i\). For \(2\le i\le n\), let \(d_j^*(i)\) be the optimal quote in state i for customer class j. For \(i=n\), comparing \(\bar{\Delta }_{i1}\) with \(\bar{\Delta }_i\),
The first inequality holds since \(d_j^*(i)\) is not necessarily the optimal action at state \(i1\), and the second inequality holds since \(\bar{\Delta }_{i1}\ge \bar{\Delta }_{i}\) for \(i\ge n+1\), and since \(L_{i1}(d_j^*(i))<L_i(d_j^*(i))\). Finally, we show \(\bar{\Delta }_{i1}\ge \bar{\Delta }_i\) for \(i=1\). Note that by the definition of \(\bar{\Delta }_0\) in (13), \(\bar{\Delta }_0\ge \bar{\Delta }_1\) holds, since \(\max \{\bar{\Delta }_0,0\}\ge 0\).
Analysis under \(i<0\) and \(i\ge 0\) shows that \(\Delta _i\) is increasing in i, \(i\in \mathbb {Z}\). The following are inferred from this result:

(i)
The operator used to determine the optimal production decision for a given nonnegative state is \(\max \{v(i1),v(i)\}\) (see Eq. 2). Rearranging the terms, one obtains \(\max \{0,v(i)v(i1)\}+v(i1)=\max \{0,\bar{\Delta }_i\}+v(i1)\). We have shown that \(\bar{\Delta }_i\) is decreasing in i. This implies that there exists a state s such that for \(i>s\) to produce is the optimal decision, and for \(i\le s\) not to produce is the optimal decision. This implies that optimal production policy is of controllimit type.

(ii)
Since \(\bar{\Delta }_i\) is decreasing in i (i.e., \(\Delta _i\) increasing in i), from Eq. 8 it is possible to infer that the term \(\max \{R_j,\Delta _i\}\) is equal to \(\Delta _i\) for sufficiently large values of i, and the term is equal to \(R_j\) for other values of i. Thus, for \(i<0\), there possibly exists a threshold \(K_j\), such that for \(i<K_j\), arriving customers are accepted and for \(K_j\le i\le 0\) arriving customers are rejected, when there is stock. If, on the other hand, it holds that \(R_j>\Delta _i\) for \(i\le 0\), then \(K_j\) does not exist and all arriving customers of class j are accepted whenever there is stock. In that case, there exists a threshold \(B_j\) at which customers of class j are rejected whenever the backlog in the system is \(B_j\) or more. It is possible to infer observing the Eqs. 8 and 12 that either a \(K_j\) or a \(B_j\) exists (but not both).

(iii)
Optimal quoted lead times increase with the number of customers in the system. To see why, observe that for \(i\ge 0\) for class j,
$$\begin{aligned} d_j^*(i)&=\hbox {arg max}_{d_j}\left\{ \bar{F}_j(d_j)(R_j\ell L_{i+1}(d_j)+\bar{\Delta }_{i+2})\right. \nonumber \\&\quad \left. +\bar{F_j}(d_j)(\bar{\Delta }_{i+1}\bar{\Delta }_{i+2}+\ell (L_{i+1}(d_j)L_{i}(d_j)))\right\} , \end{aligned}$$(14)$$\begin{aligned} d_j^*(i+1)&=\hbox {arg max}_{d_j}\left\{ \bar{F}_j(d_j)(R_j\ell L_{i+1}(d_j)+\bar{\Delta }_{i+2})\right\} . \end{aligned}$$Note that in (14) \(\bar{F}_j(d_j)(\bar{\Delta }_{i+1}\bar{\Delta }_{i+2})+\ell (L_{i+1}(d_j)L_{i}(d_j)))\) is decreasing in \(d_j\), since \((\bar{\Delta }_{i+1}\bar{\Delta }_{i+2})\) is positive, and \((L_{i+1}(d_j)L_{i}(d_j))\) is positive decreasing in \(d_j\) and \(\bar{F}_j(d_j)\) is decreasing in \(d_j\). This implies \(d_j^*(i)\le d_j^*(i+1)\), \(i\ge 0\), i.e., longer lead times are quoted to class j.
Appendix 2: Proof of Proposition 1

(1)
Consider Eq. (12). Quotes to class k in state i can be expressed as \(d_k(i)=\hbox {arg max}_{d}\{\bar{F}_k(d)(R_k\ell L_i(d)+\bar{\Delta }_{i+1})\}\), or
$$\begin{aligned} d_k(i)=\hbox {arg max}_{d}\left\{ \bar{F}_j(d)(R_j\ell L_i(d)+\bar{\Delta }_{i+1})+\bar{F}_k(d)(R_kR_j)\right\} . \end{aligned}$$The equivalence holds since it is assumed that \(\bar{F}_j(d)=\bar{F}_k(d)\). Since \(\bar{F}_k(d)\) is decreasing in d and \(R_kR_j>0\) quoted lead times to class k are shorter than lead times to class j. Shorter lead times imply \(K_k\le K_j\).

