Imagine a business that makes three changes. It reduces customer support hours, introduces a new service charge and makes its returns process more restrictive. Costs fall. Revenue per transaction rises. The next management report looks encouraging.

Whether the business has improved is a different question.

This is a hypothetical example, but the decision it illustrates is worth examining: are we creating more value, delivering it more efficiently, or simply asking customers to accept less?

Three ways to improve a margin

A better margin can come from a better product. Customers value the offer more and are willing to pay for it. It can come from a better operation: fewer mistakes, less waste and a simpler process. It can also come from changes that improve the company's immediate economics while making the customer's experience worse.

The reported result may look similar. The implications for the business can be quite different.

That is why I would want a margin discussion to include the customer consequence of the decisions behind it. Which part of the experience changed? What did the customer gain? What became harder? What evidence would tell us that we had gone too far?

Ask what the customer is buying

A customer rarely buys only the item listed on an invoice. A purchase can include confidence that something will arrive on time, the convenience of resolving a problem, or the enjoyment of an experience.

Consider a retailer. The product matters, but so can stock availability, knowledgeable staff and a straightforward return. A cut to one of those elements may initially look like a saving. Its full consequence depends on how much that element mattered to the customer's decision to shop there.

It would be equally mistaken to assume every service investment is valuable. Some activities cost money without meaningfully improving the experience. Understanding the distinction is part of the work.

Read the results together

I would examine financial performance alongside the behaviour and experience of comparable customer groups. Are people choosing to return? Are problems resolved? Are complaints changing? Are we spending more to replace customers who no longer see a reason to stay?

None of these measures is sufficient alone. A repeat purchase does not tell us everything about satisfaction. A lower complaint count could reflect better service—or a harder complaints process. Even a familiar retention percentage is difficult to interpret without understanding who is included, over what period, and what counts as retained.

The useful question is whether the evidence tells a consistent story. If margins rise while service deteriorates, that deserves investigation. It is a signal to understand, not proof that a particular decision was wrong.

The counterargument matters

A sustainable business cannot promise every customer everything. Prices sometimes need to rise. An expensive feature may need to go. A generous policy can become impractical or attract abuse.

There are also occasions when accepting some customer losses improves the business. Serving a particular need may be uneconomic, or inconsistent with the offer the company can reliably deliver.

The test is whether the choice is deliberate and understood. Can we explain the value proposition honestly? Are the trade-offs visible? Are we measuring the consequences we expect, including the uncomfortable ones?

In sectors where repeated use can cause harm, including betting, loyalty also needs careful interpretation. More activity is not automatically a better customer outcome. Responsible participation and customer wellbeing belong in the assessment.

A more useful performance conversation

Before celebrating an improvement, I would ask four questions:

1. What specifically changed to produce the result?
2. How did that change affect the customer's experience?
3. What evidence suggests the improvement will last?
4. What would cause us to reverse or adjust the decision?

Those questions do not require a complicated new reporting system. They require a willingness to connect the numbers with the decisions that produced them.

The ambition is a business whose economics improve because its offer and operation become stronger. When that happens, customers have a reason to keep choosing it—and the company has a better foundation on which to grow.

An AI demonstration can be impressive in five minutes. A commercial result takes longer to establish.

A report appears almost instantly. A customer reply is drafted before someone has finished explaining the problem. A presentation that once occupied an afternoon takes minutes.

These are useful capabilities. But they answer only the first question: can this task be done faster?

The more important question for an established South African business is what improves because it is faster. Does the customer get a better outcome? Does the business remove a constraint? Does the team produce more valuable work at an acceptable total cost?

Time saved is an input to the business case. It is not the completed business case.

What the evidence says

A study published in February 2025 examined the introduction of an AI assistant across 5,172 customer-support agents. Issues resolved per hour increased by 15% on average, with benefits varying substantially across workers. This was a particular workplace, not a promise for every business. Read the study: Generative AI at Work.

Other findings underline why context matters. METR’s July 2025 study found that 16 experienced open-source developers took 19% longer on selected tasks when AI tools were available. That result concerned early-2025 tools in a specific setting; it was not evidence that AI generally reduces productivity. Read METR’s study.

The update matters too. In February 2026, METR reported that selection effects made its follow-up results unreliable for estimating current productivity gains. The researchers believed developers were probably benefiting more than before, but could not confidently establish the size of that improvement. Read the update.

The commercial lesson is to test the actual work, with the actual people, using the tools available now.

