Cost allocation model for shared services

August 10, 2007


Earlier this year we conducted a review of the cost allocation model used to charge our local operating companies for support centre services like helpdesk, IT procurement, server management, etc. This post outlines the model we came up with and draws key principles for any agreement of this type.

The key is to keep everyone focused on costs, not on cost allocations. (Cost allocation just shifts costs around.)

When allocating costs from a shared service, the key aims are:

  1. Provide customers with transparency and control over cost drivers.
  2. Provide flexibility over the way resources can be used, while keeping a single consistent allocation model.
  3. Leave choice over resource allocation and daily control with the service provider.

The costs for a shared service can be divided into 2 components:

  1. Infrastructure
  2. People

Infrastructure costs

Infrastructure costs should be completely separated from overheads and people costs. Examples of infrastructure include data transfer, rack space charges, outsourced server monitoring, etc.

Each infrastructure item has a total cost, which must be divided among the customers according to an allocation model that best represents the cost driver.

Example: Allocation of Infrastructure Costs

Data transfer into the data center for July cost $100. The allocation model for this infrastructure item is bytes transferred by each customer company. Foo Industries generated 75% of the traffic during July, while Bar Incorporated generated the remaining 25%. As such, the data transfer bill for Foo is $75 and Bar is $25.

Racks for housing servers in the data center are depreciated at a rate of $50 per month. Costs are distributed based on the number of servers used by each company. Foo has 10 servers in place, while Bar has 15. As such, Foo’s rack charges is $20 for July while Bar pays $30.

Server monitoring is compulsory for data center servers and is charged at $200/server/month. This is billed directly to each company based on their servers in place so Foo pays $2000 and Bar pays $3000.

In the end, we have this equation to give the operating company cost for each infrastructure item:

ItemCostToCustomer = TotalCostOfItem x (CustomerUsage/TotalUsage)

People costs

People in a shared service spend their time on 3 things: 1. Project work 2. Ad-hoc tasks, maintenance and incident management 3. Managing other people

Each person working in a shared service has a specific cost. This will typically include:

Time spent on project work and ad-hoc tasks can be directly allocated to customers, but time spent on people management is harder to quantify. To solve this problem, we calculate the cost of time spent on people management and allocate it among all reports under the manager.

Example: Allocation of people costs

Alice is the manager of the shared service and spends 100% of her time managing people. She does no direct project work and does not complete any ad-hoc tasks. Her cost, including salary, building and equipment costs, is $100.

Alice has 5 direct reports, each of whom have 4 reports, giving a total of 25 staff in her team. Alice’s cost is divided evenly among all 25 reports, adding $4 to the cost of each person.

Alice has no time to allocate among customer companies (she’s done no “real work” afterall). But, her cost is effectively distributed by the work completed for customers since it is allocated to staff who do “real work”.

Bob reports to Alice. His cost, including salary, building and equipment was $80. With the management allocation from Alice, his cost is now $84.

Bob spends 50% of his time on people management, 25% on projects and 25% resolving ad-hoc issues. Per the model, 50% of Bob’s total cost of $84 is evenly distributed among his 4 reports ($42/4 = $10.50 each). The 25% project work ($21) and 25% ad-hoc work completed by Bill are billed to the customers directly.

Chris reports to Bob and spends all his time on ad-hoc tasks. His cost, including salary, building and equipment was $60. With the management allocation from Alice and Bob, his cost is now $60 + $4 + $10.50 = $74.50.

Of the ad-hoc tasks performed by Chris, 50% were done for Foo Industries and 50% were done for Bar Incorporated. As such, Chris’s cost to Foo Industries is $37.25 and to Bar Incorporated is $37.25.

In summary, the allocation of people costs follows these principles:

  1. All people costs are allocated and paid on an individual person basis. So, a company that uses 25% of Derek’s time will pay 25% of Derek’s cost. This is not the same as using 25% of total time spent by the shared service team and paying 25% of their total cost. For example, if we use resolved calls as the metric to determine ad-hoc time spent and include both L1 (average 300 calls) and L2 (average 50 calls) engineers in the cost calculation there is no potential reward for moving calls from L2 resolution to L1 resolution.
  2. Time spent on people management (an rough estimate for each manager) is added to the cost of the people being managed. So, you are only charged for actual work being done but we recognise that part of the cost of using those resources is the management team in place.
  3. Time spent on project work is directly allocated and billed to the customer requesting the project. It’s important to appropriately separate these tasks from ad-hoc time. This ensures that we can see the real cost of project activities and keeps ad-hoc tasks reasonably consistent in complexity (thus evenly cost distributed).
  4. Time spent on ad-hoc tasks is assumed to be the remainder after calculating time spent on people management and time spent on projects.

