Abstract
The most effective method for capacity planning within any organisation is to use Demand Driven Capacity Planning. This enables infrastructure capacity to accurately reflect expected demand for the services supplied, and is as proactive a method as is possible in most circumstances. There are several dependencies to using Demand Driven Planning that require careful consideration before this method is adopted, some of which may seem insurmountable in many organisations but in reality they are not. Key benefits of this approach are the ability to respond to customer requirements faster, thereby improving revenue recognition, and more accurate financial forecasting of expenditure. This paper is intended to help key decision-makers examine the benefits of demand-driven planning prior to adoption.

Background
Determining the future capacity required for networks, systems or any form of manufacturing use basically the same techniques . It is usually based on a mixture of proactive and reactive forecasting. Many organisations undertake capacity planning by trending forward from historical data, which we call reactive forecasting (it is reactive as it depends on events which have already happened). While this may help identify generic trends and organic growth over time it does not accurately reflect the following future changes to the infrastructure:

• New Customers, especially those introducing large business volumes
• New Services with no historical data to draw on
• Changes in architecture, design, technology or topology
• Changes in performance service level targets
• Non-cyclical changes, such as marketing campaigns
• Constraints, whether technical, business or financial

To transform capacity planning into an effective business support process requires proactive forecasting. This is where expected future demand is used to derive future capacity requirements, and we call Demand Driven Planning.

Reactive Forecasting & Planning
The most difficult aspects of capacity planning are, in order:

1. finding suitable business volumetric data to use
2. accurately translating business volumes into infrastructure requirements
3. accurate understanding of existing inventory
4. measurement of supply chain efficiency.

For these reasons it is often easier to use whatever historical infrastructure utilisation data is available and trend forward into the future. By trending forward it is possible to determine future infrastructure capacity requirements at any point in time. However there are two main faults with this method; firstly any forecast deviates from the actual result increasingly over time (Figure 1), and secondly using only historical data excludes known changes to demand and supply (Figure 2).

Figure 1 - Forecasting variance over time

Figure 2 - Historical vs. Forecast data

Forecasting errors will always occur so techniques have to be devised to reduce their impact, such as introducing a supply-chain management strategy to reduce vendor and internal provisioning lead-times . As can be seen from Figure 1, the deviation from the forecast can vary dramatically over time. This deviation is defined by a function f(t2/t1), which can be a linear or, more often, exponential function. The deviation from forecast at any point in time also has a statistical probability distribution. Also as unit volume increases the statistical probability is improved, hence when forecasting from a baseline of 1,000,000 units the percentage probability of accuracy is automatically better than when the baseline is 10 units, although the unit variance error may be the same. The use of simple trending mechanisms (such as linear regression) provides the ability to forecast simply future capacity requirements. However this we call reactive forecasting, as it only uses historical data and therefore can only react to past events, not future changes. The main advantage of using reactive forecasting is that reasonably accurate results can be determined with minimal effort. This reduces the cost of capacity planning, although increases the likelihood of costly errors. In a business where demand varies are at a predictable equilibrium rate, and no other major changes are occurring, it is possible to use reactive data (as per Figure 2) for infrastructure forecasting with a high degree of accuracy. However, any significant changes, such as large new customers, new products or services, marketing campaigns or changes in technology, will have a dramatic impact to the infrastructure requirements. For this reason Capacitas advocates the use of proactive data in an integrated modelling and capacity planning process.

Proactive Forecasting & Planning
To ensure future infrastructure requirements are fully captured in the planning process requires three key components (as demonstrated in Figure 3):

1. A forecast of product/services data over the chosen planning horizon
2. A model to translate products/services into components required
3. An accurate understanding of existing infrastructure inventory

The accuracy of the output of this process is a product of the accuracy of the forecast data, the translation process and the inventory data, so care must be taken to ensure each component is as accurate as possible. Each item is covered in a separate section below.

Figure 3 - The Three Key Components to Demand Driven Planning

Product Unit Forecast
To capacity plan effectively using demand driven planning requires an accurate demand forecast of each product and service that is supplied on the infrastructure being planned. So for instance a leased-line network service provider would require a forecast of each E1, E3, STM-1, STM-4, STM-16 etc., at each location (or preferably on each route) over the planning horizon with granularity of no less than three months. A suitable example is shown below in Table 1.

It should be noted that this data is provided into a suitable modelling application in the most detailed form possible. If this data was aggregated into a single measurement unit (e.g. STM-1) before passing into a suitable model it would lose value as the lower order units may have specific and unique impacts (such as electrical port requirements in his example).

