The use of artificial intelligence (AI) and machine learning is on the rise, but it’s important to know which reputation management processes should and shouldn’t be automated.
AI, using natural language processing (NLP) models, allows computers to understand and decipher what a human is saying. This is starting to be used by companies in South Africa responding to customers through online channels.
It also being employed increasingly in the reputation marketing sector.
The main reasons organisations turn to these technologies is that they dramatically improve efficiency, can reduce errors and save time. They are also used to produce exceptionally detailed and useful business performance analysis.
Businesses with online listings and reviews can automate much of the reporting and customer interaction that these platforms provide.
Customers are warming to the idea of artificial intelligence because it provides a way to get quick responses.
Review responses can be semi-automated through NLP and AI but it is important to always have a human eye to ensure 100% accuracy and personalisation. It’s important that there is some customisable wording in the response to the customer, but much of the information can be automated.
It’s also important to have strategic keywords and elements in your review responses and software can ensure these elements are dynamically added, making the responses feel personalised and relevant to the rating and review.
Some interactions should never be automated: A good example is responses to negative comments, reviews and complaints. If a customer had a bad experience at a business and has left an online review or comment, there should always be human interaction and response to that. It shouldn’t be left to a machine or bot to handle or, worse, be left in a void or ignored.
In order for automation to work, it has to be adapted to specific business sectors, behaviour, regions and languages.
Google and Amazon have some of the best pre-configured AI models but these have to be modified and trained to make them do what we want them to do.
For instance, machine learning models basically read text. But in South Africa, we have 11 official languages and a lot of local slang, as well as brand-related words. We have to teach the models South African languages and idioms and integrate them into what we have built.
One of the models is the just-launched Google Bert, which Google is calling its best-ever AI model.
Bert provides context to searches by using certain contextual words and natural language that it used to ignore, and the results are now much more accurate.
My team and I have refined and trained it further using our own data compiled over the years for South African relevance and slang. This data can then be used to create insights that can be acted upon – such as tracking sales and sentiment, fine-tuning operations and improving efficiencies.
Companies should automate reporting because this is repetitive, time-consuming and open to human error.
Reports can be produced on anything from impressions, clicks and calls to reviews, review scores, social media posts and bookings stats. Generally, anything that’s quantifiable and that’s available to analyse, can be aggregated, counted and automated.
Any measurable digital marketing stats are generally available through third-party APIs (application programming interfaces). An API is a link between two sites, for example between ours and Facebook, Google or TripAdvisor’s – which allows you to pass information back and forth.
These systems automatically build reports, so that brands don’t have to spend time managing and putting this together as that part of their strategy is automated.
Using the example of a restaurant, in the past, every review had to be categorised by a person who would check if a review was about food or service, for instance.
Now, using AI, we can understand exactly what a review or comment is about and we can streamline the process of responding. This way companies can ensure their investment in software is increasing internal ROI and efficiencies, while time spent by employees or suppliers can be used more strategically.
The human response
Humans are still vital to customer interaction, however, not only to deal with the negative sentiment but to train the models and communicate insights which have been compiled by AI.
AI can show you stats of your strengths and weaknesses, but only when this is combined with a human input will these stats truly become insights and provide value that can be acted upon.