Start with a constraint

“Where can we use AI?” produces a long list. “What is preventing this business from serving customers better?” produces a more useful one.

Consider a hypothetical retailer whose service team spends too much time finding product information. An assistant that retrieves approved information could be valuable. But if the underlying problem is inaccurate stock data, generating a fluent answer faster may simply accelerate the wrong answer.

One problem concerns access to knowledge. The other concerns the reliability of the underlying operation.

Before selecting a tool, define the constraint in plain language. Customers wait too long for a correct answer. Staff repeat the same administrative work. Managers cannot identify recurring service problems quickly enough. Then establish what successful improvement would look like.

Measure the whole job

A draft is not a completed task. The complete job may include collecting information, generating an answer, checking it, obtaining approval, sending it and dealing with what happens next.

A pilot that measures only the generation stage can miss work transferred elsewhere. A manager may spend longer checking a report. A customer may need to contact the business again. Another team may have to correct a decision based on incomplete information.

Measure from the beginning of the job to an acceptable outcome. For customer service, that could mean the cost of resolving an issue correctly, alongside repeat contact and escalation. For internal reporting, it could mean the time required to reach a supported decision.

For a South African rollout, test the languages, channels and customer circumstances the business actually serves. A polished English demonstration is not sufficient evidence for every customer journey.

Decide what happens to the capacity

Suppose a pilot releases staff time. What is the plan for it? It might be a shorter queue, better follow-up, more coaching or the capacity to grow without adding the same amount of administration.

Those are different outcomes. They should not be presented as interchangeable savings.

If the team remains the same size and spends the released time improving service, describe that as service capacity. If faster work enables growth, establish whether there is demand and whether another part of the operation can fulfil it.

Someone should own the conversion from capacity into value. Otherwise, the benefit risks remaining an estimate in a presentation.

Protect judgment while changing the work

AI adoption is also a team-design decision. Who checks uncertain answers? Who can override the system? How do newer employees learn to recognise a mistake rather than merely accept a polished response?

Experienced people should help define what good work looks like, including the exceptions. Their role should not quietly become an unlimited queue of machine-generated material to approve.

A proposed pilot should have clear information boundaries, a route to human assistance and a way to stop the process when it fails. The customer should not have to become the quality-control department.

The case for moving quickly

There is a reasonable counterargument: businesses can analyse themselves into inaction. Tools improve rapidly, and waiting for perfect evidence can mean missing useful learning.

Choose one workflow. Establish a baseline. Give a responsible person authority to run the test. Compare quality, time and total cost against the existing approach. Decide in advance what would justify expansion, revision or stopping.

Not every experiment must produce immediate savings. Learning can be a legitimate objective, provided it is labelled and budgeted as learning.

A better boardroom question

“How many people are using AI?” tells a board something about adoption. A more useful question is: which customer or business outcome improved, what did the improvement cost, and how confident are we that AI contributed?

The answer may be less dramatic than a demonstration. It will be more useful for deciding what to fund next.

Businesses create lasting value by doing worthwhile work better—and making sure the benefit reaches someone.

I run a toy retailer and an online betting business from the same desk. People assume that means the skills transfer. Most of them don’t. One thing does, and it matters more than everything else combined.

The thing that transfers is a question: what gives us the right to win this market?

Strip a business down and that is what is left. A customer chooses you for a reason you can state in one sentence, or they choose someone else. Every board meeting I chair opens the same way, before a single number: what is this business for, and what is our right to win? Only then do we go to the metrics, the value added, the compliance and regulatory items.

The order matters. Metrics tell you what happened. The right to win tells you whether it will happen again. A board that starts with the numbers is auditing the past. A board that starts with the right to win is deciding whether the past is worth repeating.

What does not transfer is culture. The temptation when you run two businesses is to run one culture across both. It doesn’t work. A toy retailer and a betting company serve different people, at different moments, with different brands, and they need different cultures to do it well. What overlaps are values: excellence, learning, doing what needs to be done. Values are shared. Culture is local. Confusing the two is how a group quietly flattens the businesses inside it.

Something did cross over, though, and it went from the toy business to betting. Toy retail taught me that a store is theatre. The whole thing is built around finding the moments where you get to shine, and making sure you shine in them. We took that idea into the betting business and reframed it as moments that matter: find the handful of points in a fan’s experience where their view of us is decided, and design for those first. It changed what we prioritised. It also changed what we stopped doing.

