Planning vs. Analytics – the Data Difference – A word of warning

At this point it is important to note that an Operational Analytics cube, perhaps with 20 or more dimensions, is NOT a Planning cube – the two requirements are very different. I have worked with companies who tried to have their sales reps plan at a detailed operational level – for example for every SKU (20,000), for every customer, outlet, warehouse, region, brand, origin company, industry segment, time and more. This becomes a nightmare for the rep, and a potential source of risk as finding the right place to put data is virtually impossible.

One strategy that I have followed is to create a simple planning cube that is pre-populated with all the real combinations of planning data (I called this Planning Nodes – each planning node represents a single real combination of many other dimensions). That gives the rep the exact data points to put the plan data into at SKU, Planning Node and Time, simplifying the planning process immensely. Then let the CPM software allocate this into the detailed cubes for them. The more automation you can get into a CPM solution, the better. In Prophix, we use Detailed Planning Manager for exactly this purpose.

Finance Breakout – Wrestling Control from Accountants by Adding Value

So, one of my original challenges was to get Finance, who typically own the CPM solutions, to allow wider use of the solution. The most effective way to achieve this is to offer value, in the name of accuracy and detail, to the data that they are already gathering. Typically Finance only sees summarised operational data in the P&L. By offering the ability to drill into some of the key P&L lines (such as Sales) and get detailed Customer profitability, or a cube that can show them the difference between expected and actual ship dates by invoice by the customer can give Finance Directors and their staff huge insight into the accuracy of their planning process. Proactive outlier analysis is also possible – push monthly issues directly to Finance using automated reporting.

There is no need to directly connect the Operational Cubes to the Finance Cubes if you don’t want to – although some CPM tools will tell you that you can directly connect many cubes together there is seldom a good business case; it’s better to connect them by process (I.e. on demand refresh rather than live) to ensure performance for both Finance and Operations. Operations data is often far more rapidly changing than Finance data, or just required on a daily basis.

Finance vs IT – where does the responsibility sit?

I could also phrase this – “who should look after CPM systems, IT or the business?”. The challenge here is that IT has technically able staff who may have database and system experience, but are generally time poor and responsible for the whole company rather than an expert in any one part. Business departments are experts in their area but typically not systems experts, and also time poor.

Projects that I have seen to be most successful have combined a number of resources from the company:

1. Business Users – the business users need to have an on-going involvement to ensure their buy-in to the future system.
2. User Requirements – the business needs to control the scope; don’t be afraid to re-think current processes rather than re-building a new version if what you have already.
3. Project Management – even if the vendor project manages their implementation you need to project manage your involvement – you should be involved in any project to ensure that you can manage the solution going forward.
4. IT – IT need to buy-in to provide you with the infrastructure and support as the project goes on.

So realistically, either for the project or on a permanent basis, you’ll establish a multidisciplinary team or ‘centre of competency’ on a shared service model, supporting the solution. CPM solutions need to evolve with the changing company and this team will be able to do it.

Navigation of Highly Dimensional Data – End User Strategies

One of the challenges of highly dimensional data is the navigation for end users – being able to analyse your data by day, SKU, customer and another 10 dimensions is very powerful, but ultimately will end up with a lot empty cells, or zero data as the sparsity reaches into the billionths of percent, or less, of filled cells. Ultimately what you need to do is to control the user’s access to data, or to limit the data made available to each user. These might sound the same but there is subtle difference.

Assuming a single large data cube, adding a comprehensive security model based around groups will allow you to limit the data in that cube that is available to end users. Typically a group’s based security model is ‘additive’ (I.e. if a user is a member of more than one group then they get the access from both groups combined). This enables you to set up groups for access to products, customers, regions, versions, departments, or anything else, and then give the users multiple groups to define their access. As users need different access, they can just be added to/removed from groups and you can even have a group for read/only read/write to quickly grant or remove data access. In the end, you can have as complex or simple security model as you like.

The second method involves creating sub-cubes from the master cube, which may have many fewer dimensions, and much fewer data. The advantage of making smaller cubes can be performance, simplicity, and flexibility. Within products such as Prophix creating a small cube with a subset of data is a simple task for business users so that cubes can be built, removed, on an ad-hoc basis. Some larger retail models have daily cubes, weekly cubes, period cubes for stock and sales information based on the different needs of their users. Again, the actual implementation will depend on your needs.

So, in summary, do it! Navigate the politics, understand what you’re trying to achieve and design your models to deliver the data for the users in an easy to use and performant way.