(2)
If a customer class is more tolerant to lead times, i.e., \(F_k(d)<F_j(d)\), this does not imply that quoted lead times to class k are longer. However, under the further (sufficient) condition \(\bar{F}_k(d)\bar{F}_j(d)\) is increasing in d, it is possible to show that quoted lead times to class k are longer. Optimal quote to class k in state i can be expressed as
$$\begin{aligned} d_k(i)= & {} \hbox {arg max}_{d} \left\{ \bar{F}_j(d)(R_j\ell L_i(d)+\bar{\Delta }_{i+1})+(\bar{F}_k(d)\bar{F}_j(d))\right. \nonumber \\&\left. (R_k\ell L_i(d)+\bar{\Delta }_{i+1})\right\} . \end{aligned}$$(15)
Note that the d value defined by (15) will be in the set of values that make \(R_k\ell L_i(d)+\bar{\Delta }_{i+1}>0\). If there is no such d value, then \(d_k(i)=d_\mathrm{max}=d_j(i)\). For states where for some \(d\in [0,d_\mathrm{max}]\), \(R_k\ell L_i(d)+\bar{\Delta }_{i+1}>0\), lead times for class k are longer if \(\bar{F}_k(d)\bar{F}_j(d)\) is increasing in d. The reason is that \(R_k\ell L_i(d)+\bar{\Delta }_{i+1}\) is increasing in d which implies \((\bar{F}_k(d)\bar{F}_j(d))(R_k\ell L_i(d)+\bar{\Delta }_{i+1})\) is an increasing function of d. Thus lead times to class k are longer than lead times to class j.
Appendix 3: Proof of Lemma 1
We show that under a basestock level \(s\ne s^*\) quoted lead times to arriving customers are shorter, and stock rationing levels are closer to 0. We analyze \(s<s^*\) and \(s>s^*\) separately.
\(\underline{\hbox {Case }1.\; s<s^*}\) Let \(v_s(i)\) denote the bias and \(g_s\) the expected average gain per unit time under basestock s. Let \(\bar{\Delta }_i(s)=v_s(i)v_s(i1)\). First we express the Eqs. (12) and (13) under the optimal basestock, \(s^*\).
Similarly, the following equalities are defined under the nonoptimal basestock, s,
In the proof, we first show that \(\bar{\Delta }_i(s)>\bar{\Delta }_i\), for \(i>s\). As we will show, this implies quoted lead times are shorter and stock rationing levels are closer to zero. AktaranKalayci and Ayhan (2009) analyze the effect of parameters on the pricing decisions in a maketoorder setting under average reward criteria. Proof follows similar lines.
\(\underline{\hbox {For }i>s, \bar{\Delta }_i(s)>\bar{\Delta }_i}\) Consider Eq. (16). Let \(d_j^*(i)\) denote the optimal quote to class j in state i under basestock \(s^*\). Then \(\bar{\Delta }_i(s)\) can be expressed as
In (18), note that \(\frac{g^*}{\mu }\frac{(h)i^}{\mu } +\sum _j\frac{\lambda _j}{\mu }\bar{F}_j(d_j^*(i)) (R_j\ell L_i(d_j^*(i))+\bar{\Delta }_{i+1})\) is simply \(\bar{\Delta }_i\). Furthermore, \(\frac{g^*}{\mu }\frac{g_s}{\mu } +\sum _j\frac{\lambda _j}{\mu }\bar{F}_j(d_j^*(i))(\bar{\Delta }_{i+1}(s)\bar{\Delta }_{i+1})\) is always positive (through induction assumption, and since \(\frac{g^*}{\mu }\ge \frac{g_s}{\mu }\)). Thus for \(i>s\), \(\bar{\Delta }_i(s)>\bar{\Delta }_i\).
\(\underline{\bar{\Delta }_i(s)>\bar{\Delta }_i \hbox { implies shorter lead times}}\) Now we show that under s, lead times are shorter for \(i\ge s\). Under \(s^*\), optimal quote in i is \(d_j^*(i)=\hbox {arg max}_{d}\{\bar{F}_j(d) (R_j\ell L_i(d)+\bar{\Delta }_{i+1})\}\), whereas under s, optimal quote is \(d_j^s(i)=\hbox {arg max}_{d}\{\bar{F}_j(d) (R_j\ell L_i(d)+\bar{\Delta }_{i+1}+\bar{F}_j(d)(\bar{\Delta }_{i+1}(s)\bar{\Delta }_{i+1})\}\). Since for \(i\ge s\), \(\bar{F}_j(d)(\bar{\Delta }_{i+1}(s)\bar{\Delta }_{i+1})\) is decreasing in d, it holds that \(d_j^s(i)\le d_j^*(i)\).
