Getting started with AI sounds simple. We know it’s not.
There’s a point most customer service teams seem to reach on their quest to reduce costs where it’s no longer a question of whether AI will play a role – that part is, for the most part, already decided – it’s a question of how. As in, how do you actually turn your people-heavy, process drowning customer service operation into an AI powered solution?
Tools like ChatGPT have made AI feel immediate and intuitive. You ask a question, and you get a coherent answer almost instantly. In isolation, it feels like something you could just plug into a customer service environment and ‘voila’.
In practice, it doesn’t quite work like that.
AI doesn’t automatically understand your business. It doesn’t have access to your systems unless you explicitly connect them. And it won’t fix processes that are already unclear or inconsistent. What it does do is follow structure, which means that the effectiveness of any AI solution is closely tied to how well the underlying operation is defined. The clearer your processes, the more useful your AI becomes.
So step one begins to get a bit clearer – you start by taking a close look at your existing operation.
Step 1: Defining “day one”
A common instinct when starting out is to think big. If AI can handle customer queries, then why not have it handle all of them?
It’s an understandable goal, but in reality it tends to introduce too much complexity too early. Customer service operations are full of edge cases, exceptions, and dependencies on internal systems. Trying to replicate all of that in one go can quickly become overwhelming.
Rather than aiming for full automation, it’s more useful to think in terms of getting to day one. By that, we mean a point where AI is successfully handling a single, real customer service process from start to finish.
That first use case should be relatively contained. Typically, it will be something that is:
- high in volume
- repetitive in nature
- clearly defined
- and relatively low risk
For many retail businesses, “Where is my order?” (WISMO) is a natural candidate. It follows a predictable structure and relies on data that can usually be accessed in a consistent way.
It may not be the most exciting place to start, but it’s a practical one.
Step 2: Assigning ownership
Like any major project, this one is going to benefit from clear ownership. Without it, you’re going to quickly lose momentum – particularly because a demanding project like this one straddles a wide range of skillsets. You’re going to need people who are technical, and organised, and close to the existing processes. You’re going to need people who think strategically – but also people who see the details clearly. That’s unlikely to culminate in one person. Here’s who you’re going to need to get to a successful ‘day one’.
- A primary owner / project lead: Someone who is ultimately responsible for the AI solution. They will shape how AI is applied, deciding that first use case, the channels to launch on, setting general tone and behaviour rules for the AI itself. This is your major decision-maker.
- Technical support: You’ll need someone who understands the system(s) your chosen use case relies on, how they connect, and what data is available. Much of the early work in an AI project revolves around integrations, so having this in place early on is going to remove a lot of blockers.
- Process owner: AI can only follow what has been clearly defined. Someone needs to take responsibility for documenting the chosen use case in detail. All the specific variations, decision points and edge cases to solve that customer problem, so that your AI has accurate answers for every question.
In some organisations, these roles may be three people – or they may overlap. What matters most is that each of these responsibilities are covered.
Step 3: Choosing a vendor
At this stage, most teams begin evaluating platforms.
There are plenty of options available, and as the team behind Gnatta, we naturally believe our approach is a strong one. If you’d like to explore that further, we’d love to chat to you → [Get started with Gnatta]
More broadly, though, the important thing is choosing a platform that aligns with how AI actually needs to be implemented in a customer service environment.
A few considerations tend to be particularly important:
- How much control do you have over its behaviour? You should be able to define what the AI does, what it doesn’t do, and precisely when it should hand over to a human
- Can you split off high-risk processes? You should be able to separate sensitive conversations and apply different logic and processes – often, that means a multi-agent architecture. i.e. Avoid one AI Agent answering ALL queries. Check this blog for more info.
- How easily can it connect to your existing systems? If the AI can’t access the same data your advisors rely on, its ability to resolve queries will be limited.
- Is it easy to use? You need your team – your existing team – to be able to use and manage this system without long training courses or technical experience.
Choosing carefully here can make a big difference later on.
Step 4: Mapping the process
For AI to work reliably, the process it’s following needs to be clearly defined.
That usually means documenting the current workflow in more detail than you might expect — including different variations of the same query, decision points, and any conditions that affect the outcome.
If you’re approaching this for the first time, we’ve created a simple template to help structure that process:
👉 Download the AI Process Capture Template
Step 5: Building integrations
Most customer interactions depend on access to information — order details, account data, delivery updates, and so on. AI is no different in that respect.
A useful question to ask early on is: what does the AI need to see in order to resolve this query? Closely followed by: how will it access that information?
If that data isn’t readily available via an API, or is difficult for a human to retrieve, it’s unlikely to be straightforward for AI either. Because of that, integrations often become a key part of the early work.
Your tech support lead will need to identify each required data point and begin configuring those integrations in your chosen AI vendor as soon as possible.
Step 6: Launching ‘day one’
With your process defined and necessary data made available to the AI, you’re ready to introduce it to a live environment. At this stage, it’s important to keep all contacts that aren’t part of your selected use case in their existing workflows. Use the automation and filtering options available in your chosen vendor to route only the relevant queries to your AI Agent so you can establish:
- how often it resolves queries successfully
- how often it escalates to a human
- how customers respond
- how much volume is deflected from your existing operation
Analyse interactions to identify process and tone of voice tweaks, upgrade integrations where necessary, build in additional logic for unforeseen edge cases (it will happen!) and iterate until you’re confident your AI Agent is handling the query robustly.
Achieving full automation
Once your first use case is working effectively, the process for expanding becomes repeatable. You’re no longer starting from a blank page, you have a framework. You understand what’s required, and your team is equipped to pick a new use case, map out those processes and integrations and conquer it together.
It’s simply a case of selecting the next use case, and applying the same principles.
The role of human teams
As AI takes on more process-driven work, the role of human advisors begins to shift. Your team becomes responsible for:
- maintaining and improving AI-driven processes
- designing new journeys and use cases
- handling complex or sensitive interactions
- contributing to retention and customer loyalty
Rather than being replaced, human teams move closer to shaping and overseeing the experience, while AI handles the more repeatable aspects at scale.
AI in customer service is inevitable — but your existing team are the ideal architects for it, and their knowledge will remain invaluable.
We’ve helped a number of organisations navigate that process, from identifying the right first use case through to launching and scaling AI in a controlled, practical way.
If you’d like to explore what that could look like for your team, we’d be happy to talk it through.