The strongest objection to all of this is obvious: if you run two businesses, neither gets your full attention. That is true, and it is exactly why the right-to-win question earns its place. I cannot be in every room. A question that every business can answer in one sentence does the work I would otherwise have to do by being present.

So the honest answer to “what transfers?” is not a skill or a playbook. It is a discipline of asking one question first and refusing to move on until it is answered.

A growing business can look healthier than it is. More orders, more locations and more customers give everyone something to celebrate. The harder question is whether each new layer of activity makes the business more valuable or simply more demanding to run.

Growth deserves ambition. It also deserves a clear explanation. Before calling it good, I would want to understand what caused it, what it consumed and what remains when the initial excitement passes.

What did we actually sell more of?

Revenue can rise because more customers bought, because existing customers bought more, because prices increased or because the mix of products changed. Those are different stories. An acquisition adds another distinction: a larger group does not necessarily mean the original operation improved.

Separate the causes before debating the quality. If price increases account for most of the movement, look at customer behaviour alongside revenue. If a promotion brought in new customers, ask what they choose once the promotion ends. If one unusually large order changed the period, avoid treating it as a repeatable pattern.

None of these sources of growth is automatically bad. The point is to know which business is being built.

What did the growth require?

Consider a hypothetical retailer opening more stores. Sales increase, but so do stock commitments, fit-out costs, recruitment and the demands on experienced managers. A store can report promising trading while the wider business carries the cost of making it possible.

A useful discussion follows the cash as well as the income statement. When must inventory be paid for? When does the customer pay? What happens if stock moves more slowly than expected? Which costs remain if sales disappoint?

The same discipline applies to digital businesses. Serving another customer may appear inexpensive, yet acquisition, support, incentives and the systems needed to maintain service can change the economics. Averages deserve scrutiny when different customers or channels behave differently.

Is the customer proposition getting stronger?

Growth can put pressure on the very experience that earned the customer. Availability becomes less reliable. Support queues lengthen. The people who once knew the product well are stretched across too many locations or responsibilities.

That makes customer experience part of the growth discussion. Examine whether new and existing customers receive the promise that brought them in. Look at repeat choices, problems resolved and the reasons people leave. Interpret those measures together; no single metric proves loyalty.

In businesses where repeated use can cause harm, more activity is not a sufficient measure of customer value. Commercial success needs boundaries that protect customers, even where those boundaries constrain revenue.

Can the organisation carry the next stage?

An expansion plan is also a plan for people and decisions. Who will coach new managers? Which choices can move closer to the customer? Where is specialist judgment still necessary? A business that grows only by adding decisions to the founder’s desk has a capacity problem hidden inside its sales forecast.

The question is not whether every process is perfect. It is whether the few processes that protect the proposition are strong enough for the next step.

What would change our mind?

There is a fair objection: insist on certainty and a business will never expand. New markets require investment before the evidence is complete. Some capabilities only improve through use.

That is a reason to make the bet explicit. Define what the expansion is expected to teach, what the business can afford to commit and which results would justify another stage. A bounded investment with a clear learning objective is different from repeatedly funding a plan whose assumptions are never revisited.

Good growth need not be smooth, immediately profitable or free of mistakes. It should have a credible route to a stronger customer proposition and sustainable economics. The board’s job is to keep those two ambitions in the same conversation.

Every business depends on capabilities it does not own. It may use somebody else’s brand, distribution network, technology or specialist expertise. The useful question is which dependencies strengthen its offer—and which leave it unable to protect that offer when circumstances change.

Ownership is not a guarantee of control. A capability can sit inside a business and still be poorly understood, underinvested in or dependent on one person. Equally, a well-run partnership can provide expertise and reach that would be difficult to build alone.

Start with the customer promise. Then decide how best to deliver it.

Identify what makes the offer distinctive

Suppose a hypothetical retailer wins through thoughtful product selection and knowledgeable service. It might use a partner for delivery while keeping decisions about the range and staff development close to the business. Another retailer might compete on reliable fulfilment, making delivery capability far more central to its proposition.

The same activity can therefore deserve different treatment in different businesses. There is little value in an abstract rule that everyone should outsource logistics or own technology. The decision depends on where the business earns the customer’s choice.

Describe the capability precisely. “Technology” is too broad. Customer insight, product design, hosting, payment processing and service tools are different things. A business can retain judgment about the experience while buying much of the infrastructure beneath it.