\(\underline{\hbox {Case 2}.\; s>s^*}\) For \(i>s^*\), showing that \(\bar{\Delta }_i(s)>\bar{\Delta }_i\) follows similar lines as in Case 1. And thus the conclusion of shorter lead times and lower stock rationing levels follows.
Appendix 4: Proof of Theorem 2
We show that under s, the structural results for rationing and lead time quotes hold as stated in Theorem 1: there exists a rationing level (possibly negative) for each customer and lead time quotes increase with the number of customers in the system. To show that the structure under s does not change, we equivalently show that \(\bar{\Delta }_i(s)\) is decreasing in i in the “relevant region”. Note that if \(\bar{\Delta }_i(s)\) is not decreasing in i, then lead time quotes are not necessarily monotone increasing, and there might exist several stock rationing levels for class j (where the highest of those levels would be lower than \(K_j\) under he optimal policy, as stated in Lemma 1). We analyze \(s<s^*\) and \(s>s^*\) separately.
\(\underline{\hbox {Case 1}.\;s<s^{*}}\) We show that for \(i> s\) \(\bar{\Delta }_i(s)\) is decreasing in i. If \(s=0\), then similar analysis as in Case 2 of Theorem 1 yields \(\bar{\Delta }_i(s)\ge \bar{\Delta }_{i+1}(s)\) for \(i>0\).
If \(s>0\), then for \(i\ge 0\) similar analysis as in Case 2 of Theorem 1 yields \(\bar{\Delta }_i(s)\ge \bar{\Delta }_{i+1}(s)\). For \(s<i<0\), due to the term \(\frac{(h)i}{\mu }\) it is not immediate that the inequality holds. We show for \(s<i\), \(\bar{\Delta }_i(s)>\bar{\Delta }_{i+1}(s)\) by showing that for \(s<i\), \(\bar{\Delta }_i(s)\bar{\Delta }_i>\bar{\Delta }_{i+1}(s)\bar{\Delta }_{i+1}\). Since \(\bar{\Delta }_i>\bar{\Delta }_{i+1}\), this implies \(\bar{\Delta }_i(s)>\bar{\Delta }_{i+1}(s)\).
Let \(\delta _i=\bar{\Delta }_i(s)\bar{\Delta }_i\). There exists a state \(m>0\) at which rejecting all arriving customers is the optimal decision under basestock s and \(s^*\). Let \(\delta _m=\bar{\Delta }_m(s)\bar{\Delta }_m=\frac{g_s}{\mu }(\frac{g^*}{\mu })\). For \(s<i\) we check whether \(\delta _i>\delta _{i+1}\). In other words we check
Since for \(i=m\), optimal decision is to reject all arriving customers, RHS and LHS of the inequality in (19) are equal to \(\frac{g^*}{\mu }\frac{g_s}{\mu }\). We make the proof by induction. Suppose for state \(i\ge n\), \(\delta _n\ge \delta _{n+1}\), i.e., \(\bar{\Delta }_n(s)\ge \bar{\Delta }_{n+1}(s)\). We show, \(\delta _{n1}>\delta _n\),
To show that (20) holds, we introduce the following lemma:
Lemma 3
Let f(x) be a decreasing function of \(x,~x\in X=[0,\bar{x}], f(x)\ge 0\) where \(f(\bar{x})=0\). Let \(c_1,~c_2,~c_3,c_4\in \mathbb {R}\) such that \(c_1>c_2>c_4\), \(c_1>c_3>c_4\), and \(c_1c_2>c_3c_4>0\). Let \(L_i(x)\) be a function of \(i\in \mathbb {Z}\) and x, as defined in Eq. (1). Then for \(i\in \mathbb {Z},~x\in X\),
Proof
We prove the lemma in three steps:

(1)
Show that for \(i\in \mathbb {Z},~x\in X\),
$$\begin{aligned}&\max _x\{f(x)(c_1L_i(x))\}\max _x\{f(x)(c_2L_i(x))\} \nonumber \\&\quad \ge \max _x\{f(x)(c_3L_{i}(x))\}\max _x\{f(x)(c_1(c_3c_4)L_{i}(x))\}. \end{aligned}$$(21) 
(2)
Show that
$$\begin{aligned}&\max _x\{f(x)(c_1L_i(x))\}\max _x\{f(x)(c_1(c_3c_4)L_i(x))\} \nonumber \\&\quad \ge \max _x\{f(x)(c_3L_{i}(x))\}\max _x\{f(x)(c_4L_{i}(x))\}. \end{aligned}$$(22) 
(3)
Show that
$$\begin{aligned}&\max _x\{f(x)(c_3L_i(x))\}\max _x\{f(x)(c_4L_i(x))\} \nonumber \\&\quad \ge \max _x\{f(x)(c_3L_{i+1}(x))\}\max _x\{f(x)(c_4L_{i+1}(x))\}. \end{aligned}$$(23)Note that \(L_i(x)\), and \(L_{i+1}(x)L_i(x)\) are positive decreasing functions of x.

