DMS201 Module-3

Aggregate planning

Aggregate unit: Aggregate planning is about planning the labour requirement for production.
It starts with forecasting future demand of a product family. One question arises immediately
– how does one forecast demand of a product family? We have the past demand data of
individual products, which can be used to forecast individual product demands. How can we
forecast demand of a product family using the product-level demand data? We can simply
add up the product demands and refer to the sum as the demand data for the product family.
Then we can use that data and obtain demand forecast for the product family. We can see
why the forecast of aggregate is less erroneous than individual forecasts.

The above approach of aggregating individual product raises one concern, though. What is
this aggregate unit? What are its features, e.g., labour requirement, cost? It is a hypothetical
unit, and its features can be obtained as weighted averages of the features of the individual
products with the sales shares as the weights. An illustration is given below.

Past demand Average Labour-hr

Model Cost
Month-1 Month-2 Month-3 Month-4 share (%) required
M1 143 147 145 150 30.46 4.2 285
M2 96 90 85 82 18.38 4.9 345
M3 210 190 225 200 42.89 3.7 230
M4 35 39 41 44 8.28 5.5 410
Aggregate 484 466 496 476 100 4.22 282.8

Past demand data of the hypothetical aggregate unit and its features are mentioned in the last
row of the table. In the rest of this note, our focus is on this aggregate unit. Note that a similar
principle is applied for capacity planning, where the aggregation happens at the plant level. It
is worth noting that there are other ways of aggregation.

Costs in aggregate planning: Let us consider the following demand forecasts of an aggregate
unit at the start of April. We need to come up with a production plan to fulfil the demand in a
cost-effective manner. It shall be noted that our purpose is to obtain labour requirements for
the future. The production quantities are decided in the next level of planning, i.e., materials
requirement planning. At that stage, labours are already decided and hence act as a limit on
production. Similarly, equipment is already deployed at the time of aggregate planning and
hence acts as a limit on aggregate production. Let us ignore the limit for now.

Month May Jun Jul Aug Sep Oct

Forecast 300 400 500 450 400 350

The total demand for the next six months is 2400 units. We can fulfil it in different manners,
leading to different types of costs. Suppose we produce the entire 2400 units in May itself.

DMS201 Notes prepared by Dr. Avijit Khanra

Then we must have hired many labours for May. We would have incurred huge amount of
hiring cost, which involves cost of recruitment and training. In the next five months, we are
going to pay the large labour force without extracting any work. The wage without work is
referred to as the idle time cost. We can reduce the idle time cost by firing excess workers,
which has its own costs. Firing cost includes compensatory payments and loss of reputation
of the firm, which is difficult to measure compared to the other two.

Apart from the three labour related costs, there are two materials related costs. If we produce
2400 units in May, 300 units will be consumed in May and the remaining 2100 units will be
carried forward to June. Similarly, 1700 units will be carried forward from June to July, 1200
from July to August, 750 from August to September, and 350 from September to October.
When we carry large inventory, we need large working capital, which comes with interest.
This cost is referred to as the inventory holding cost.

Suppose the nature of the product and the market is such that the customers are willing to
wait if the production is not sufficient meet the demand. This allows greater flexibility to the
planner. For example, we can choose to produce only in July and August – 1200 units in both
the months. Then the demands of May and June are backlogged and fulfilled in July. Backlog
is 300 units at the end of May and 700 units at the end of June. Backlogs are responded with
expediting delivery, and sometimes discounts/gifts. In a competitive market, backlogging is
not possible. Then a stock-out leads to loss of sales (to competitors) and loss of reputation.
Backlogging and stock-out related costs are referred to as shortage cost.

A cost-effective aggregate production plan fulfils the aggregate demand such that the hiring,
firing, idle time, holding, and shortage costs are minimized. It is difficult to come up with a
simple solution method with so many types of costs. Next, we see two extreme strategies to
fulfil the demands. The best plan is somewhere in-between them.

Demand chase strategy: In the demand chase strategy, every month we produce just enough
to meet the months’ demand. Then the production quantity is same as the demand in every
month. Consequently, there is no inventory holding or shortage. We only incur labour-related
costs. Suppose a product requires a total of 0.5 labour-month. Then the table below shows the
labour requirement in different months. We assume that April ends without any inventory or
shortage. We represent shortage by negative inventory.

Month Opening May Jun Jul Aug Sep Oct

Forecast -- 300 400 500 450 400 350
Production -- 300 400 500 450 400 350
Inventory 0 0 0 0 0 0 0
Labour -- 150 200 250 225 200 175

Suppose 200 workers are employed in April. If we choose to maintain the workforce size as
shown in the last row of the above table, then we need to follow the hiring/firing scheme as
shown in the first part of the table below. There is no idle worker at any stage in this scheme.

DMS201 Notes prepared by Dr. Avijit Khanra

It appears that we are hiring/firing too frequently. Allowing some idle workers for a month or
two may not be a bad idea. The second part of the table below shows a scheme to achieve the
same workforce level with some idle worker in May.

Month Opening May Jun Jul Aug Sep Oct

Strategy
Labour 200 150 200 250 225 200 175
Hiring -- 0 50 50 0 0 0
No idle
Firing -- 50 0 0 25 25 25
worker
Idle -- 0 0 0 0 0 0
Hiring -- 0 0 50 0 0 0
With idle
Firing -- 0 0 0 25 25 25
worker
Idle -- 50 0 0 0 0 0

Let 𝑐ℎ and 𝑐𝑓 denote the cost of hiring and firing a worker and 𝑐𝑤 denote the idle time wages
of a worker per month. Then the cost of ‘no idle worker’ strategy is 100𝑐ℎ + 125𝑐𝑓 and the
cost with idle worker is 50𝑐ℎ + 75𝑐𝑓 + 50𝑐𝑤 . The second strategy is better if 50𝑐ℎ + 75𝑐𝑓 +
50𝑐𝑤 ≤ 100𝑐ℎ + 125𝑐𝑓 , i.e., 𝑐𝑤 ≤ 𝑐ℎ + 𝑐𝑓 , which is typically true. If no idle worker strategy
suggests firing of a worker in a month and then hiring him back after a gap of 𝑔 months, then
keeping the worker idle is better if 𝑔𝑐𝑤 ≤ 𝑐ℎ + 𝑐𝑓 ≡ 𝑔 ≤ (𝑐ℎ + 𝑐𝑓 )⁄𝑐𝑤 .

