Analytics – Is It About Sample or Size?

When generating a sample on which to run speech analytics in a contact centre, there is the idea that the larger the sample, the more accurate the results would be. However, this is not always the case.

The size of the sample is important for sure, but equally critical is the quality of the sample. In other words – does it accurately reflect all the variables in the organisation?

Who Is Alf Landon?

Few people know that name, yet it is associated with one of the biggest polling failures in American history.

In the 1936 election, Alf Landon was the Republican candidate challenging Franklin D. Roosevelt’s re-election for president. As it had done successfully for five previous elections, the Literary Digest had begun conducting a poll to predict the outcome. This was done by sending survey forms to more than 10 million voters, of which they received 2.4 million back. The results from this substantially large sample predicted that Landon would win by a majority.

However, as the history books reflect, the Literary Digest was wrong and Roosevelt won with 61% of the popular vote. Interestingly enough, there was another researcher, George Gallup, who also conducted a survey with a much smaller sample size. He was able to accurately predict the outcome of the 1936 elections within 5%.

The Literary Digest had polled their readers, as well as a directory of automobile owners and telephone users. This sample, however, was of a higher income bracket and not representative of the average American voter.

The reason Gallup succeeded where the Literary Digest failed was due to the quality of the sample, despite it being much smaller. It was more representative of the voting population.

The 1936 election poll had significant impacts. For one, it proved to be the demise of the Literary Digest; but more importantly, it was the catalyst to developing more refined public polling techniques.

Gallup’s American Institute of Public Opinion received national recognition following the accuracy of its predictions and this opened the way for a more scientific approach to sampling.

Today Gallup as an organisation is a respected world leader in analytics and research.

Scientific Sampling Approach

Companies like Ember take a scientific approach to sampling, which helps clients get a representative view of what is happening in their organisation, without having to analyse every single call.

Here’s an example of Ember’s analytics at work:

A large client was running a sample that represented just 18% of their total call volume through speech analytics. However, when the sample was viewed as a split of agent skills, it was an indication that the sample may not be an accurate representation of what was going on in the organisation.

Because the volume was concentrated in a few skills only, the majority of the calls were very similar. This meant that the sample provided only a one-dimensional view of what was really going on in the contact centre. The detail focused on the high-volume work and discounted rare call types.

Now some people may be of the opinion that the majority of the call volumes are more likely to hold the most value for the business and that the only way to include the rare call types would be to increase the sample size. However, the client had contracted to a set number of hours for the analytics, so that would not have been an option.

A more scientific approach was needed so that all the agent skills and call types could be accurately analysed. It was particularly important to include the rare call types because they potentially represented high-value and high-impact calls.

In other words, these call types could potentially be the key to unlocking significant revenues.

By applying a rigorous scientific sampling strategy, the sample was balanced to reflect enough of each skill without having to increase the sample size. The client had the assurance that no more processing in terms of volume or time would be required, because the sample size hadn’t changed. The strategy had just adapted the sample so that a more meaningful analysis could be conducted.

The results were more accurate and representative of the opportunities that existed for the organisation. They had an overview not only of what they were doing, but also where they could exploit potential new opportunities.

The moral of the story is that when it comes to analytics samples, more is not necessarily better. Instead, a more scientific approach aimed at generating a sample that is a true cross-section of the business, irrespective of volume, will provide more accurate results.

When the purpose of analytics is to generate information on which strategic business decisions are based, then a strategic and scientific approach is needed to classify the right type of samples.