(1)
Equation (21) simply follows from \(c_1(c_3c_4)>c_2\), and \(\max _x\{f(x)(c_1(c_3c_4)L_i(x))\}>\max _x\{f(x)(c_2L_i(x))\}\).

(2)
Consider the function \(f(x)L_i(x)\), which is a negative increasing function with value 0 at \(x=\bar{x}\). Let \(c_{2'}=c_1(c_3c_4)\) and define \(x_1, x_{2'}, x_3, x_4\) as \(x_k=\mathrm{arg\,max}_{x\in X}f(x)(c_kL_i(x)),~k\in \{1,2',3,4\}\). Since \(c_1>c_{2'}>c_4\) and \(c_1>c_3>c_4\), it holds that \(x_1<x_{2'}<x_4\) and \(x_1<x_3<x_4\). Then two cases are possible, \(\underline{x_{2'}<x_3}\) Observe that
$$\begin{aligned}&f(x_1)(c_1L_i(x_1))f(x_2)(c_{2'}L_i(x_{2'}))\\&\qquad>f(x_{2'})(c_1L_i(x_{2'}))\\&\qquad f(x_{2'})(c_{2'}L_i(x_{2'}))\\&\quad>f(x_3)(c_3L_i(x_3))f(x_3)(c_{4}L_i(x_{3}))\\&\quad >f(x_3)(c_3L_i(x_3))f(x_4)(c_{4}L_i(x_{4})), \end{aligned}$$where the first inequality is due to \(x_1\) being the maximizer of \(f(x)(c_1L_i(x))\), the second inequality is due to \(c_1c_{2'}=c_3c_4\) and \(f(x_{2'})>f(x_3)\), and the last inequality is due to \(x_4\) being the maximizer of \(f(x)(c_4L_i(x))\). This implies (22) holds. \(\underline{x_{2'}>x_3}\) Observe that
$$\begin{aligned}&f(x_1)(c_1L_i(x_1))f(x_3)(c_{3}L_i(x_{3}))\\&\quad>f(x_{3})(c_1L_i(x_{3})) f(x_{3})(c_{3}L_i(x_{3}))\\&\quad>f(x_{2'})(c_{2'}L_i(x_{2'}))f(x_{2'})(c_{4}L_i(x_{2'}))\\&\quad >f(x_{2'})(c_{2'}L_i(x_{2'}))f(x_4)(c_{4}L_i(x_{4})), \end{aligned}$$where the first inequality is due to \(x_1\) being the maximizer of \(f(x)(c_1L_i(x))\), the second inequality is due to \(c_1c_3=c_{2'}c_4\) and \(f(x_{2'})<f(x_3)\), and the last inequality is due to \(x_4\) being the maximizer of \(f(x)(c_4L_i(x))\). This implies that (22) holds.

(3)
Note that \(f(x)(c_kL_{i+1}(x))=f(x)c_kf(x)L_i(x)f(x)(L_{i+1}L_i(x))\), \(k\in \{1,2,3,4\}\), and that \(L_{i+1}(x)L_i(x)\) is a positive decreasing function of x. Let \(x_3+\) and \(x_4+\) be defined as \(x_3+=\hbox {arg max}_{x\in X}f(x)(c_3L_{i+1}(x))\) and \(x_4+=\hbox {arg max}_{x\in X}f(x)(c_4L_{i+1}(x))\). Since \(f(x)(L_{i+1}(x)L_{i}(x))\) is increasing in x, and \(c_3>c_4\), it holds that \(x_3<x_{3}+<x_4+\) and \(x_3<x_4<x_{4}+\). Then two cases are possible. \(\underline{x_4<x_3+}\) Observe that
$$\begin{aligned}&f(x_3)(c_3L_i(x_3))f(x_4)(c_{4}L_i(x_{4}))\\&\quad>f(x_{4})(c_3L_i(x_{4}))f(x_{4})(c_{4}L_i(x_{4}))\\&\quad>f(x_3+)(c_3L_{i+1}(x_3+))f(x_3+)(c_{4}L_{i+1}(x_{3}+))\\&\quad >f(x_3+)(c_3L_{i+1}(x_3+))f(x_4)(c_{4}L_{i+1}(x_{4})), \end{aligned}$$where the first inequality is due to \(x_3\) being the maximizer of \(f(x)(c_3L_i(x))\), the second inequality is due to \(f(x_{4})>f(x_3+)\), and the last inequality is due to \(x_4\) being the maximizer of \(f(x)(c_4L_{i+1}(x))\). This implies that (23) holds. \(\underline{x_4>x_3+}\) Observe that
$$\begin{aligned}&f(x_3)(c_3L_i(x_3))f(x_3+)(c_{3}L_i(x_{3}+))\\&\quad>f(x_{3}+)(c_3L_i(x_{3}+))f(x_{3}+)(c_{3}L_{i+1}(x_{3}+))\\&\quad>f(x_{4})(c_{4}L_i(x_{4}))f(x_{4})(c_{4}L_{i+1}(x_{4}))\\&\quad >f(x_{4})(c_{4}L_i(x_{4}))f(x_4+)(c_{4}L_{i+1}(x_{4}+)), \end{aligned}$$where the first inequality is due to \(x_3\) being the maximizer of \(f(x)(c_3L_i(x))\), the second inequality is due to \(f(x_{4})(L_{i+1}(x_4)L_i(x_4))<f(x_3+)(L_{i+1}(x_3+)L_i(x_3+))\), and the last inequality is due to \(x_4+\) being the maximizer of \(f(x)(c_4L_{i+1}(x))\). This implies that (23) holds.
