9

Learning Curves
This page intentionally left blank
 9
Learning Curves

LEARNING OUTCOME
! Apply learning curves to estimate time and cost for new products and services.

9.1 Introduction
In this chapter, learning curves will be explained together with its application to
decision-making.

9.2 The learning curve

9.2.1 Introduction
In practice, it is often found that the resources required to produce given amounts of a prod-
uct decline as output accumulates. It costs more to produce the first unit of a product than it
does to produce the one thousandth unit. Various factors contribute to this relating to labour,
material and overhead costs. One significant element in this is the learning curve effect.
The learning curve relates to the observed tendency for workers to become more adept
at a task the more times they perform it.

9.2.2 The nature of the learning curve

According to the Official Terminology:

Learning curve: The mathematical expression of the phenomenon that, when

complex and labour-intensive procedures are repeated, unit labour times tend
to decrease at a constant rate. The learning curve models mathematically this reduc-
tion in unit production time.

239
240 STUDY MATERIAL P2

LEARNING CURVES The recognition of the so-called learning curve phenomenon stems from the experience
of aircraft manufacturers, such as Boeing, during the Second World War. They observed
that the time taken to assemble an individual aircraft declined as the number of aircraft
assembled increased: as workers gained experience of the process, their proficiency, and
hence speed of working, increased. The learning’ gained on the assembly of one plane
was translated into the faster assembly of the next. The actual time taken by the assembly
workers was monitored, and it was discovered that the rate at which the learning took
place was not random, but rather was predictable.
It was found that the cumulative average time per unit decreased by a fixed percentage
each time the cumulative production doubled. In the aircraft industry, the percentage by
which the cumulative average time per unit declined was typically 80 per cent. For other
industries, other rates may be appropriate. Further, the unit of measurement may more
sensibly be taken as a batch of product, rather than as an individual unit. This does not, of
course, affect the underlying principle.
Let us take as an example a learning rate of 90 per cent:
● If the first batch of a product is produced in 100 hours:
● The cumulative average time taken to produce two batches (a doubling of the cumulative
production) would be 90 hours.
● This gives a total production time of 2 " 90 ! 180 hours.
● The actual time taken to produce the second batch (the batch being the unit of measure
in this case) would thus be 80 hours, the cumulative total time taken to produce two
batches – 180 hours – less the time taken to produce the first batch – 100 hours.
As a doubling of cumulative production is required, in order to gain the benefits of
learning in the form of reduced average labour hours per unit of cumulative production,
the effects of the learning rate on labour time will become much less significant as produc-
tion increases. Table 9.1 shows this.
The graphs in Figure 9.1 show the data from Table 9.1 plotted both on an ordinary and
on a double logarithmic scale. It will be noted that the double logarithmic form is linear;
the ordinary form is not.
The learning percentage is usually somewhere between 70 per cent and 85 per cent. The
more complicated the product the steeper the learning curve will be. For example, research
on Japanese motorbikes found that the small 50cc bikes had a learning curve of 88 per
cent, the 50–125cc bikes, one of 80 per cent and the 250cc bikes, one of 76 per cent.

Table 9.1 Cumulative average time learning rate: 0.90

Cumulative average Cumulative total

Batches time per batch (hours) (hours)
1 100.00 100
2 90.00 180
4 81.00 324
8 72.90 583
16 65.61 1,050
32 59.05 1,890
64 53.14 3,401
128 47.83 6,122
256 43.05 11,021
512 38.74 19,835
 PERFORMANCE MANAGEMENT 241

LEARNING CURVES
Cum. 100
average
time per 90
batch, hours
80

Yx 70

60

50
X
12 4 8 16 32
Batches (or units)

Log of cum.
average
time per
batch, hours

Log of no. of batches (or units)

Figure 9.1 Cumulative data graphs

Table 9.2 An 80 per cent learning curve

No. of Average time Total time Additional Additional Time per

units per unit in hours units time unit
1 10 10
2 8 (10 " 80%) 16 1 6 hours 6 hours
4 6.4 (8 " 80%) 25.6 2 9.6 4.8 (6 " 80%)
8 5.12 (6.4 " 80%) 40.96 4 15.36 3.8 (4.8 " 80%)

