There is a moment familiar to almost every South African consumer. It is half past eight at night, the geyser has packed up, and you are trying to log a claim on an insurer's app. The chatbot asks you to describe the problem. You do. It offers you three options, none of which is your problem. You type "agent". It offers you the same three options. Somewhere in that loop, a decision made eighteen months ago by a product team you will never meet, working to a cost target set by a finance committee you have never heard of, has just told you exactly what that company thinks you are worth.
Nobody in that company would describe the chatbot's decision tree as a brand decision. The brand people were busy with the sponsorship, the tone of voice document, the campaign. The chatbot belonged to operations, or digital, or whoever owned the cost-to-serve line. And that separation, which made a certain amount of sense when brand lived mostly in advertising, has quietly become the most expensive assumption in the business.
The operating layer is now the brand layer
Brand used to be something a company said about itself, transmitted through channels it controlled, and experienced by customers in a handful of set-piece moments. A shop floor. A radio ad. A person on the other end of a phone.
That is no longer how it works. Most of what a customer now experiences of a company is produced by a system rather than a person: the ranking of search results, the recommendation on the home screen, the fraud model that declines the card at the till, the pricing engine that quotes one number today and a different number tomorrow, the routing logic that decides whether your query is worth a human. These are not communications. They are operations. But they are what the customer actually meets, and they arrive far more often than any advertisement does.
Which means the brand is no longer being authored by the people who think they are authoring it. It is being authored, thousands of times a day, by whoever tuned the model.
Recent work by Accenture researchers, published in Harvard Business Review in July 2026, puts numbers to the cost of ignoring this. They benchmarked sixty companies in athletic apparel and in hotels, scoring what they call "brand debt", the accumulated drift between what a company promises and what its products, policies and systems actually deliver. Firms carrying the least of it were between two and four times more likely to beat their industry's average revenue growth over the following one and three years. Their scoring drew on public evidence rather than internal data, and the authors have a consulting practice to sell, so treat the multiples as directional rather than precise. The direction is still worth taking seriously.
Why this bites harder here
South African companies are exposed to this in ways that companies in higher trust markets are not.
Trust starts lower. A customer who has spent years dealing with a municipality's billing department, or waiting on a medical aid authorisation, does not extend the benefit of the doubt to your automated decline. They read it as the system working exactly as intended, against them. The recovery cost after a bad automated interaction is therefore higher here than the global averages suggest.
The distribution of failure is also faster and less forgiving. Complaints do not sit in a call centre queue anymore. They go to X, to Hellopeter, to a WhatsApp group of four hundred people in the same estate, often with a screenshot attached. An algorithm that treats a category of customers slightly worse than others does not produce one complaint. It produces a pattern, and patterns get noticed, aggregated and written up.
Then there is the regulatory floor. POPIA already requires that a person be able to contest a decision made purely by automated means where it significantly affects them. Most companies treat that as a compliance box. It is more useful read as a warning about what automated decisions do to a relationship when they cannot be questioned.
Borrow the discipline you already have
Here is the part that should be uncomfortable for most executive teams. We already know how to govern this kind of risk. We do it in two other domains, seriously and continuously, and we do not do it for brand.
Financial risk has owners, thresholds, delegated authority, a schedule of review, an audit function that reports independently, and consequences when limits are breached. Technical risk has change control, staging environments, rollback plans, severity ratings, incident reviews and a register of known debt that somebody is accountable for paying down. Neither system is perfect. Both are vastly better than nothing, and both work because they attach a named person to a defined exposure and force a periodic conversation about it.
Brand exposure has none of that. It has a set of guidelines, a quarterly tracker that arrives too late to act on, and a chief marketing officer who is held responsible for trust erosion caused by systems they were never asked to approve.
Applying the same discipline is not complicated in principle. It means an automated system that touches customers gets a brand review before deployment, in the same way it gets a security review, and the reviewer has the standing to say no. It means someone owns the question of what the system's outputs actually feel like on the receiving end, and monitors that continuously rather than checking uptime and calling it quality. It means keeping a register of deliberate shortcuts, because some shortcuts are worth taking, and the danger is not the individual trade-off but the compound interest on twenty of them nobody wrote down. It means treating a customer facing automation failure with the seriousness of an outage: a post-mortem, a root cause, a fix, a record.
The obvious objection is that this adds friction, and friction is the enemy of shipping. Fair. Financial controls also add friction, and no serious business proposes doing away with them, because the alternative has been demonstrated too many times. The question is not whether to accept friction but where to put it. Right now most organisations put none of it in front of decisions that reach millions of customers, and quite a lot of it in front of a press release that reaches nobody.
The uncomfortable conclusion
If most customer experience is now produced by systems, then the people configuring those systems are the brand team, whatever their job titles say. The cost target set in a budget meeting, the threshold chosen by a data scientist, the decision to route a category of query to a bot rather than a person: each of these is a statement about who the company values and how much. Customers read those statements accurately, and they read them long before they read your campaign.
Governing that with the same seriousness applied to money and to code is not a marketing initiative. It is basic operational hygiene for a business whose reputation is now being written, continuously and at scale, by machines.
