Congratulations, you’ve decided to leverage an AI platform for faster, accurate, and more consistent analysis of your customer feedback data.
Manual data tagging and reporting are a good starting strategy for customer insights. However, as you’ve likely recently discovered, as the volume of customer conversations grows, your team’s capacity and desire for manual work become limiting. Unscalable customer insight is often a bottleneck that works against CX improvements.
Recent advancements in AI and NLP are an effective answer to this problem and there has been a lot of hype around their use in the customer experience insights space. Unfortunately, that’s also led to a lot of hyped-up claims around AI’s capabilities and what makes ‘our’ AI ‘great’ for uncovering insights—a difficult environment to pick a tool that works for you.
In this guide, we discuss what steps you must take to select the right AI for customer experience tool. When it comes to analysis and reporting of your customer conversation data on channels like email, chat, phone calls, surveys, and social media, there are 7 questions to need to ask before you make an investment.
7 questions you need to ask before you press the BUY button on AI for customer experience
1. Are the insights granular and actionable?
The main reason an organisation invests in a data analytics and reporting platform is to be able to act on the insights. But one of the biggest complaints with most analytics platforms is that the insights are not actionable enough. It’s therefore critical that you are confident about the quality of insights before you make that buying decision.
What makes insights ‘quality’?
Remember these three words when evaluating the quality of insights, and you will never go wrong: Hierarchical, Granular and Contextual (HGC).
Look at the below example, which one provides you with the information you need to actually make a decision?
At scale, the analysis on the left has two main issues:
- It’s not hierarchical or granular: it shows that data tag ‘missing ingredient’ but fails to provide deeper insight like which ingredient is missing. It’s not enough to know there’s a problem, you need to know what problem you have (granularity) and what level it is. In an ideal AI product, you’d be able to kick off a project to tackle ‘missing ingredient’ complaints and then understand what is driving it—for that you need at least two levels (hierarchy).
- It’s not contextual: Not only does it miss multiple topics mentioned by the customer, reducing the richness of your analysis, but it lacks context. You need to know the ‘who, what, where and when’ of the drivers of customer complaints.
When you have a handful of conversations, you can apply these points manually. But, when they’re in the thousands per day, you need to be able to filter by location, topic, sentiment, and any other aspect you need.
Expert tip: If you see the phrase ‘word cloud’, run!
2. Does the AI work on your data?
The real test of an AI tool is not how many patents it holds, but if it can work on your data—not just on the training data used to teach it.
If you expect the tool to uncover granular, contextual and actionable insights, it needs to understand your data and business domain. A human who is an expert in the insurance domain may not be able to provide a similar level of expertise and insight in the travel domain, and you can expect an AI tool to do that either. An AI tool trained on travel customer data may not work well on your customer’s data—your data is unique.
Any AI software claiming to be ‘out of the box’ needs to be questioned. In the majority of cases, the machine learning models need to be customised if it’s going to provide accurate, granular insights. A lot of vendors rely on generic machine learning models to uncover insights from your data. Although this saves a lot of setup time for the vendor, it's not customer-centric and the insights are likely to lack granularity and accuracy.
In most cases, with generic models, the output is words and phrases which then need to be turned into a taxonomy at your end. This defies the purpose of using smart technologies like NLP and machine learning if you need to do all the hard work of creating a taxonomy.
The only way to ensure that the AI tool works on your data is to test it with your data. Ask the vendor for a free trial where you can assess the quality of insights.
3. Is it easy to setup?
Pre or post-covid, one thing that hasn’t changed is the availability of your IT team. There are multiple teams always competing for their time. You, therefore, don’t want to invest in a new tool which needs significant effort from your IT team for integration, deployment and maintenance, or you’ll have to wait another year to start using it.
One-click integration: The simplest and fastest way to integrate and share your data is if there is a native integration. Check your help desk’s marketplace to ensure their integration is listed, they may not so ensure that you check with the vendor if they have native integration plugins for the platforms you use.
API Support: Any CIO should not approve a new platform if it doesn’t have API support. It not only makes any integration easy but enables easy sharing of insights across different systems and teams within your organisation. If a Zendesk chat analytics integration has API support, not only can you get better understanding of the reasons for customer contacts but you will also be able to leverage those automated tags to facilitate native ticket triage, priortitisation and reporting.
4. Can it analyse every ‘conversation’ channel?
To gain useful customer insight, you need to analyse every customer interaction with your brand. If an AI for customer experience doesn’t work across all channels, like email, chat, NPS and CSAT surveys, reviews, social media, and voice, you’ll have potentially damaging blindspots.
Investing in multiple tools for multiple channels often leads to low tech adoption, disparate KPIs, siloed insight and an unclear understanding of the customer’s journey.
Many tools claim to be multichannel, but you need to take that with a pinch of salt. Automated analysis of NPS surveys using Natural Language Processing (NLP) is an entirely different ball game to customer support ticket tagging. The data held within support tickets is more unstructured, rich and complex in support tickets, which makes them so insightful, but that much harder to analyze.
Things become more complex still when including every voice call to get a 360-degree view of every customer support conversation. Traditional speech analytics fails to go beyond basic sentiment analysis and alerts on predefined words and phrases.
The only way to find out if a tool can analyse different types of unstructured data accurately and consistently is by testing it with your data channels. Ensure you trial the software with a minimum of two data sources, one being a customer support channel like email and other being a customer feedback channel like NPS.
5. Is it simple to use?
Improving customer experience is a collaborative effort. Because a customer’s journey weaves in and out of different business departments, it requires stakeholders from across the company to be on the same page and working together.
While it’s easy to buy new, shiny AI for customer experience analytics dashboard, it’s a significant challenge to promote companywide adoption of the tool. It, therefore, must be intuitive, need little user training, and be tailored and accessible for teams of all technical ability.
You don’t want to buy a technically complex software which requires hiring dedicated data analysts to understand and report the insights. This distribution bottleneck defies the very purpose of investing in the latest CX platforms—which should facilitate a rapid speed-to-insight for every teams.
6. Is reporting flexible for analysts and senior management?
Senior executives are interested in understanding core customer metrics. Whether that’s NPS, CSAT or ticket volume, customer KPIs are integral to high-level decision-making.
Any AI for customer experience software needs to be flexible to the different needs of these senior executives, as well as more operational team members. One user needs an overview, the other needs actionable insights.
When buying an AI tool, ensure that it supports the simple distribution of insights for consumption by different levels of stakeholder. Some mediums which facilitate easy distribution of insights are daily emails with top-level metrics, alerts to anomalies in the data, and easy weekly/monthly management reports. Product’s should also be designed with a high level view and a granular, root cause view, for ease of navigation.
7. Is your data secure?
Safety and security of personal data should be top priority for any business. It’s not only a legal requirement, but any compromise with a customer’s personal data is a PR storm waiting to happen. You must be comfortable with the data privacy and security measures in place for the shortlisted vendors.
GDPR compliance, data encryption, security testing and certifications are all a good starting point. You may also want to pay attention to information governance policy, incident response procedures, business continuity and disaster recovery plans.
As your legal teams will tell you, it’s better to be safe than sorry.