Table 9.2 illustrates how an 80 per cent learning curve works, as units are doubled from
1 to 2, to 4, to 8.
In constructing Tables 9.1 and 9.2, it was assumed that we already knew the learning rate
that applied to this particular situation. However, it must be appreciated that, in the real
world, this rate can only be established by observation. Records must be kept of the number
of units/batches produced and the associated time taken, in order to construct the equivalent
of Table 9.1 (although it is likely that fewer observations would actually be taken). It is then
the job of the engineer or accountant to deduce the learning rate from these observations,
242 STUDY MATERIAL P2

LEARNING CURVES which will require the specification of an equation to fit the data. For example, the observa-
tions in Table 9.1 are plotted in Fig. 9.1, and can be described by the equation:

Learning curve formula:

Y ! axb where Y ! cumulative average time taken per unit

a ! time taken for the first unit
x ! total number of units
b ! index of learning

the log of learning log 0.8 (for 80%) #0.0969

where b ! i.e. ! !# 0.32
log of 2 log 2 0.3010

Thus, in Table 9.2, the average time taken per unit, for 8 units, is given by:
Y ! 10 " 8#0.32
! 5.14

which, allowing for rounding, is the same average time that was derived in Table 9.2. Note
that the log is easily found by using the log button on your calculator. You are not expected
to have a working knowledge of logs for the exam.

Exercise
Derive the learning curve formula for a 90 per cent learning rate, and apply it to confirm
the cumulative average time per unit for 64 units, as shown in Table 9.1.

Solution
Y ! ax b
log 0.9
b, the index of learning ! ! #0.152*
log 2
*This is calculated using a scientific calculator, but you can use log tables if you pre-
fer. Inserting the relevant data into the formula, we get:
Average time taken per unit ! 100 " 64#0.152
! 53.14
This agrees with Table 9.1 above.
The learning curve equation is not normally used to check earlier calculations in the way
that we have just done, but is rather used to assess the time that will be required for an output
level that does not represent a doubling of the cumulative production total, and thus cannot be
determined by simply creating a table such as the one used earlier. For example, let us assume
that the manufacturer above has the opportunity to bid for a contract to produce ten batches
 PERFORMANCE MANAGEMENT 243

of his product, and wishes to estimate the time it will take to complete the contract, in order

LEARNING CURVES
to help set the tender price. If the cumulative total production of his product to date is 32
batches, the learning curve equation can be employed to calculate the cumulative average time
per batch to produce 42 batches: the 32 already produced, plus the ten under consideration:
Yx ! ax b
Yx ! 100 " 42#0.152
Yx ! 100 " 0.5666
Yx ! 56.66

that is, the average time per batch to produce 42 batches is 56.66 hours, giving a total pro-
duction time of 56.66 " 42 ! 2,380 hours.
Inspection of Table 9.1 above reveals that the average time per batch to produce 32
batches is 59.05 hours, with a total production time of 1,890 hours. The total time taken
to produce the ten batches under consideration will thus be 2,380 – 1,890 ! 490 hours,
that is, an average time of 49 hours for each of the batches 33–42. This may be compared
with the average of (3,401 – 1,890)/32 ! 47.20 hours indicated by Table 9.1 for the next
doubling of a full 32 batches from 33 to 64. Obviously, if the current level of production
does not lie on a table such as Table 9.1, that particular average time, and the correspond-
ing cumulative hours to date, will need to be calculated from the same formula.

