Call Mining: Insights and Lessons Learned

Recently, my colleague Matthew Berk posted about the marketing benefits of being able to move beyond call transcription to a more meaningful level of customer insight provided by a deeper and scalable understanding of advertiser – consumer conversations.

Today we are announcing the availability of Call Mining, a technology-based speech analytics product designed to capture the most useful call data for advertisers that we are currently rolling out in beta to Marchex Call Analytics customers. I’ve been deeply involved with developing the product and on-boarding customers, and I wanted to share some trends and early insights that Call Mining is showing us.

The first conversation with an advertiser that leads them to become a customer of Marchex Call Analytics often starts like this: “I know that my search and display campaigns are generating phone calls. I just don’t know how many.” These conversations often beg deeper questions, such as “Where are my customers calling from?” or “How many of these phone calls are longer than a couple of minutes?” Marchex Call Analytics effectively answers these questions for millions of conversations each month.

Over the past year, though, questions from advertisers have increasingly been about the contents of the actual conversation between callers and agents (the person answering the phone). “Are callers booking appointments? What features are they asking about? Why aren’t they buying?”

When we set out to build Call Mining, these were the types of questions we wanted to answer, and the product allows advertisers to understand the types of conversations that their customers are having with their business. Coincident with our announcement, I wanted to share a few insights and surprises that came up as we started to trial this product with customers.

Our product data suggests that longer calls aren’t necessarily better calls.
One of the questions we had as we were developing Call Mining was whether or not there would be a strong correlation between the duration of a phone call and the likelihood of a purchase or conversion. In other words, is knowing the average duration of a call “good enough”?

In most cases, the answer is “no”. Generally, when a call is “very short” or “shortish” it indicates that a conversion probably didn’t happen and the call can be chalked up to a wrong number, unqualified lead or a simple question about directions or store hours. However, “longish” or “very long” calls do not indicate a likeliness to convert. Let’s say that you’re calling an insurance company. You provide information about yourself, your driving record and your insurance needs. The price is quoted at the end of the call - which may be the single biggest factor in whether or not you decide to buy!

For most customers, it seems that a longer call means a reservation or appointment or purchase may have happened, but the duration of a call is not a reliable enough indicator upon which an advertiser can base ROAS calculations.

Calls Are Not the Same Across Media Channels.
If a consumer sees your advertisement in the newspaper, on their PC, or on their mobile device it can be for very different reasons. Take this customer, who received the following number of calls from two different online marketing sources.

Source #1: 281 Calls
Source #2: 202 Calls

This data allowed us to calculate an effective cost per call for each source. However, these calls also generate additional useful data for this customer. Of particular importance to the customer are the calls where the customer specifically referenced one of several different problems or objections:

Source #1: 91 Calls (32%)
Source #2: 18 Calls (9%)

Our conclusion? For source #1, a much greater percentage of the calls were from existing customers reporting a problem and much less important to a marketer held accountable for new customer sign-ups.

The Magic Word for Payment?
My favorite part of Call Mining is building topics of conversation to understand about a set of calls. Call Mining provides suggestions on topics to build, based on the set of calls and what we know generally about conversations, and each time I use the feature I discover something new.

For example, many phrases can indicate the potential of a purchase over the phone. “Credit card” and “American Express” are two good examples. If you’re evaluating car rentals or reservations at your B&B, “confirmation number” is an important phrase to look for.

In most cases, though, there is one word that I’ve found that is a virtual lock to indicate a telephone purchase. The word is “expiration”. In almost every conversation where someone discusses a credit card purchase, the agent asks the caller for the expiration date on his or her card. There is almost no other term or expression that refers to expiration, and there is almost no other context where people frequently use the term expiration in a conversation.

Further Insights.
In the process of building our Call Mining technology, I’ve arrived at the conclusion that conversations between your business and your customer provide the richest and deepest clues into your business, and I expect to be following up with further insights as more customers begin to extract the qualitative conversion data they are looking for. If phone calls are an important part of your business, and you’d like to know more about how your customers are interacting on the phone, please reach out to me at johnb(at)marchex(dot)com.