Lemma 3 implies that (20) holds as follows: Observe that \(\bar{\Delta }_n(s)>\bar{\Delta }_n\), \(\bar{\Delta }_{n+1}(s)>\bar{\Delta }_{n+1}\) (by Lemma 1), \(\bar{\Delta }_n(s)>\bar{\Delta }_{n+1}(s)\) (by induction assumption \(\delta _n>\delta _{n+1}\)), and \(\bar{\Delta }_{n}>\bar{\Delta }_{n+1}\) (by Theorem 1). Thus letting \(c_1=R_j+\bar{\Delta }_n(s)\), \(c_2=R_j+\bar{\Delta }_n\), \(c_3=R_j+\bar{\Delta }_{n+1}(s)\), \(c_4=R_j+\bar{\Delta }_{n+1}\), and \(f(x)=\bar{F}(x)\), Lemma 3 implies that (20) holds.
Thus the proof of for \(s<s^*\), \(\bar{\Delta }_i(s)\) is decreasing in i for \(i>s\) is complete. We conclude that the structure of rationing and lead time quotation policy is the same with that of the optimal policy, except that the decisions taken are different. Under \(s\), lead times are lower, and stock rationing levels are closer to zero.
\(\underline{\hbox {Case 2.}\; {s>s^*}}\) For \(i>s^*\), showing that \(\bar{\Delta }_i(s)\) is decreasing in i follows similar lines as in Case 1. For states \(i<0\), lead time quotation corresponds to the stock rationing decision. Analysis shows that \(\bar{\Delta }_i(s)\) is not necessarily decreasing in i for states \(s\le i\le s^*\) (see Fig. 7). However, immediate reasoning reveals that in all states \(s\le i <s^*\), no rationing of stock should take place for any class of the customers even if there exists a customer class k with \(R_k=0\). Any stock over \(s^*\) is a burden which causes unnecessary stocking cost. Customer arrivals help getting rid of the unnecessary stock; thus when stock level exceeds \(s^*\) rejecting a customer cannot be the optimal action. Actually, in numerical analysis we observed that even customers with negative unit revenue might be allowed to the system in states \(s\le i<s^*\).
In conclusion, when \(s\ne s^*\), the rationing and lead time quotation policies have the same structure as of the optimal policy, with lower lead times and stock rationing levels close to zero.
Appendix 5: Proof of Lemma 2
We show that under a given basestock s, a decrease in quoted lead times results in an increase in W and in L. For \(i<0\), “decrease in quote” implies lower stock rationing levels (i.e., stock rationing level will be closer to 0). Let policy P and P define two policies with P corresponding to the policy with shorter quotes. Let s be the basestock level implied by the policies. Let \(\underline{d}_k(i), k\in C, i\ge s\) be the lead time quote in state i for class k under P. We first introduce the following lemma (Puterman 1994):
Lemma 4
Let \(\{x_i\}\), \(\{x_i'\}\) be realvalued nonnegative sequences satisfying
for all \(t\ge 0\), with equality holding in (24) for \(t=0\).
Suppose \(v_{i+1}\ge v_i\) for \(i=0,1,\cdots \), then
In the problem under consideration, let \(x_i\) stand for \(\frac{\sum _j\lambda _j\bar{F}_j(d_j(i))}{\lambda _\mathrm{eff}}\pi _i\) and \(v_i\) stand for \(E_i[wait]=(\frac{i+1}{\mu })^+, for~i\in \mathbb {Z}\). Let \(\{x_i\}\) and \(\{x_i'\}\) be the sequences defined under P and P, respectively. Note that under a given policy, \(\pi _{i+1}=\frac{\sum _j\lambda _j\bar{F}_j(d_j(i)))}{\mu }\pi _i\). For the simplicity of expression and without loss of generality, we assume \(\mu =1\). Let \(\sum _j\lambda _j\bar{F}_j(d_j(i))\) be shortly denoted as \(\underline{G_i}\) and \(G_i\) under policies P and P, respectively. The \(G_i\) is the total effective arrival rate to state i. For \(t\ge s,\sum _{i=t}^{\infty }x_i\) would be expressed as \(\sum _{i=t}^{\infty } \frac{G_i}{\lambda _\mathrm{eff}}\pi _i\), where \(\lambda _\mathrm{eff}=\sum _{i=s}^{\infty }G_i\pi _i\). Note that \(\sum _{i=s}^{\infty }x_i=1\).