9.2.3 Uses of the learning curve

(i) In circumstances where the learning curve is likely to operate, that is, in complex
assembly operations, knowledge of the rate of learning can help in price setting.
(ii) When setting budgets, the effects of the learning curve should be taken into account.
Standard costs should reflect the point that has been reached on the learning curve,
so that those set on the basis of the ‘steady state’ (i.e. that point at which learning has
become insignificant) will be lower than those set during the learning period. This is
necessary if variances arising during the learning period are to be meaningful.
(iii) The learning curve has been found to be particularly useful in determining the likely
costs to be incurred in fulfilling government contracts. This provides a rational basis
for price negotiation and cost control.
A number of points about learning curves must be stressed:
(i) Learning curves chart the reduction in time per unit as experience is gained; they do
not measure a reduction in cost per se. However, if workers are paid by the hour and
hourly wages are constant, the labour cost per unit will decline as a result of the learn-
ing curve phenomenon.
(ii) In addition to direct labour costs, only those costs that are directly related to direct
labour time, that is, any overheads that vary directly with those hours, can be expected
to decline as a result of the learning curve.
(iii) Learning is likely to be greatest in complex assembly environments, of which aircraft
assembly is a prime example. If labour is working in a machine-paced environment,
there is no significant opportunity to alter the rate of working, and thus the learn-
ing phenomenon in terms of direct labour time cannot exist. However, the experience
that plant managers gain in scheduling work in such an environment may result in a
reduction in the direct labour hours required for a given level of production.
244 STUDY MATERIAL P2

LEARNING CURVES
9.2.4 Deriving the learning rate using logs
You must be able to use logs to derive the learning rate from basic data:

Example
Data relating to the production of the first 12 units of a new product are as follows:
Time taken to produce the first unit 15 hours
Cumulative time taken to produce first 12 units 81 hours

What is the learning rate?

Solution
Formula:

Y ! axb
81
Y ! average time per unit/batch ! ! 6 . 75 hours
12

a ! time for 1st unit/batch ! 15 hours

x ! total number of units/batch ! 12

log of the learning rate

b ! index of learning !
log 2
To find b:

6 . 75 ! 15 " 12b

6 . 75
12b ! ! 0 . 45
15

The challenge now is to solve the equation 12b ! 0.45 and to do that we need to use logs. As stated earlier,
this simply requires an ability to use the log button on your calculator and apply some basic rules.
You have to log both sides of the equation – but note that the power (here b) changes and becomes a
multiple:

12b ! 0.45 becomes

b log 12 ! log 0.45
Using the normal rules of equations:

log 0.45
b!
log 12

Use your calculator now to find each of the log values. Do not ignore any minus signs:

#0 .3468
b! ! #0 .3213
1 .0792

Now we know that b ! index of learning ! log of the learning rate

log 2

And log 2 is simply found using a calculator ! 0.3010

log of the learning rate

So, #0.3213 !
0.3010

Rearranging gives
Log of the learning rate ! 0.3010 " #0.3213 ! #0.0967
 PERFORMANCE MANAGEMENT 245

The final stage is to find the learning rate itself and this is done by using the antilog or 10x button (you may

LEARNING CURVES
well find it is the log button used with the shift key although in some models it is a different button entirely). Don’t
forget the minus when you enter the number.
The figure #0.0967 converts to 80% – the learning rate!

9.2.5 Learned behaviour

Learning is often assumed to be automatic and just volume related, but it is not that sim-
ple. Learning is not a matter of law, it is achieved by focusing attention on, and striving
for, improved performance. Two organisations with identical staff and resources would not
necessarily achieve the same amount of learning, as management’s attitude and the organi-
sation’s culture create learning.
Why do all employees achieve the same level of performance? Why does all learn-
ing stop at the same point? It seems it is all to do with the expectations of the company.
Cyert and March (1963) said that ‘the behaviour of an organisation is created through rou-
tines and routines stabilise an individual’s behaviour making all workers’ behaviour similar
and routines are developed and maintained through interpretation of the past’. By routines
Cyert and March meant habits, that is paperwork, weekly meetings, rules, roles who does
what, etc. Levitt and March (1988) argued that organisations learn by ‘encoding inferences
from history into routines that guide behaviour’. So routines, roles, habits, conventions, etc.,
have beliefs embedded in them. In this way routines can be made sufficiently independent
of individuals and survive personnel turnover.

9.3 Summary
In this chapter we have learned about learning curves and their implication for
decision-making.