Implementing a conversational AI platform is often seen as a huge undertaking. We outline some of the risks involved and how to mitigate them to ensure successful virtual agent deployment
If you are like most executives in the banking, financial services, insurance and telecom sectors, you are probably under enormous pressure to transform your organization into a more automated and digital, yet customer-centric one.
Conversational AI is a great way of bringing a business’ customer service offerings into the 21st century, giving customers instant access to information, products and services without any of the traditional pain points such as lengthy wait times and discrepancies in accuracy due to human error.
Gartner reported that chatbots will become the primary customer service channel for roughly a quarter of organizations by 2027. And the trends in the industry appear to support this claim. Within our client base of 170+ companies and organizations, we are already seeing a massive shift towards the positive adoption of conversational AI by end-users. Norway’s largest bank reported a 51% drop in traffic to human support in the six months since launching its virtual agent, while 75% of another of our banking clients’ customers choose to get help from its virtual agent, even when presented with the option to talk to a human.
The potential for conversational AI to change the face of customer interactions is huge but, as with any new leap forward in technology, there is always a certain amount of hesitancy to jump in guns-blazing. The same Gartner market guide that is so bullish on AI being a key component of future customer service interactions, also states that it is expected that 40% of chatbot/virtual assistant applications launched in 2018 will have been abandoned come 2020.
This raises important questions about how the risks of deploying a conversational AI project can be mitigated, and what is involved in successfully bringing an AI-powered chatbot online that can:
- Answer any question with no waiting time
- Perform complex tasks on behalf of customers
- Forward sensitive requests to the right human employees when necessary
Identifying the risks
Time, money and scope are among the top concerns from businesses looking to roll out a conversational AI solution. Anyone with prior project delivery experience understands that these are typical risk parameters, but when dealing with a nascent technology such as conversational AI, the latter point is worth exploring further as it is pivotal to both the others.
In more traditional projects, the objective is usually clear from the outset, but in such a fast-changing and highly competitive industry, the challenge often lies in simply understanding and agreeing on exactly what you want to get out of your new digital employee. If you don’t start with a clear and agreed-upon scope, your project will likely miss its launch targets (and cost more money!) or you will end up not delivering a virtual agent at all, let alone one that is fit-for-purpose.
Without a well-defined scope you will quickly find yourself well behind the AI curve and catching up then becomes an increasingly lengthy and expensive endeavor.
Key factors to mitigating risks
Having standardized tools and processes in place is a good start to successful deployment, however, there are other factors that are equally important and unique to conversational AI. We’ve already touched on why defining a clear scope is crucial (we’ll get back to how to do this later), but the third key factor in mitigating risk is specifically how to work through the delivery process and build a competent, capable and enthusiastic team of AI trainers, who can improve and maintain the virtual agent long after it has gone into production.
Part of building up this AI trainer competence means equipping your staff with the tools and resources to work independently (or at least with minimal supervision) once a project has gone live. A key objective for a vendor should be to confidently leave a project at the end, ensuring that the AI training team is able to assess the quality of the virtual agent and make continuous improvements. This allows you to retain full ownership of the product without being reliant on the vendor to make the project a success.
Another powerful way to quash qualms around deployment can be to hold a 2-day accelerated POC workshop. We have found that, in circumstances where an organization may have had a less-than-favorable experience with conversational AI in the past, this can help debunk some of the common myths around AI while giving employees and key decision-makers the opportunity to get hands-on experience with the technology.
Coming back to scope, an accelerated POC is also the perfect opportunity to identify and prioritize use-cases to design an optimal implementation strategy for a virtual agent. You can look at it as a two-day micro-project in which you learn how to identify, solve and implement specific use-cases. This also has the added benefit, if the project is greenlit, that some of the initial work has already begun and therefore time to delivery is reduced.
The three phases of a virtual agent project
Although no two projects are ever the same, our previous experience of successfully launching over 170 virtual agents, we have established that a typical conversational AI project will pass through a roadmap with three distinct phases:
Phase 1 - Information (3-6 months)
Dependable information, assistance and guidance through chat in any language, at enterprise scale, available 24/7.
Phase 2 - Transaction (6-12 months)
Core business functions, such as orders, transactions and upgrades, can be completed by customers directly in the chat window.
Phase 3 - Transformation (+12 months)
Analyze, advise, sell and upsell products and services through direct messaging in voice and text based on customer unique data (that you own).
The timeframes mentioned above are not meant to be seen as the duration of each phase. They are, instead, a suggestion of when to begin implementing each phase based on your project’s maturity timeline. In many cases, it’s possible to start at the ‘Transaction’ phase, even if it’s for simple API integrations such as a password reset, for example.
Criteria for success
Depending on your specific business goals and risk profile, a conversational AI project should be flexible enough to help you meet the targets you set out to achieve. A typical project can take between 12-14 weeks, from setting KPIs and confirming project scope, to flipping the ‘go-live’ switch after proper QA and testing is complete.
While there is a huge amount of benefit to automating interactions, it’s not always (and in most cases, never) preferable to eliminate humans from the equation entirely. Instead, conversational AI should be looked at as a tool to help make existing human employees more effective, and one of the best strategies to mitigating risk is to integrate a virtual agent with a live chat service.
The cornerstones to creating a solid conversational AI foundation are:
- A robust intent hierarchy that allows a virtual agent to reliably answer inquiries and automate processes that would typically take time for human staff to complete
- Integration with experienced customer service reps via a live chat service, who are available to tackle more nuanced inquiries outside of a virtual agent’s defined scope
- Prominent placement of a virtual agent on your website - if your customers can’t find it, they can’t use it
- Proactive capabilities that allow a virtual agent to offer help (either to customers or behind-the-scenes to support agents) when it identifies a need
Additional criteria for ensuring successful deployment
Launching and maintaining a successful conversational AI solution requires more than just good planning and preparation. It is important to have plenty of tools at your disposal, not only to help ensure that you are getting the most out of your virtual agent, but also that you are able to keep step with the market as it continues to evolve alongside the needs of your customers.
Future-proofing your project is crucial and there are a number of ways that you can achieve this:
Built-in quality assurance tools
These allow your AI trainer teams to continually improve the system. It places control of the virtual agent’s output directly in their hands and ensures it remains consistent and on-brand.
KPI dashboard with qualitative analytics
Identify trends and convert your findings into actionable outcomes. Track conversations, user sentiment, most frequently-triggered intents, user feedback and much more.
Community and ownership
A dedicated e-learning platform means that support staff are certified to update, expand and train the virtual agent. Best-in-class online courses upskill non-technical staff to AI Trainers, becoming part of a burgeoning global community.