Andrew White at Contexta360 explains that how we build services, technology and solutions has radically changed over just a very short period of time.
Much like the automotive industry in the late 1800s and early 1900s, the product was built from scratch. Nowadays, cars are built using a mix of industry components and off-the-shelf capabilities or micro-parts.
It is a similar journey for the raw coding world. In the beginning, we coded from scratch, and some still do. Increasingly, we use abstraction software, drag-and-drop builders, and the auto equivalent of sub-components that do a specific task within the whole product.
This methodology directly impacts the customer interaction analytics world, specifically in the voice and chat analytics sector, and is having a marked impact on ease of deployment and cost of deployment.
In order to build a conversational analytics strategy, we first need to understand what it is. Conversational analytics is the convergence of speech analytics, chat analytics and wider text analytics across any medium (voice, video, app, mail or good old-fashioned letter). Ultimately, it is about:
- the voice of the customer
- what the customer is calling/texting/chatting about
- why the customer is calling/texting/chatting
- customer intent
- customer actions
- customer questions
- context of the conversation
- friction within the conversation
- emotion within the conversation
In short, it is the digital synthesis of a single indexed conversation, or trends and patterns across millions of indexed conversations.
Interaction analytics is very similar and typically focuses on the transactional and meta-data interactions.
Ultimately, it is about:
- when the customer called/entered a chat
- what the customer bought/inquired/requested
- where the customer clicked
- what options the customer chose
- the manual survey results
- journey paths
- security steps taken
- automation sessions (IVR, IVA, chatbot)
- channel hops made (for example chat to voice)
- IP/SIP packet data
When we blend conversational and interaction analytics, that is where the real magic starts.
But wait, we are still talking about tech. Let us move on to strategy as tech, which is the final part of the process.
What is it we are trying to achieve? The mission is not to be an expert in deploying conversational analytics, this is a tech capability that is needed to drive the original strategy.
So here I would like to assemble the framework of building the strategy. This includes:
- The business results we strategically need
- The business drivers
- The influencing factors of the business drivers
- The components of the influencing factors
- How you resolve the influencing factors
But actually, it is quite a simple model, so let us turn the theory into practice.
- I want to grow revenue by 20 per cent or I want to reduce costs by 10 per cent. (I’ll choose revenue)
- The strategic drivers may be geo-expansion/product expansion, increase customer acquisition/customer retention and upsell. (I’ll choose increase retention and upsell)
- This could be new products, new pricing, new packaging, new promotion. But these are typically nailed down, so service would be a good driver. We need to ensure we have the right knowledge, service availability, soft skills, CX and C-SAT indicators to score our performance unemotionally and completely.
- In order to do this we need data from a) conversational analytics and b) interaction analytics. This includes what was bought, when it was bought, the key topics in all conversations, the intent, action, sentiment, emotion, question – and what insights we extract.
- Finally, we get to the point of deciding how we get this data. In short, code from scratch, look at ready-made components (NLU and AI components off the shelf) or invest in a fully integrated solution.
Another dynamic to consider in your strategy is who in your business needs this insight? There is no fixed answer here, but increasingly we see two centres of value, namely:
- The data-science teams
- The operations teams that interact with customers.
Historically, everything mentioned above was the preserve of the data-science teams, and the technology was highly customised, hand-integrated and manually “crunched”.
A few months ago I met a big-brand potential client. It was interesting to see just how immature their capabilities were. Their process, put simply, was:
- A business unit (BU) would request a report along the lines of “can I get the conversational analytics for product group X, in call centre team Y between Jan 1 and March 31 with indicators for CX, customer effort, and C-Sat (please)?
- The data-science team would then:
- Pull the recordings
- Match this data to agent attendance data for the team in question
- Match this data to agent attendance data for the product lines in question
- Run the conversations through a speech-to-text tool
- Run the transcriptions through their own natural language understanding application
- Link this to transactional data and interaction data
- Produce a report (in circa two weeks)
- This would be received by the original BU owner who would say “thank you but this is not what I was looking for”
This is probably a bit extreme, but it highlights the fact that our insights need to be way closer to the BU owners.
There is still a very real place for data-science teams to do more and more advanced analytics, but the example above can be executed in precisely 30 seconds in a modern conversational and interaction analytics solution.
Additionally, data-science teams should not be too concerned about building from scratch. We have amazing supervised and unsupervised topic models off the shelf, including QM, C-SAT and customer effort score models and our own hyper-accurate speech engine that is tuned to your lexicons.