In the following, we first show \(\sum _{i=t}^{\infty }x_i\ge \sum _{i=t}^{\infty }x_i'\) for \(t\ge s\):
Note \(G_i\pi _i=\prod _{n=s}^iG_n\pi _i\). Thus (25) can be expressed as
Note that showing
is sufficient for (26) to hold, since \(\underline{G}_i\le G_i,~ i\ge s\). We show through recursion that the inequality in (27) holds. Assume (27) holds for \(t=r1\),
If (28) holds, then (27) holds for \(t=r\), since
holds, due to \(G_r\ge \underline{G}_r\), and denominator of the fraction in lefthandside (LHS) is smaller due to (28). For the initial state \(r=s+2\), (28) holds. Thus the inequality (25) holds. Finally, since \(v_i=E_i[\hbox {wait}]=(\frac{i+1}{\mu })^+\) is increasing in i for \(i \in \mathbb {Z}\), Lemma 24 implies \(W\ge \underline{W}\).
For the comparison of the number of outstanding orders waiting to be processed, L, we show
Since \(\underline{\pi }_i\) and \(\pi _i\) can be shown to follow \(\sum _{i=t}^{\infty }\underline{\pi }_i\le \sum _{i=t}^{\infty }\pi _i\) as in Lemma 24, it is trivial that (30) holds.
Appendix 6: Proof of Proposition 3
We define g and \(g^2\) as the optimal expected average gain under dynamic, and static quotation schemes, respectively. Note that the definition of the static quotation scheme implies that the stock rationing decisions are given optimally. Then for \(i<0\) it is possible to define the following functional equation under the static scheme:
Note that for \(i<0\) the functional equation has the same structure with that of the original dynamic quotation problem. Thus, assuming without loss of optimality that there will be a state \(L_B\) at which stopping production is the optimal decision, for \(i<0\) it is possible to show that \(\Delta _i^2=v(i)v(i+1)\) is increasing in i. This implies a controllimit type production policy is indeed optimal under the static and myopic quotation scheme.
Next, we show that the basestock level is higher under the static quotation scheme compared to the optimal scheme. For \(i<0\), it is possible to rewrite (7) and (8) as follows:
The structure of the equations are the same under dynamic, zero lead time, static, and myopic schemes; however, note that \(\frac{g^2}{\Lambda }\ge \frac{g}{\Lambda }\) since \(g^2\le g\). This implies \(\Delta _i^2\ge \Delta _i\) for \(i\le L_B\) and \(L_B<i<0\). This implies the level at which the optimal decision is to stop production will be higher under nonoptimal lead time quotation policies, i.e., \(s_2\ge s^*\).
Appendix 7: Proof of Proposition 4
The proof follows from the observation of the equation in (10). From the equation, the optimal quotes under dynamic quotation scheme is expressed as \(d_j(i)=\hbox {arg max}_{d_j\in D_j}\{\bar{F}_j(d_j)(R_j\ell L_i(d_j)+\bar{\Delta }_{i+1})\}\). Since \(\bar{\Delta }_{i+1}<0\) for \(i<0\), and since \(\bar{F}_j(d_j)\) is decreasing in \(d_j\), \(\hbox {arg max}_{d_j}\{\bar{F}_j(d_j)(R_j\ell L_i(d_j)\}\le \hbox {arg max}_{d_j}\{\bar{F}_j(d_j)(R_j\ell L_i(d_j)+\bar{\Delta }_{i+1})\}~\forall ~j\).
Appendix 8: Proof of Proposition 5
To show the effect of the change in parameter on the optimal policy structure, we use the value function. For a related study on the effect of system parameters on the structure of the optimal policy see Cil et al. (2009) and AktaranKalayci and Ayhan (2009). The value function, \(V^n(i)\), denotes the total expected reward (profit) when there are n remaining transitions, and the state is in i. We obtain \(V^n(i)\) as follows:
where \(\tau _0\) and \(\tau _j,~j\in C\) are defined on realvalued functions as
Before the proof, note that under the optimal policy the chain has a single recurrent class plus some transient states, and thus a constant gain is obtained at all states. Puterman (1994, p. 339) states that under constant gain the bias, v, is the relative difference in total expected reward as n goes to infinity. In other words,
This implies the structural behavior of \(v(i)v(i+1)\) (identified in Theorem 1) also reflects the behavior of \(V^n(i+1)\) asymptotically.
