When I was brought into a fast-growing software company to build a customer success function, the first thing I did was read support tickets. Hundreds of them. I wanted to understand why customers were getting in touch.
What I found wasn’t a customer success problem.
There were no measured response-time targets, follow-ups were diaried inconsistently, there was no escalation path for when a customer complained, and no way to tell whether a case had been resolved or simply closed by the agent. Revenue was growing, so nobody saw a problem. But when I looked at the numbers, the count of customers staying beyond their first three months was flat.
I couldn’t talk about customer success until the basics of support and complaints handling were standardised. So that’s where we started: response templates, a clear split between first and second-line support, and a proper complaints process. Only then did we bring in AI chatbots to handle the simple requests.
That order mattered. A chatbot dropped onto the original chaos would have answered faster, but haphazardly, and closed cases nobody had resolved.
Not everyone agreed. Some of the longest-serving support staff felt they already knew their customers well and that the new process was unnecessary. What changed the conversation was customer journey mapping. We ran workshops across the company that followed a customer from first contact to renewal, and the gaps were easier to see together than in any report: the handoffs nobody owned, and the points where nobody followed up. It turned an argument about process into a shared view of the customer’s experience.
AI amplifies what’s underneath it
I’m genuinely excited by what AI can do. But I keep meeting leaders who ask, “Where do we start with AI?” and the honest answer is usually, “Not there.”
AI is a multiplier. Put it on top of clear processes and trustworthy data, and it multiplies value. Put it on top of unclear ownership, inconsistent data and undocumented workarounds, and it multiplies those instead.
I’ve seen the same lesson in other settings. On a long-running data quality problem at a regulated utility, one that had been a stated priority for over a decade without improving, we considered off-the-shelf and AI options. Neither was the answer. What worked was understanding the process that created the bad data, changing it, and building a bespoke tool in-house. Accuracy reached 99.7%.
A readiness check you can run this week
Before you spend on AI, ask three questions:
- Can you write the process down? If two people would describe it differently, an AI will learn one of the versions at random. A customer journey map is a quick way to find out: walk the process from the customer’s side and see where it breaks.
- Do you trust the data? If people routinely work around the numbers, a model will treat the workarounds as truth.
- Does someone own the outcome? If nobody is accountable when it goes wrong, nobody will notice that it has.
If the answer to any of these is no, that’s your starting point. It’s less glamorous than a pilot, but far more valuable.
Where AI then earns its place
Once the foundations are there, AI really does deliver. In the support example, chatbots took the simple, repetitive requests, and the team had more time for the complex cases, the ones where a customer needed a person.
The shortcut isn’t skipping the basics. It’s doing them first.
So here’s my question for you: if you handed your most important process to a new hire today, would it work?
Katie Tether – Business Analyst & Process Manager
Katie is a senior business analyst and programme lead with deep experience translating complex, multi-stakeholder requirements into delivered outcomes – across utilities, hospitality technology, health, financial services, and insurance.