Since 2016, the golden age of chatbots, improvements in natural language processing (NLP) technologies and the refinement of use cases have given virtual agents renewed arguments in favor of mass adoption by businesses and users.
Indeed, the automation of customer service through chatbots, also known as intelligent conversational agents, is becoming the norm for companies seeking to digitalize their customer experience.
2020, described by Forbes as the year of renewal for conversational agents, became the year of mass adoption of these technologies.
In this context, we share the keys to understanding what makes chatbots more effective and, above all, more profitable.
Chatbot & Customer Service: R.O.I. is King!
First, thinking that a conversational interface could suit any use case was a mistake. The proliferation of POCs launched by large corporations and startups led to massive disillusionment around the performance of an emerging technology. Equally, this was followed by a complete questioning of the usefulness of such tools.
Four years later, we observe that virtual agents have genuine value when used to:
- automate the handling of recurring questions from users (customers / employees / customer service agents),
- qualify / route a support request to a specific department,
- schedule a support request with a human agent,
- resolve issues related to logging in to a customer portal,
- help a user navigate their customer portal,
- measure user satisfaction in a specific context,
- capture / qualify / enrich a commercial opportunity,
- collect information in place of a form.
All these use cases share 3 common points:
- they are easily automatable,
- they require brief and focused exchanges,
- they are not subject to any constraint in their formulation.
That said, it is pointless to think that chatbots are the solution to all your problems. Exchanges that are too complex or too in-depth require strong and dynamic contextualization. For the time being, they are poorly suited to the use of a conversational interface.
As a reminder, this is precisely what led “early adopters” toward widespread disillusionment.
Chatbots That Are Finally Intelligent!
There are two major families of chatbots for customer service:
- chatbots capable of constrained exchanges, often modeled as static decision trees,
- chatbots capable of freer and more humanized exchanges, using automated natural language processing technologies.
Use cases related to customer service (self-care) have propelled intelligent virtual agents into the spotlight. However, the technologies used (NLU or NLP) are highly demanding in terms of structured and labeled data. Indeed, this poses a problem, particularly when training models on the desired area of expertise. This requirement generates uncertainty about the performance of the virtual agent during the first months of operation.
At The Chatbot Factory, we have been working on this challenge for several years in order to offer a conversational intelligence that is as “plug-and-play” as possible. The goal is to guarantee bot performance from the deployment phase onward, while minimizing training, optimization, and language model enrichment operations.
One of the major differentiating factors of our technology lies in our ability to pit multiple language analysis and processing models against each other, classified into two main families:
- approach based on machine learning algorithms (NLU),
- approach based on semantic analysis linked to ontologies and dictionaries.
Our Tolk platform integrates a proprietary technology that selects the most appropriate model based on performance and context parameters.
This technological innovation improves bot performance by enabling the interpretation of up to 70% of requests related to a user intent, without any initial training.
As a result, our clients can deploy their conversational agents in under a week, compared to 3 months previously, and achieve strong performance without delay.
Ensuring Success!
Business teams often set themselves clear and quantifiable objectives:
- improve the quality of customer service, in particular by making it accessible 24/7 and immediate,
- reduce the cost of handling a customer interaction.
And the reality?
In reality, the measured benefits are somewhat out of step with these expectations. The observation is that each channel creates its own audience and that channel cannibalization proves to be a rather long and tedious process.
This rests on a change in user habits. It is the benefit around the value obtained from the new tool or service that motivates this evolution.
The goal is therefore to design a chatbot that is not a copy of your FAQ, but a service designed around users’ problems.
Start from the assumption that your knowledge bases and FAQs address only a small portion (approximately 40% to 50%) of your customers’ actual requests. It is limited, but that is the reality.
It is therefore essential to accept launching your virtual agent without covering the entire intended scope. To address this uncertainty, it is imperative to mobilize business teams around the enrichment and creation of new knowledge bases.
Return on investment objectives can only be achieved by adopting an agile approach. It is often preferable to shorten the design and deployment time of the agent in order to quickly confront actual user requests. Entering rapid iteration cycles to focus on real customer needs and deliver relevant, precise, and satisfying responses is key.
People play a central role in the bot’s learning process. They use listening and monitoring tools designed to accelerate the virtual agent’s learning and increase its overall performance.
Finally, measuring satisfaction and listening for weak signals in the conversation are good effectiveness indicators for assessing the usefulness of a virtual agent.
In Summary…
Customer Service Chatbots are effective when they are well designed and well positioned within your customer journey. Do not expect too much from them; be pragmatic and agile.
The benefits linked to using a chatbot can be rapidly observed:
- promotion of the self-care approach,
- immediate reduction in the number of emails, medium-term reduction in the number of calls,
- measurable economic efficiency (10x cheaper than a human agent)
- scalable.
All that remains is having access to an easy-to-deploy technology, tools that identify automation levers, and performance indicators aimed at increasing the bot’s performance.
