AI is reshaping credit and collections in South Africa, this we know. We have to also understand the real winners aren’t the banks that rush the hardest into new tech.
The winners are the banks that use AI to make customer interactions feel more human, not more robotic.
Right now a lot of lenders are looking at AI mainly as “intelligent automation.” They want to push more work through the system, cut bottlenecks, speed up decisions and grow the book without adding people every time volumes spike.
This all makes sense in a world where cost-to-income is under pressure. The risk is that “automation” quietly turns into “automation only,” and customers start to feel like they’re dealing with a machine, not a partner. In collections, that’s a quick way to create mistrust, complaints and broken relationships.
In countries where unemployment is high and debt stress is everywhere, how you treat people when they’re under pressure matters just as much as how much you collect.
When someone answers a call or opens a message from their bank, they’re often anxious, embarrassed or irritated.
If the first touch feels cold and scripted, you might get the instalment, but over time, you’ll probably lose the customer.
If it feels like a real conversation with someone who actually gets their situation, you’ve got a chance to sort out the immediate problem and strengthen the relationship.
This is where AI can actually help you be more human at scale, if you use it properly. It can take care of the repetitive, mechanical work that clogs up collections teams: pulling data from different systems, building worklists, prioritising queues, logging outcomes and sending standard communications on time.
It can pick up patterns you’d never see by hand and suggest the next best action for thousands of accounts at once. Used this way, AI buys your people the time and context they need to do what they do best.
The starting point is working out which parts of the journey really need a human touch and which don’t. Most customers don’t need to speak to an agent just to get a reminder or a confirmation that their promise to pay is captured.
Self-service, chatbots and simple digital journeys can sort out these low-complexity interactions quickly and conveniently.
Where AI really adds value is deciding who can safely stay in these low-touch paths and who needs more attention.
For example, a customer with a clean track record and a payment miss can get a light digital nudge.
Someone showing multiple stress signals, or flagged as vulnerable, should be escalated to an agent or a field team.
Once you’ve drawn this line, the next step is to give your people better tools. An agent shouldn’t have to click through half a dozen systems to see a customer’s history, current products, previous promises and recent behaviour.
A human-centred collections setup brings this into a single view, together with practical AI suggestions. Instead of a rigid script, the agent sees context and guidance: likely reasons for the arrears, offers that work for similar customers and risk indicators that suggest a softer or more cautious approach.
The agent still makes the call, but the system gives them a much smarter starting point.
Keeping AI human also means being honest about its limits. Models are only as good as the data and assumptions you feed them.
They can be biased, they can be gamed and they can produce recommendations that look clever in a slide deck but clash with your values or your brand.
This is why human oversight isn’t optional. Governance needs to spell out who’s accountable for AI-driven decisions, how models are tested and monitored, and when people are expected to override the system. Training teams to question the technology, not just obey it, is a big part of staying in control.
On the customer side, transparency goes a long way. People are much more likely to accept AI-supported decisions if they feel they’ve been treated fairly and can understand what’s going on.
This might mean explaining in plain language why a specific offer is on the table or why certain options aren’t.
It might mean giving customers a simple way to contest a decision or ask for a human review.
In collections, it often means acknowledging the reality of their situation and working with them on a plan that lines up with both your policies and basic common sense.
In practice, a human-centred AI strategy in collections usually follows a few core principles. Design journeys that start with automation and escalate to humans, not the other way around.
Use AI to prioritise, predict and prompt, while letting people handle the negotiation and the human side of the conversation.
Measure success with more than just cash collected; track customer satisfaction, complaints, roll rates and even staff engagement.
Make sure frontline teams have a voice in how tools are built and rolled out, so the tech actually fits the way they work.
It’s also worth accepting the fact that this is a journey, not a switch you flip overnight. South African banks and retailers are at very different stages in their AI journey.
Some are still running small pilots, others are building enterprise-wide platforms. Wherever you sit, the key question isn’t “How much can we automate?” It’s “Where will technology genuinely help us serve customers better, and where do we need humans in the loop to stay authentic and trusted?”
If you can use AI to strip out friction, spot trouble earlier and give your teams richer context, you create space for more real conversations with the people who need them most. In collections, this is where the real value sits.
The institutions that keep human focus front and centre will be the ones that earn durable loyalty, even when the conversation starts with a missed payment.