Price the relationship as it will operate

A proposal can make a partnership look simple because it prices only the visible service. The operating relationship also requires integration, training, coordination, quality checks and someone to resolve exceptions.

Compare complete approaches, rather than a supplier fee against a conveniently narrow estimate of internal cost. Building in-house requires recruitment, management attention, maintenance and the ability to keep improving. Buying requires enough internal knowledge to be an informed customer.

Neither route removes work. It changes the work and who must be good at it.

Give the customer one accountable business

A customer with a problem should not need to understand the partnership structure. If a delivery fails or a service gives conflicting answers, someone must own the resolution across the boundary.

Before launching, agree how exceptions move between teams. Who can make a decision? Who communicates with the customer? What happens when the parties disagree about the cause? A service promise without a practical escalation route can leave the customer carrying the coordination burden.

This is particularly important when a recognised international brand meets local execution. The name may attract attention, but availability, service and relevance determine what the customer experiences in South Africa. The partnership has to support those everyday choices.

Preserve the ability to make a different choice

A strong partnership does not require pretending that circumstances will never change. Discuss how knowledge, customer service and essential operations would continue if the relationship ended or materially deteriorated.

That includes understanding what information can be recovered, how another provider could take over and which skills the business needs to retain. Specific arrangements need appropriate professional review; the commercial principle is to understand the dependency before relying on it.

An exit plan is not a prediction of failure. It is a test of whether both parties understand how the relationship works.

Make room for commitment

The strongest counterargument is that excessive concern about dependency can undermine partnership itself. Some opportunities require shared investment, patience and confidence that neither party will walk away at the first difficulty.

That is right. A partnership cannot be managed entirely through defensive clauses and competitive tenders. It needs people who can solve problems together and incentives that reward a successful customer outcome.

The balance is informed commitment: understand what is essential, retain the judgment needed to protect it, and invest in a relationship that gives both parties a reason to perform.

The business does not need to own everything the customer values. It does need to know how that value is created, who is accountable for delivering it and what it will do if the promise stops being met.

The first AI decision in an established business is often framed as a choice of software. A more useful starting point is a piece of work: something that happens often, matters to the business and has an outcome people can assess.

That makes the first experiment easier to learn from. It also avoids building a business case around a demonstration that was designed to impress rather than represent the work.

Choose a constraint people recognise

Ask a team where time disappears or customers get stuck. Then narrow the answer. “Administration” is too broad. Finding approved product information before answering a customer is specific. So is preparing an internal summary from a defined set of documents.

A good candidate has enough repetition to compare results, a clear owner and a way to identify mistakes before they cause serious harm. The most consequential decision in the business is rarely the easiest place to establish those conditions.

It is also worth asking whether the problem needs AI. An inaccurate stock record, an unclear policy or a duplicate approval may be better addressed directly. Automation cannot compensate for every weakness in the underlying operation.

Write down what good work means

Consider a hypothetical service team testing an assistant that finds product information. A useful answer must be accurate, relevant to the question and drawn from an approved source. It must also recognise when the available information is insufficient.

Speed is one measure. The number of corrections, repeat contacts and unresolved questions may matter more. Decide which standards are essential before seeing the tool’s output. Otherwise, an impressive response can quietly redefine success.

For a South African business, use examples from the customers and channels actually served. Include the languages, spelling variations and incomplete questions the team encounters. A demonstration built around ideal inputs tells only part of the story.

Start with a limited role

Retrieving information or preparing a suggestion gives people a different degree of control from allowing a system to send answers, alter accounts or commit the business to an action. Those choices should be explicit.

For an initial trial, define what the tool may access, what it may produce and which actions require a person. Give the team a clear way to report an error and continue the work without the tool. Use approved information and the organisation’s agreed data controls.

The person checking the output needs time, context and authority to reject it. Adding a nominal approval step does little if the reviewer is expected to accept everything immediately.

Compare the complete job

Establish how the work is performed today. Then compare similar tasks, including the time required to check and correct outputs. Record difficult cases as well as easy ones. A pilot that quietly excludes the awkward work can give a misleading picture of readiness.

Measurement deserves care. In its February 2026 update, METR explained that selection effects limited what its follow-up developer study could establish about current AI productivity gains. That research concerns software development; it is not a forecast for retail or customer service. It is a useful reminder to examine which people and tasks enter a comparison. Read METR’s research update.