It’s only a model

As always, this model is an approximation of reality. It will never be perfect, nor should we aim for it to be perfect. It’s important to remain pragmatic and remember that a lot of the small inconsistencies and errors will correct themselves. (Two slightly wrongs can make a right in this context.)

You’ll need to think about how to handle events like extended sick leave or annual leave.

The key is to remember to keep all cost drivers transparent and controllable. Try not to let generic buckets like “overheads” or “maintenance” creep into the model.


After 6 months use on a team of 25 people divided among 5 operating companies over 2 countries, we’ve found this to be a simple, flexible model that has given us unprecedented insight and high level of control over cost drivers.

We’re now dealing with the hard (and important) problem of seeking real process improvement and cost control rather than looking for temporary advantage by playing with cost allocations to get temporary local advantage.

Comment by Anonymous on November 01, 2007 5:24 AM
How do you address,measure,and allocate increases in server usage and storage the shared server environment? For instance, in the event that one client's needs exceeds the available capacity on an existing server but only needs an additional 15% of an additional server do you pay for and assume the cost of the server? Is there a % or assumption in the planning stages that allos for or creates a bufer zone for growth?
Comment by Nathan @ e-gineer on November 01, 2007 10:05 AM
For idle server capacity there are two choices, either the 3rd party provider builds a cost structure that includes growth capacity (e.g. Amazon EC2) or the exact costs are dynamically shared among the users. We use each of these models in different cases.

When dynamically sharing costs we calculate the total % capacity used and allocate the total cost on that basis. For example, there is 300% available capacity through 3 servers costing a total of $300/month. Company A uses 75% of a server while company B uses 150%. So, company A pays 33% of the total cost 75%/(75%+150%) which is $100. Company B pays $200.

This may mean that company A has a cost increase due to actions taken by company B. This is less predictable than a fixed cost model.

Unfortunately, in the end, someone must pay for spare capacity. The key thing is to find a model where actions to reduce my own cost (in this case reducing capacity) are the same actions that we need to take to reduce the total cost. That way companies are discussing cost, not cost allocation.
Comment by Anonymous on January 15, 2008 9:40 AM
I like your model but was wondering how you treat un-allocated capacity that occus from time to time, i.e. what if an existing FTE is not being fully utilised in a particular month on ad-hoc or project work? Where does that piece of his/her total cost get allocated? Is it a fair rule to simply allocate out any residual costs evenly across businesses?
Comment by Nathan @ e-gineer on January 15, 2008 10:25 PM
There are two cases of time to be paid which has not been productive, idle time as you suggest but also periods of extended leave.

We've not seen idle time as a problem, as that gets absorbed into the "issues time block". So, those issues are particularly expensive that month, but at least the costs are split by the companies usage (even if high).

With extended leave periods, we typically split the cost based on recent allocations of that resource. So, if you normally use them a lot you will pay the majority of their cost during the unavailable period.

Of course, in the end hopefully these are the minority of your allocation costs so an even split may be a simple & fast way to achieve a similar result. If idle resource costs are a large part of your allocation, then there are bigger problems at hand than the allocation of costs ;).
Comment by Anonymous on January 16, 2008 1:01 PM
Thanks for the clarification Nathan. I agree that if you have large unallocated porsions of time - something is wrong. Where I work we're surrounded by accountants, so the focus is more on transparency of every minute cost basis for our businesses, so even small portions of time are relevant.
Comment by Anonymous on February 17, 2009 5:09 AM
I am in the process of developing a cost allocation policy for my I.T department. Currently all computer and server costs were paid by I.T. We now plan to allocate costs to user departments in an eqitable manner. My questions can i measure cost allocation by some metric or index of sorts. This is to show that we have made some improvement in allocating costs to some industry standard or something.
Comment by Nathan @ e-gineer on February 24, 2009 3:35 PM
Benchmarking not the costs themselves, but the model and process of cost allocation...

I've not seen any information about this and imagine its an area ripe for best practice rather than benchmark comparisons as it's all about making costs transparent without adding to them just to track funny internal money.

Other than that I guess you could benchmark against billing services that providers (e.g. telco's) provide to customers. Is your bill timely, clear, etc?

Would be interested if anyone has any links or experience to share on this...
Comments are closed.