Tech and tactics for the future contact centre today

By Cheryl Allebrand, CGI UK, Managing Consultant, specialising in Conversational AI & Automation

Running a contact centre that meets customer and market demands without breaking your business case isn’t an overnight endeavour, but it is doable, and it is worth doing.  The positive impact on your customer and employee experience will make the effort worthwhile and will pay dividends in terms of customer advocacy and brand reputation, not to mention all the time and cost savings that come with automation.

With all the hype about ChatGPT and Large Language Model (LLM) disruption, you’ll be excused for wondering whether it’s worth changing now or where to start.  Despite daily announcements about coming benefits and risks of the newest AI, the bulk of the basic techniques for building a data-enabled contact centre still apply.

Here are four steps to help you figure out what tech mix will best serve your organisation’s goals, where LLMs can play a role, and where to start first.

Step 1: Snapshot your current state

If you don’t know your starting point, you won’t understand your big challenges in enough depth to strategise fitting solutions.

• Query your data

Today’s Speech-to-Text (STT) capabilities open a whole new realm of possibilities when it comes to analysing old call recordings. Instead of sampling only a fraction of calls manually, you can now gain full insight into what customers most frequently ask for, along with how they ask for it and the corresponding agent responses as well. 

Developments in tech, including the right application of LLMs, mean you can now automate the bulk of categorisation and labelling work, while knowing the words and phrasing your customers use means you can train your NLP for better recognition. LLMs are great at figuring out alternate phrasings, so you can further improve your recognition rates.

• Understand the issues

You might already have some ideas about what your biggest challenges are but check if your new data backs this up and quantify it. 

• What are your goals?

Solving challenges is a great starting point, but what other goals do you have as an organisation and what will be your North Star when it comes to prioritising action? 

Once you understand what you want to achieve, you can select the technology mix that will deliver that experience to your customers and employees.

• How will you measure success?

Pick your metrics first and establish your baseline so you can check back regularly to manage progress. 

Step 2: Tackle first things first

It may seem counterintuitive, but your real starting point should be informed by solid data and discussions. Yes, laying the groundwork is necessary, but now is when real work begins.

• Automate the easy

While it may feel tempting to take on everything at once and speed up your success, that’s not the way to get the most out of automation. Find a narrow use case – preferably one that’s high volume or high value – and automate that work. This may mean building a chatbot to handle it, using RPA for a process, or a combination of the two (along with OCR or image recognition where it makes sense). Do that well, refine, use the learnings, and apply them to the next use case.

LLMs can be applied to auto-generate flows that build FAQ bots trained on your data. Remember to mitigate the risk of hallucinations in post-processing and keep a human in the approval loop. You won’t want to relinquish control of your responses and take on risk of public relations and regulatory disasters.

• Authenticate

Breaking the rule about not solutioning before understanding the problem, if identity and verification (ID&V) is part of the bulk of your customer interactions, you should be offloading that work from your agents. Cumulatively this will produce huge time savings and quite frankly technology handles this for your customers faster, easier and more securely. 

• Banish IVR

If you’re still forcing customers to sit through your list of instructions and guess at your tree structure, stop it now. Trying to make them classify their issues based on your organisational structure is an annoyance for them. On top of that you’re incurring losses due to double work when they find their way to the wrong department. Intelligent Routing helps eliminate double work and minimise customer frustration going into a call. 

Step 3: Build for outcome

One of the big mistakes companies make when implementing chatbots is believing that building basic FAQ bots will relieve pressure on their contact centres. What they fail to recognise is that most customers don’t make contact for general information, they’re fundamentally trying to achieve an outcome. Here are some ways to make the in-system connections that deliver results:

• Re-examine your routines

Particularly under the pressure of the pandemic, companies were forced to quickly digitise their processes to fit the constraints of our new circumstances. Now it’s time to go back and digitalise work, ensuring the process takes advantage of opportunities afforded by technology.

• Don’t ask what you already know

Check your silos. Customers hate re-explaining themselves, so why ask them to repeat information you already know about them, even if it is in a different part of the organisation? Conversational AI is great at de-siloing information by only accessing the input you need at that moment. It’s fine to ask customers to confirm that nothing has changed, just don’t assign them form-filling busywork instead of helping them. 