Decide what the next investment depends on

Before the trial begins, agree what would justify extending it, changing it or stopping. Set a review point and a limit on the initial commitment. If the objective is learning rather than immediate savings, say so.

A reasonable objection is that narrow pilots can become an excuse to avoid larger change. The answer is to connect the experiment to a decision. A successful test should identify the next workflow, the capabilities needed and the conditions under which wider adoption makes sense.

The first experiment need not transform the business. It should produce trustworthy evidence about a useful piece of work—and leave the team better able to choose what happens next.

A board can receive a detailed AI presentation and still struggle to understand the decision in front of it. Demonstrations show capability. Adoption figures show use. Neither necessarily explains what the business should fund, what it is exposing to risk or who is accountable for the result.

The discussion becomes more useful when management connects AI to a specific operating choice. What will change, for whom, and on what evidence?

Which business problem are we addressing?

An AI initiative should have a purpose a director can describe without naming the product. Reducing the time needed to resolve a customer problem is a purpose. Improving the reliability of an internal review is another. Giving everyone access to a tool is an activity that may support those purposes.

Separate projects intended to improve an existing operation from experiments intended to explore a new opportunity. Both can deserve funding, but the evidence and expectations should be different. An exploratory budget should not be presented as a proven saving.

What supports the business case?

Ask for a baseline, the result of any trial and the limits of the comparison. Which tasks were included? Who participated? What counted as acceptable quality? Did faster production create more checking elsewhere?

Benefits also need an owner. If the business expects time savings, what will happen to that capacity? If it expects better service, how will customers experience the improvement? A projected reduction in effort is not automatically a reduction in expenditure.

The board does not need to run the experiment. It needs enough information to distinguish an observed improvement from an assumption that still requires testing.

Who can act, and who can intervene?

An assistant that suggests wording is different from a system that changes a customer record or takes an action. Ask where those boundaries sit and who is permitted to change them.

Management should be able to explain which information the system can access, what it is allowed to do and how an error reaches someone with authority to respond. The operating owner, technical specialists and relevant control functions need a shared understanding of those limits.

“Human oversight” is only useful when the human has the competence, time and authority to exercise it. A growing queue of approvals can turn a sensible control into a rubber stamp.

What happens to the team and the customer?

A technology decision can redesign work before anyone formally redesigns a role. Newer employees may have fewer opportunities to learn the underlying task. Experienced employees may spend more time reviewing exceptions. Managers may face a different mix of work and accountability.

Ask how training and performance measures will change. Retaining the ability to recognise a bad answer matters as much as learning to produce an answer quickly.

For customers, examine both the normal journey and the failure. Can someone obtain help when an automated response is wrong? Is there a workable alternative when the service is unavailable? In sensitive settings, would speed or increased activity conflict with a good customer outcome?

What could become difficult to reverse?

A small trial and a business-wide commitment create different dependencies. As adoption expands, examine what the business relies on: a supplier, access to information, particular staff knowledge or integration with essential processes.

Ask how the service would continue if a provider changed its terms or performance deteriorated. The answer need not be to own every component. It should be a proportionate understanding of the dependency and the choices available.

How do we avoid slowing everything down?

There is a legitimate concern that board attention can turn a modest experiment into a lengthy approval exercise. Oversight should match the consequence. A bounded internal trial should not require the same treatment as a system making consequential customer decisions.

Agree decision limits, reporting thresholds and circumstances that require escalation. Then let management operate within them. The board’s contribution is clarity about objectives, accountability and acceptable exposure—not involvement in every software choice.

A strong AI discussion should end with a decision: proceed within defined limits, gather missing evidence, expand what has worked or stop what has not. The value lies in making that decision well.

A business can make a customer promise in a sentence. Keeping it may require decisions across the entire operation. Available stock. A realistic delivery window. A partner who can fulfil their part. Someone with authority to put things right when the normal process fails.

That is why a promise belongs in the operating plan as well as the advertising. It commits resources before the customer arrives. And it provides a useful test of a business’s right to win: can the operation repeatedly deliver the reason customers choose it?

Start with the moment the customer is buying

Consider a hypothetical toy retailer. A parent ordering a birthday present is buying a product, but also relying on it arriving before a particular day. A broad range and an attractive price may win the order. Neither repairs the missed birthday.

This does not mean every retailer should offer immediate delivery. It means the offer needs to be specific enough for the customer to make a sensible choice. Delivery tomorrow and dispatch tomorrow are different promises. A product that appears on a website and a product available for fulfilment are different things too.

