Strategy is always the meeting point of need and capability. The starting point in a text analytics strategy is the identification of business processes where text analytics might deliver benefit, and an awareness of what is possible. Business processes that are candidates obviously need access to relevant text data. At the current time this usually applies to customer and marketing applications, although as text analytics becomes more prevalent so businesses will collect more text data to enable the analytics process. As far as capability is concerned it is usually a given that provided good quality text data is available, so value will be created through the analytics process, although this is nearly always an iterative process.
Once candidate business processes have been identified it becomes a matter of fleshing out some form of cost/benefit analysis. With analytics technologies it becomes much more difficult to estimate benefits unless data are available from other organizations, or suppliers who have experience in the domain. Typically, an increase in lift (value created with new information divided by value created without information, multiplied by 100) of just a few percent will more than adequately reward many text and data mining activities. The cost of developing and deploying text analytics applications depends very much on the route taken. Text analytics may be part of a much larger ‘big data’ project, in which case it becomes more difficult to allocate costs. Costing discrete projects is usually much easier.
This does not mean however that determining costs will be necessarily straightforward. Unlike traditional process automation projects (e.g. ERP, CRM) where deployment is essentially linear, analytics projects are usually iterative in nature. Again it is very useful to have access to people who have knowledge of building and deploying analytics processes, although the nature of text data will be particular to each organization and this inevitably introduces some variability. The cost components will include hardware and software (unless a cloud service is used), skill sets, domain expert involvement, performance management and monitoring and possibly the acquisition of new data. This latter point is more important than is immediately obvious. Sourcing external data (social data, market data etc.) is an obvious cost, but it is more than likely that the results of analysis will imply that greater data capture when dealing with customers for example might deliver more accurate insights and predictive models. There is a cost associated with this and it needs to be taken into account.
Finally there is a cost associated with the management and monitoring of processes involving analytics. The insights and models derived from analysis usually decay with time, simply because markets and customers change. Monitoring performance and feeding this back into the analytics process is not a trivial matter and will impose its own overhead.
While we have been conservative by suggesting just a few percent increase in lift, it does happen that the benefits can be considerably greater than this. A more useful model for modelling the return from an investment is Expected Return. This allows for multiple scenarios and will generate an expected return from an investment. While this is not widely used at the current time other than in some specific industries (petrochemicals for example), it does give a good feel for risk and is more appropriate for analytics projects where there are more unknowns.
Analytics projects do need a somewhat different approach to risk management than traditional IT systems. It really is not enough to develop a model and leave business users to get on with it, the whole process needs much finer integration between business, IT and data analysts.
Text Analytics: a business guide
A report for business and technology managers wishing to understand the impact of rapidly evolving text analytics capabilities, and their application in business.
The Business Value of Text Analytics
What is Text Analytics?
Text Analytics Methods
Unstructured Meets Structured Data