This methodology applies similarly to any other form of capacity planning, from IT systems to Cans of Baked Beans on Supermarket shelves. Another example would be of a Retail Bank wishing to forecast how many millions of each type of transactions an IT system may have to process per month:

There are two common formats for these forecasts, either using cumulative or incremental figures. While capacity planning can be undertaken using either method, the most effective for an installed base system (such as a network or computer which grows to support product demand) is to use cumulative inputs - see Figure 4. If capacity planning of a throughput based system is required (such as stock required to build widgets) then incremental inputs are often more effective. The reasons behind the preference for cumulative input units are not immediately intuitive, as it would seem that incremental units should be as effective, especially as incremental output changes are the business deliverable of capacity planning (to calculate additional costs and when they are required). However the use of cumulative input figures in calculating an installed base system offers several advantages:

• Automated base-lining of inventory against demand every iterative calculation cycle preventing sequential cumulative errors from accumulating
• Cumulative system size is automatically calculated:
• enabling identification of sizing constraints and cost break-points (e.g. equipment purchases)
• enabling calculation of installed base charges (e.g. software licence costs)
• enabling calculation of other operating expenditure (OPEX - e.g. Staff)
• Reduced data integrity problems due to data version control issues

Product unit forecasts are often the most difficult data for a capacity planning department to get hold of, as this data is held by the Sales and Marketing department. Political issues surround the retrieval of this data, but it is essential to planning any business effectively and, despite usual claims, it almost certainly exists somewhere within all organisations for Marketing and Finance planning.

Figure 4 - Comparison of incremental versus cumulative input units

Capacity Planning Model
This paper is not intended to cover the many modelling packages on the market as only the principle of modelling is discussed here, not the practicalities which vary for each problem domain. What is vital is that the modelling package has flexibility to allow the modelling data to be input and altered easily, as needs dictate, reflecting potential changes over the planning horizon. This data is the model’s representation of the real world (for the domain being modelled), specifically the relationship between input and output data.

Example of Model and Data:
A simple example is a computer program used to calculate how a solid such as a metal may expand with increasing temperature. The "modelling package" is the code that takes input data, calculates the output from the internally held understanding of the relationship and the presents the output. The "modelling data" is the data that identifies the relationship between temperature and the volume.

The modelling data needs to accurately represent the relationship of inputs to outputs within the system being modelled, and is usually expressed in terms of a set of rules. As with the product unit forecast data described above the model should accurately describe the system down to the required level of output granularity. So when modelling a network the granularity may be simply circuit capacity between nodes, or could be as detailed as cards in routers for each node. Table 3 below indicates a typical model for three components within a network.

Inventory Data
Capacity planning (both reactive and proactive) also relies on accurate and complete inventory data. This can often be difficult to get hold of, especially in large, mature, distributed environments, such as a PTT. Asset Management, Purchasing and Change Control processes should be linked with the capacity planning function within an organisation to ensure a consistent view of the existing and future inventory. In reality this inventory data, if it exists at all, is usually spread across an organisation with small databases held by individuals in many teams. For this reason a centralised inventory management database system is a mandatory requirement for any organisation serious about responsible financial planning.

Business Benefits
Demand Driven Planning delivers many business benefits over the reactive forms of capacity planning, such as trending or threshold driven capacity management, due to the increased accuracy of this method. These break down into two main categories - financial and customer service.

Financial Benefits

• Accurate financial modelling over the planning horizon
• Ability to respond quickly to changes in the business environment
• Accurate capacity planning ensures expenditure not incurred before needed
• Accurate capacity planning reduces risk of stranded capital
• Faster time-to-market of new products and services
• Faster revenue recognition as capacity available when needed
• Reduced contingency budgets freeing capital
• Lower unused inventory levels providing improved capital efficiency (ROTA)

Customer Service Benefits

• Ability to provide vendors with long-term forecasts thereby reducing vendor lead-times
• Accurate capacity planning ensures capacity always available when needed reducing customer turn-up times
• Ability to deliver changes to service levels and new products quickly

The increased accuracy using Demand Driven Planning enables the pre-provisioning of long lead-time infrastructure with a higher level of confidence. This is the key differentiator over other reactive forms of planning.

Conclusion
Demand Driven Planning provides businesses with a powerful tool for planning infrastructure requirements in advance and hence improving customer service and financial planning. A process is used in manufacturing called Materials Requirements Planning (MRP), which is a component of successful Demand Driven Planning. The other aspects of successful Demand Driven Planning are accurate Demand Forecasting, Capacity Modelling and Inventory Management. Combined the effectiveness of these four components is multiplied to offer organisations a major competitive advantage over organisations using Reactive Planning processes. The key dependencies for effective Demand Driven Planning are:

• Accurate Forecasting
• Accurate Modelling
• Accurate Inventory