In the proof, we use the structure that \(\Delta V^n(i)=V^n(i)V^n(i+1)\) is increasing in \(i,~\forall i\in \mathbb {Z}\) as n goes to infinity.
(1) Effect of \(R_k\). We show the impact of \(R_k\) on the structure of the optimal policy. Let \(R_k^+\) denote the increased revenue for class k, \(R_k^+>R_k\). Let \(V^n_+(i)\) denote the value function under \(R_k^+\). Our aim is to show that under \(R_k^+\), quotes to class k are shorter while quotes to classes \(j\ne k\) are longer, compared to those under \(R_k\).
To show the effect of \(R_k\) on the optimal policy, we first show that under \(R_k^+\) the following hold:

(i)
\(V^n(i)V^n(i+1) < V^n(i)V^n(i+1)\), i.e., \(\Delta V^n(i) <\Delta V_+^n(i),\)

(ii)
\(\Delta V^n(i)R_k>\Delta V_+^n(i)R_k^+, ~\forall i.\)
We use induction in the proof. Initial step of induction assumes that \(\Delta V^{n1}(i)<\Delta V_+^{n1}(i)\).
(i) Show \(\Delta V^n(i)<\Delta V_+^n(i)\).
We first show that under each operator \(\tau _j\) and \(\tau _0\), \(\Delta \tau _jV^{n1}(i)<\Delta \tau _jV_+^{n1}(i),~\forall ~j\in {0}\cup C\).
\(\underline{\hbox {For }\tau _j,j\in C}\) Let \(d_i, d_{i+1}, d_i^+,d_{i+1}^+\) be maximizers of \(\tau _j V^{n1}(i), \tau _j V^{n1}(i+1),\tau _j V_+^{n1}(i),\tau _j V_+^{n1}(i+1)\), respectively. Let \(\varepsilon = R_k^+R_k\). We check whether
In (33) \(\varepsilon I_k\) takes value \(\varepsilon \in \mathbb {R}^+\) if the operator under consideration is \(\tau _k\) and 0 otherwise. The analysis is made under \(j\ne k\) and \(j=k\) separately.
\(\underline{\hbox {For }j\ne k}\) If in (33) \(d_{i+1}\) is replaced with \(d_{i+1}^+\) and \(d_i^+\) with \(d_i\):
Under the induction assumption of \(\Delta V^{n1}(i)<\Delta V_+^{n1}(i)\), the inequality in (34) and thus in (33) are satisfied.
\(\underline{For j=k}\) Note that when \(j=k\), \(\tau _jV_+^{n1}(i)\) is defined with \(R_k^+\) and let \(d_i^+\) be the corresponding maximizer of \(\tau _jV_+^{n1}(i)\).
\(\underline{\hbox {When }d_{i+1}^+<d_i}\) In (33) replace \(d_{i+1}\) with \(d_i\) and \(d_i^+\) with \(d_{i+1}^+\). Then,
Note \(d_{i+1}^+<d_i\) implies \(F_j(d_{i+1}^+)<F_j(d_i)\). Since \(L_{i+1}(d)L_i(d)>0\) and decreasing in d, and \(\Delta V^{n1}(i)\) is increasing in i, inequality (35) and thus, (33) hold.
\(\underline{\hbox {When }d_{i+1}^+>d_i}\) In (33) replace \(d_{i+1}\) with \(d_{i+1}^+\) and \(d_i^+\) with \(d_i\). Then as in (34),
The inequality holds since \(\bar{F}_j(d_i)\bar{F}_j(d_{i+1}^+)>0\).
\(\underline{\hbox {For }\tau _0}\) For \(i> 0\), \(\tau _0V^{n1}(i)=V^{n1}(i1)\) and thus \(\Delta \tau _0V^{n1}(i)<\Delta \tau _0V_+^{n1}(i)\) immediately holds since \(\Delta V^{n1}(i)<\Delta V_+^{n1}(i)\) by assumption.
For \(i\le 0\), \(\tau _0V^{n1}(i)=\max \{V^{n1}(i),V^{n1}(i1)\}\). Let \(u_i,~u_i^+\in \{0,1\}\) be the indicator variables associated with \(\tau _0V^{n1}(i)\) and \(\tau _0V_+^{n1}(i)\). We check whether
Replacing \(u_{i+1}\) with \(u_{i+1}^+\) and \(u_i^+\) with \(u_i\) gives
Since \(\Delta V^{n1}(i)<\Delta V_+^{n1}(i)\), inequality (37) and thus (36) holds.
That \(\Delta \tau _jV^{n1} (i)<\Delta \tau _jV_+^{n1}(i),~j\in \{0\}\cup C\) is sufficient for \(\Delta V^n(i)<\Delta V_+^n(i)\).
