Automatic Optimisation of temporary staff in
a limited environment - some simple facts.
This is a non mathematical view of recruitment software
for temporary staff which seeks to optimise cost for the end
user.
Optimising people is not the same as optimising machines.
This was written by Ian Pettman who has a degree in Physics from
Oxford University. It was written after a number of lengthy
discussions with the following:
Dr Peter Kelen of Power Optimisation. Power Optimisation
specialises in providing routines which cycle power stations down
time and work schedule for major UK generating companies for the
lowest cost of generating the electricity.
https://www.powerop.co.uk
Dr David Nelson who's Doctoral thesis was on optimisation of
Nurses within a Hospital Trust in New Zealand. https://researchspace.auckland.ac.nz/bitstream/2292/332/9/02whole.pdf
Dr Barry Stoker who is heavily involved in extensive analytical,
modelling, business and research consultancy including employment
optimisation for large commercial groups.
https://www.jigsaw-consultants.co.uk/aboutus.html
Overview
There are three sections to this document: Mathematical factors
and Human factors and Other factors. In spite of its name,
Mathematical factors section is very non mathematic in its content
being largely descriptive of the issues: The maths required is 5x4
level and we even give the answer! Please give it a read,
aspirin not a requirement.
Mathematical Factors
There are some straight forward rules when considering how
difficult it is to perform optimisation. Optimisation of any sort
rapidly becomes a complex issue when the number of items optimised
grows. It is really just simple multiplication.
When it comes to choosing non interchangeable people or objects
then if I am choosing one of five I have five choices. If I am
choosing two of five then I have five choices for the first item
but only four (remaining) choices for the second. The total is
choices are 5x4 or 20. If I am choosing from ten items then 10 for
the first choice and nine for the second. In this case I have a
total of 10x9 or 90 choices. Only twice the number items to choose
but over four times the range of choices. It rapidly gets worse as
the numbers increase. In the case of an average day when 200 temp
staff are booked spread over 7 or 8 categories with around a
quarter HCA's (50) then the total possibilities are simply
huge! 50x59x48x47x...7x6x5. In fact as numbers go this
is a big one by anyone's standards.
To calculate every possibility will not take forever, it will
take many lifetimes. Some drastic short cuts are needed for a
computer program to make even a simple first choice.
Basically we need to guess what might be a good choice to start
with. If we don't try a short cut the computer will not finish on
the first day's shifts till after we are dead and gone. (It is
actually a lot longer than that)! Of course a staff Bank
administrator faced with such a task simply makes what they
consider a good first choice for the first vacancy, fills it then
goes on to the next. The Nurse Bank administrator can actually play
an Ace card when they get to the end. They can phone someone up who
has previously said they were unavailable and persuade them to
work, thus not only filling shifts more quickly, but filling more
in the end.
Actually for the computer things are a lot worse when it comes
to optimising for cost in an NHS environment. Generally
optimisation programs work by making a first guess, then swapping a
couple of people or objects around. If things get better- good
keep. If are not better, swap back and try entirely different pair
of choices. This works well when each object is slightly
different. Skip the following if you think this might be
reasonable.
(Example: say a bunch of sticks all of different lengths being
put in storage boxes of different sizes maximum 1 stick per box.
The problem: have the fewest sticks over which won't fit in any
box. We have an empty box but the stick we have over is too long
for the box we have empty. We have a long box with a short slick in
it. Solution: swap sticks: put the short stick in the short box and
the long stick in the long box. Keep doing this until we have the
smallest number of sticks over and when we do a swap thing don't
change (all the longest sticks are over). Now we think we have the
best solution because we can see things getting better. However in
the case of nurses being scheduled for lowest cost, HCA1 is on the
same pay as HCA2. At first sight this does not make a huge
difference. However, when we swap HCA1 and HCA2 there is no change
in cost. The optimisation program knows things aren't changing and
because this is what it tests for to find out if it has arrived at
a good solution, it thinks it has arrived at the best solution.
Wrong! This is known as the (double) valley problem. It is
especially severe when optimising for overall cost when large
numbers have the same individual cost.)
To sum up: if there are lots of identical values (people on the
same grade) optimisation is hard.
Human Factors
Because by definition scheduling programs are inhuman they will
lack the touch of a good staff Bank administrator. In the medium
and long term, in the nature of things, the automatic program will
generate higher degrees of disaffection than good Human interaction
and leadership. The inevitable consequence of this will be lower
levels of availability. Because of this the efficiency of staff
optimisation need to be measured over an extended period and not
just an initial blip of implementation. Also results need to be
returned for ALL implementations.
We have all heard the one about "the emperor has no clothes". So
by definition software that costs £100,000 has to be better than
software that costs less than £10,000? The only problem is that if
you analyse the level of success against the cost of the project
across all IT projects then there is an extremely high correlation
between the cost of a project and the chances of it NOT producing
the desired results. However, in very simple terms it is a lot
easier for decision makers to decide to write a large cheque and go
home satisfied that they have set in motion a big project, rather
than to tune a small project for improved results even though
evolution may be the most certain route. Unfortunately this is just
human nature.
By definition sales choose the most successful implementations
as their references. Unfortunately when talking about statistical
improvements, this is the same as flipping a coin and choosing the
best sequence of heads and tails to prove their software makes
heads come up 75% of the time.
When writing a cheque for £50,000, expensive software must
absolutely guarantee a saving of double this to cover the high cost
of ownership. Otherwise you might as well spin a coin to see if you
will save money.
So is there a case for optimisation software? Surprisingly
considering all that has gone before, the answer is a clear yes.
There are a number of household names that have implemented such
projects with significant payback. However when you analyse such
scenarios you invariable find that the starting point was a chaotic
"system" with multiple large departments operating their own
individual policies, some not worried about overall cost: just
making sure bodies were available so target were met; others
exactly the opposite: minimising personnel costs but missing
deadlines.
Other Factors
People are not machines. One of the "methods" of proving
optimisation is cost effective is to take an historical set of data
and then optimise it and show a net saving. Unfortunately this is
not a real world scenario for two reasons. One: requests come in
over extended periods. Optimise once (to give people reasonable
notice) and all later requests will be ignored limiting the range
and effectiveness of the process. Inevitability a staff bank will
resource less costly staff first. Then (later scheduled) expensive
staff (where savings can be made) will not be optimised. Ok so
optimise twice before shift patterns. Now the very nature of the
process is that staff will be moved around, some cancelled (at
short notice): others booked (who now at short notice may not be
able to fill and have to then be replaced by more expensive agency
staff). This will inevitably be an unfeeling process and lead
to dissatisfaction. Compared to electricity power generators where
optimisation can be a rolling process because machines do not get
in a huff, people are by nature homeostatic: they like a degree of
organisation.
Summary
What is achievable for a reasonably efficient temporary staff bank
office employing between 2 and 30 consultants to allocate and
manage staff temporary staff and a budget of £50,000 to £100,000 pa
for software?
Ava can save you £40,000 to £90,000 guaranteed.