What other information can you capture? There’s a lot of input available to you based on phone number and device details, contextual information, or account history. Look to reduce steps for your customer, anticipating their needs, and delivering a more enjoyable experience with quicker outcomes.

• Handhold

Help customers take the right actions to achieve their goals, step by step. Well-built processes don’t need to collect information in a specific order and can remind users of what’s still missing. 

LLMs handle longer, more complex requests better than NLP models in general. Many Conversational AI platforms and tools now include ChatGPT integration points, so test to see if it makes better sense of complex requests than what you’re currently using.

• Analyse and refine

Conversational AI generates and builds on data by default; and it also enables continuous, light-touch analysis that uncovers actionable insights.

Using pre- and post-processing techniques enables safe use of LLMs in analysis. They can be used to handle PII masking up front, allowing you to train on your organisation’s own data, then check in post for hallucination detection and elimination.

Step 4: Help agents help you

With the increasing push to self-serve and customers’ growing digital literacy, the contact centre is rarely their first point of call when they need help. So stop trying to discourage customers from reaching out – the last thing an agent wants to face is a needlessly unhappy customer on every call.

We know that attrition is a big problem right now. Let’s look at what you can do to pre-empt that.

• Access via APIs

While you might not want to give everyone full access to every system in your organisation, you can allow read/write access to specific information via API calls. This type of targeted access for a purpose helps relieve organisational bottlenecks. In Conversational AI, it means that fields are updated in real time as information is provided and agents don’t need to spend time flipping between multiple systems to hunt down the right fields.

• Assist in assistance

Whether up front or behind the scenes, adding bots the right way can lighten agent workload. Agent Assist technologies empower agents with the information they need to help customers, shortening both their training and response time, increasing their confidence and performance, and their job satisfaction.

Particularly for newer agents, knowledge graph-type responses that provide agents with suggested reading can help them find the steps they need to take to help the customer without escalation. LLMs can suggest content here. It’s safer having a human in the loop instead of risking unmediated contact with customers, so make sure you require an approval step.

Next Best Action (NBA) suggestions can help agents prevent future calls and provide value by recommending further actions that might benefit the customer. See if LLMs can provide any suggestions here, with the earlier caveats. 

• Don’t send them in blind – summarise, surface, sentiment

Most customers expect contact centre agents to know who they are and be aware of any previous issues or interactions. But agents often can’t easily access that type of information, which adds to the stress of the job.

It’s important to empower agents with warm handover that includes transcripts, summarisation and surfacing of contextual information for quick digestion. That way customers don’t have the added frustration of re-explaining themselves once they reach an agent.  

Adding sentiment analysis gives agents the opportunity to respond in the right tone to meet the customers where they are from the start of the call or chat. 

• Tag team – calculate, collect, connect

Efficient experiences take a tag-team approach where bots collect information or connect to calculators when customers need help. The bot can use this to get them part-way to their goal before bringing in an agent to finish. Remember from above, agents aren’t left on their own once the customer is in contact with them. They continue to receive assistance from bots to help customers reach their goals. 

Here’s another place where LLM’s ability to understand more complex requests can be employed. It’s a good idea to check understanding and give users the ability to amend the received interpretation, but that’s true regardless of the underlying language model employed.

Bots can even schedule appointments or call-backs or hand off to another channel if a customer isn’t prepared to actively wait in a queue – any place a customer might drop off, offer assistance.

• Facilitate agent influence

Agents work on the front lines and have the deepest insight into where your business processes are going wrong. Rather than letting that insight leave with them, enable them to improve their workplace and retain both them and their experience, while shortening handle times and providing a better total experience. 

Reassure your agents that their roles are not in jeopardy because of the growth of AI. It’s worth ensuring they understand there’s plenty of work for them for the foreseeable future – and that they can help shape their workplace into a place they thrive. 

Becoming a bot trainer can be an attractive career progression path as well. Helping agents to improve bots catalyses a virtuous circle. 

These four steps will allow you to implement a reliable and effective strategy, ensuring the success of your Contact Centre. By adopting them, you adopt a guide that can be referenced to support further development over the coming years.