In the long-term,
there’s no doubt that AI will be used as a key part of handling customer interactions in most businesses, but the question is: how can we avoid the pitfalls that can come with the implementation of a poorly-understood, heavily hyped technology?
The use of AI should
be focused on use cases where the AI
does a better job than a human, whether
that’s being quicker, more accurate, improving first contact resolution
, being available 24/7, or able to see patterns in data
that no person could find.
AI is going to be easy to get wrong – elevated
expectations, limited reference sites, murky data, the risk of customer
pushback…the list goes on.
With that in mind, here’re 5 tips on
successful AI implementations:
1) AI is not a silver bullet.
Expectations of what
the AI implementation can actually achieve must be closely managed. There may
be the expectation from senior management that headcount will immediately begin
to drop, but in the majority of instances this is not why AI is being
implemented. Focusing on a tightly-defined use case would reduce the risk of
implementation delays and expecting too much, too soon from AI. However it is
important not to see even a relatively modest implementation of AI as being a
point solution, rather than a single step in a long-term strategy
2) If it isn't broken, don't fix it.
There are areas of
customer interaction where AI cannot come close to matching a human agent.
Machines simply are incapable of feeling empathy, and even sophisticated
sentiment detection at its best comes close to what an ordinary human being can
do naturally. Use cases for AI should be focused upon areas where there is a
gap in functionality, rather than trying to replace something that isn’t broken
3) AI skills are few and far between.
AI in the contact centre
is relatively new, and with it being so popular, there is a shortage of skills,
support and resource within the industry as a whole. In-house technology
departments are less likely to have capability, expertise and experience,
meaning that the risk of suboptimal deployment and the requirement for
third-party assistance may be higher than with other more traditional IT
4) Clean your data.
Businesses' data assets
must be in place before implementation of AI, as this is a technology that
relies upon having large, clean pools of data that it can be trained on and
learn from. Without this in place, it will be virtually impossible for any AI
implementation to get close to its potential. The preparation of data will
involve having an organized, non-siloed data architecture, a consistent data
vocabulary, the means of accessing this data securely and quickly, and the
ability to access other pieces of relevant information (e.g. customer-related
metadata) in order to include greater context. Without this, it will be difficult
for a machine learning process to train itself effectively, or for a chatbot to
be able to use all of the relevant data in order to reach a correct conclusion
5) Don't trap customers.
Always have a
well-designed and clear path out of the AI-enabled service and onto a human agent.
Trapping a frustrated customer in a self-service session runs the risk not only
of training them not to use self-service again, but also poisons the well for
other companies using AI. This is what happened in the early days of email
support – customers would try to communicate with one or two businesses via
email, and when they didn’t receive a response for days (or ever), they decided
that the whole email support channel was unworthy of their time. It took many
years to change this perception and to get them to trust the channel again.
The eventual overall
roadmap of the contact centre industry will lead to significant levels of AI
involvement in customer contact, but in the foreseeable future this is likely
to improve self-service and assisting agents, rather than having seismic
effects on headcount. An AI implementation whose success is to be measured mainly
by the reduction in HR resource is unlikely to do well.