Constant workforce strategy: In the demand chase strategy, the production follows demand,
which results in frequent variations in production. Stable production is desirable for multiple
reasons. In the constant workforce strategy, a constant production level is maintained, which
is 400 per month for the example. It leads to inventory holding and shortages and related
costs, which were zero in the demand chase strategy. We may need to hire/fire a few workers
at the beginning to reach the desired workforce level. In the example, opening workforce is
same as the desired workforce of 200. So, no hiring/firing is needed. Closing inventory (and
shortage) of different months are shown in the table below.

Month Opening May Jun Jul Aug Sep Oct

Forecast -- 300 400 500 450 400 350
Production -- 400 400 400 400 400 400
Inventory 0 100 100 0 −50 −50 0

Let 𝑐𝑖 and 𝑐𝑠 denote the inventory holding and shortage costs per unit-month. Then the cost
of the constant workforce strategy is 200𝑐𝑖 + 100𝑐𝑠 . We can compare costs of these opposite
strategies and pick the better one. However, there are innumerable other strategies which are
a mix of these two extremes and there is no simple heuristic to find one.

Mixed strategy and linear programming: Linear programming, the most popular Operations
Research technique, can help identify the best strategy for the aggregate planning problem. A
mixed strategy is defined by the following decision variables.

𝐻𝑡 – Number of workers hired at the start of month-𝑡

DMS201 Notes prepared by Dr. Avijit Khanra

𝐹𝑡 – Number of workers fired at the start of month-𝑡
𝑊𝑡 – Workforce level in month-𝑡
𝑊̅𝑡 – Idle workers in month-𝑡
𝑄𝑡 – Production quantity in month-𝑡
𝐼𝑡 – Inventory holding at the end of month-𝑡
𝑆𝑡 – Shortage at the end of month-𝑡

Apart from the cost parameters 𝑐ℎ , 𝑐𝑓 , 𝑐𝑤 , 𝑐𝑖 , 𝑐𝑠 introduced earlier, let 𝐼0 denote the opening
inventory, 𝑆0 denote the opening shortages, and 𝑊0 denote the opening workforce level. Let 𝑙
denote the labour-month requirement of a product. Let 𝑑𝑡 , 𝑑2 , … , 𝑑𝑇 , where 𝑇 is the length of
the planning horizon, denote the demand forecasts. We assume that the closing inventory or
shortage shall be zero after fulfilling the demands. Then the aggregate planning problem can
be represented as a linear programming problem, as stated below. We will learn to solve the
linear programming problem towards the end of the course.

Min ∑𝑇𝑡=1(𝑐ℎ 𝐻𝑡 + 𝑐𝑓 𝐹𝑡 + 𝑐𝑤 𝑊 ̅𝑡 + 𝑐𝑖 𝐼𝑡 + 𝑐𝑠 𝑆𝑡 )
such that (1) 𝑊𝑡 = 𝑊𝑡−1 + 𝐻𝑡 − 𝐹𝑡 for 𝑡 = 1,2, … , 𝑇 (worker balance)
(2) 𝑊𝑡 − 𝑊 ̅𝑡 = 𝑙𝑄𝑡 for 𝑡 = 1,2, … , 𝑇 (worker-production relation)
(3) (𝐼𝑡 − 𝑆𝑡 ) = (𝐼𝑡−1 − 𝑆𝑡−1 ) + 𝑄𝑡 − 𝑑𝑡 for 𝑡 = 1,2, … , 𝑇 (material balance)
(4) 𝐼𝑇 = 𝑆𝑇 = 0 (closing inventory/shortage)
̅𝑡 , 𝑄𝑡 , 𝐼𝑡 , 𝑆𝑡 ≥ 0 for 𝑡 = 1,2, … , 𝑇 (variable sign)
and 𝐻𝑡 , 𝐹𝑡 , 𝑊𝑡 , 𝑊

Rolling horizon planning: Aggregate planning tells the hiring/firing requirements for next 𝑇
month. In April, we solve the aggregate planning problem for May-October and prepare for
hiring/firing of May so that it starts with the desired workforce. In May, we solve the problem
for June-November and prepare for hiring/firing of June. This process of repeated planning
with overlapping horizons is known as the rolling horizon planning. It helps adapting to
changing demand forecasts. A longer planning horizon, i.e., greater 𝑇, smoothens production
process, but forecasting errors increase with 𝑇. An appropriate 𝑇 is chosen balancing these
conflicting factors. Also, the gap between the current time and the first period of the planning
horizon, which we took as one month in the example, shall be just sufficient to complete the
hiring/ firing for the first period.

Homework: Sometimes, it is a good idea to employ overtime production to accommodate

minor increase in demand than hiring more workers. Let 𝑐𝑜 denote the over-time production
cost per unit. Let 𝑄̅𝑡 denote the over-time production quantity in month-𝑡. Modify the linear
programming formulation to allow overtime production.

Further reading: Chapter 3, Production and Operations Analysis by Nahmias and Olsen.

DMS201 Notes prepared by Dr. Avijit Khanra