(ii) Show \(\Delta V^n(i)R_k>\Delta V_+^n(i)R_k^+\).
We show that under each operator \(\tau _j\), \(\tau _0\) \(\Delta \tau _j V^{n1}(i)R_k>\Delta \tau _j V_+^{n1}(i)R_k^+\) and \(\Delta \tau _0 V^{n1}(i)R_k>\Delta \tau _0 V_+^{n1}(i)R_k^+\).
For \(\tau _j,j\in C\), let \(d_i\) and \(d_i^+\) be maximizers of \(\tau _j V^{n1}(i)\) and \(\tau _j V_+^{n1}(i)\). We check whether
\(\underline{\hbox {For }j\ne k}\) We know that \(d_i<d_{i+1}<d_{i+1}^+\). Similar to inequality (34) the following inequality is obtained:
Observe that \(1F_j(d_i)\bar{F}_j(d_{i+1}^+)>0\), so inequality (39) and thus (38) hold. \(\underline{\mathrm{For} j=k.}\) This case is analyzed separately for \(d_{i+1}^+>d_i\) and \(d_i<d_{i+1}^+\).
\(\underline{\hbox {When }d_{i+1}^+<d_i}\) In (38) replace \(d_{i+1}\) with \(d_{i+1}^+\) and \(d_i^+\) with \(d_i\), then as in (34) we obtain
By the induction assumption the inequality holds and thus (38) holds.
\(\underline{\hbox {When }d_{i+1}^+>d_i}\) Replace \(d_{i+1}\) with \(d_i\) and \(d_i^+\) with \(d_{i+1}^+\) in (38). Similar to (40) the following is obtained:
Since \(d_{i+1}^+>d_i\), \(\bar{F}_j(d_i)>\bar{F}_j(d_{i+1}^+)\). Induction assumption implies \(\Delta V^{n1}(i)R_k>\Delta V^{n1}(i)R_k^+\) and thus in (41) the sum of first two terms in the lefthand side (LHS) is greater than the sum of first two terms on the righthand side (RHS). Furthermore \(L_{i+1}(d)L_i(d)>0\) and decreasing in d and thus the inequality (41) holds. Therefore (38) holds.
For the analysis of \(\tau _0\), let \(u_i, u_{i+1}\in \{0,1\}\) be indicator variables associated with \(\tau _0 V^{n1}(i)\) and \(\tau _0 V_+^{n1}(i)\), respectively. We check
Replacing \(u_{i+1}\) with \(u_i\) on LHS of (42) and \(u_i^+\) with \(u_{i+1}^+\) on RHS yields
the inequality and thus (42) holds.
(2) Effect of \(\lambda _k\). We show the impact of \(\lambda _k\) on the structure of the optimal policy. Let \(\lambda _k^+\) denote the increased arrival rate for class k. We show that under \(\lambda _k^+\), \(\Delta V^{n1}(i)<\Delta V_+^{n1}(i)\). This condition is sufficient to conclude that under \(\lambda _k^+\), for all classes basestock and rationing levels increase, and the lead time quotes are longer. For \(\tau _j,~j\in \{0\}\cup C\), it is shown that \(\Delta \tau _jV^{n1}(i)<\Delta \tau _j V_+^{n1}(i)\).
\(\underline{\hbox {For }\tau _j,~j\in C}\) Let \(d_i\), \(d_i^+\), \(d_{i+1}\), \(d_{i+1}^+\) be maximizers of \(\tau V^{n1}(i)\), \(\tau V_+^{n1}(i), \tau V^{n1}(i+1)\), and \(\tau V_+^{n1}(i+1)\), respectively. \(\underline{\hbox {For }\tau _0.}\) Let \(u_i\), \(u_i^+\), \(u_{i+1}\), \(u_{i+1}^+\) be maximizers of \(\tau _0 V^{n1}(i)\), \(\tau _0 V_+^{n1}(i), \tau _0 V^{n1}(i+1)\), and \(\tau _0 V_+^{n1}(i+1)\), respectively.
Similar analysis as in (33), (34), (36) and (37) would imply that \(\Delta V^{n1}(i)<\Delta V_+^{n1}(i)\) holds.
From (8), this implies as \(\lambda _k\) increases, the basestock \(s^*\), and the rationing levels increase. From (12), a decrease in \(\bar{\Delta }_{i+1}\) leads to an increase in the optimal quote for state i. Thus, as \(\lambda _k\) increase so do the optimal quotes.
Rights and permissions
About this article
Cite this article
Savasaneril, S., Sayin, E. Dynamic lead time quotation under responsive inventory and multiple customer classes. OR Spectrum 39, 95–135 (2017). https://doi.org/10.1007/s002910160445z
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s002910160445z
Keywords
 Lead time quotation
 Maketostock queue
 Revenue management
 Multiple customer classes
 Inventory rationing
 Control of queues