This doesn’t mean reinventing the wheel each time, but adopting an analyst-led approach that arrives at the best marriage of tools and techniques for any business problem.
We do this because it makes sense
It stands to reason that you will get better results if an analyst has heard your business problem, built the text analytics models with the aim of answering that problem, and had a hand in presenting and interpreting the results.
This analyst-led approach ensures a focus on your problems, as well as richer interpretation of the results simply through the analyst’s familiarity with the contents of the comments. The result is a more efficient set up and better insights.
We also do this because it helps to manage expectations and mitigate risks
Text analytics is not a perfect solution. It should, however, be good enough to base business decisions on – and ultimately investment.
With this in mind, analysts are able to highlight problem areas, suggest solutions and work to create results that everyone can trust. Analysts are also able to provide a human quality check, ensuring systematic errors are avoided, and any other issues are minimised, giving you the confidence to really use the results.
In comparison, press-button solutions – although not without value for the instant findings that they deliver – ultimately risk misleading their users through a lack of transparency about how and what comments have been classified in each category.
We can pick the right combination of analytical techniques to meet your needs
Most text analytics projects involve a combination of fully automated and analyst-driven techniques. The balance between these techniques will depend on the nature of your question, which will lead to a focus on one or the other – and potentially lead the analyst to pick one tool above another (in this event, Ipsos has a small portfolio of tools available for text analytics).
1. The fully automated approaches focus around data exploration. Concepts and themes are identified automatically by the tool and the relationship between them is represented graphically in maps. This helps to create a wider picture of what the customer has been saying, and build up a greater depth of understanding of the customer experience.
The lack of human intervention at this stage provides an unbiased view of what the comments are saying.
2. The analyst-led approaches provide quantification of topics and sentiment, in many cases building on what has been identified during exploration (with checks to make sure that the concepts are sensible and meaningful).
During quantification the analyst builds the structure, ensuring that the structure and categories make sense to your business and provide actionable insights through a combination of the right level of granularity and sensible category creation (e.g. rude staff rather than simply poor service).
We suggest additional techniques that extend beyond text analytics
Sometimes quantification is not enough and, in order to build full action plans, further techniques are required. These may involve, for example, text-based drivers analysis to help prioritisation, correlations to understand the relationship between themes, or cluster analysis to group the categories into something more manageable – either for tracking purposes or to enable users to focus on key ‘pillars’ (i.e. groups of categories) rather than single categories alone.
We also involve the analyst at reporting
Reporting and dissemination of the results are key to get right for action to be taken when any analytics or research are undertaken.
Whether you opt for ad-hoc Voice of the Customer (VoC) reports, tracking results or online reporting, a sound understanding of the techniques involved in the analytics and the comments that have gone into each category is required for accurate, insightful results. Involvement of the analyst at this stage ensures both of these things, making the results more reliable and richer in terms of their interpretation.
Long live the analyst!
Text analytics is a fantastic tool for delivering insight over high volumes of text data, and multiple sources. And its automated brain allows it to do this in a quick and scalable manner. However, it still needs a human heart to deliver meaning and interpretation to the results.