The commercial question is which commitment matters sufficiently to the intended customer to justify its cost. Speed might matter. Certainty might matter more. A narrower promise that the business can keep may be a stronger proposition than an ambitious one supported by hopeful assumptions.

Translate the words into operating choices

Once the promise is clear, work backwards. What must be true about stock information, staffing, order cut-offs and supplier performance? What happens during a busy period? At which point should the business stop accepting a commitment it can no longer fulfil?

These questions expose trade-offs that disappear inside general ambitions to improve customer experience. Holding extra stock ties up cash. Reserving capacity can mean turning away other work. A wider delivery area creates more opportunities, but may also create commitments the existing network cannot reliably support.

For a South African business, the test should reflect the places and customers it actually serves. A service that works in one dense urban area is not automatically an economical national proposition. Publish the relevant boundaries clearly and test expansion against the complete cost of serving the next area.

Measure the promise across the hand-offs

A customer does not experience the organisation chart. They experience whether the order arrives, whether it is correct and whether someone helps when it is not.

A useful management view therefore follows the commitment from acceptance to completion. An order processed quickly can still arrive late. A support ticket closed can still leave a problem unresolved. Departmental efficiency matters, but it cannot substitute for the outcome the customer was promised.

Assign one accountable owner for that outcome, with authority to work across the relevant teams. This does not remove specialist responsibilities. It gives the hand-offs somewhere to be resolved instead of leaving the customer to discover which department has the problem.

The same principle applies to partnerships. If a supplier or delivery partner performs part of the service, the customer still needs a clear route to resolution. The commercial relationship should establish how information moves, who can intervene and how a failure is handled. A strong brand cannot outsource the customer’s experience of its promise.

AI makes the question sharper

Sixty60’s company-issued announcement of Pixie described personalised product recommendations and a Smart Basket that customers could edit. It announced an initial beta for Xtra Savings Plus subscribers, with further conversational capabilities described as future plans. This is evidence of a retailer developing AI-assisted shopping, not proof of its effect on retention or profitability.

The wider operating implication is worth examining. A more relevant recommendation can make choosing easier. But if the product cannot be supplied, or the delivery commitment is unreliable, the recommendation has improved only one part of the experience. Better persuasion creates little lasting value when fulfilment cannot support it.

For businesses considering AI-assisted selling, test the connection between the recommendation and the information behind the offer. Is availability current? Are service boundaries accurate? Can the customer correct the suggestion? These are operating questions with commercial consequences, not simply questions about how natural the assistant sounds.

The case against promising less

There is a fair objection: a business that promises only what is comfortable may become timid. Competitors stretch expectations. Customers do not reward operational caution for its own sake.

The answer is to distinguish a deliberate stretch from an unsupported claim. A new service can be tested in a defined area, with a limited range or a capped number of orders. That creates learning while keeping the commitment understandable. Expansion should follow evidence that the promise can be sustained at an acceptable cost.

Nor should the business buy perfect service regardless of economics. Some customers will choose a cheaper, slower option. Offering that choice transparently can be better than building an expensive universal standard that customers neither need nor fund.

Give the board a promise it can examine

Alongside the financial results, management could bring five short answers: what are we promising; to whom; what must work to deliver it; where did it fail; and what decision is now required? The purpose is to identify choices about investment, capacity and accountability, not to turn directors into dispatch managers.

Include recovery in that discussion. A commitment needs a response when it fails: prompt information, a realistic alternative and an owner who can resolve the problem. Recovery should be designed before customers need it.

A right to win becomes more convincing when it is supported by operating choices competitors would have to match. The promise attracts the customer. The ability to keep it gives them a reason to return.

Source: Sixty60: Pixie product announcement, 9 April 2026. Product plans are described as announced, not as verified later outcomes.

South African online retail is expected to account for an average 10% of national retail turnover in 2026. The estimate comes from the Online Retail in South Africa 2026 study, produced by World Wide Worx with Mastercard, Peach Payments and Ask Afrika.

Ten per cent is still a minority of retail sales. But it is too large, and growing too quickly, to be managed as a side project. The study estimates online spending of about R159 billion this year and growth of 22.5%, compared with nominal retail growth of roughly 4% through May.

The strategic question is no longer whether a retailer needs an online channel. It is whether the business knows how its stores, digital journeys, fulfilment, loyalty programme and partners work together to create a profitable customer relationship.

The channel scoreboard has reached its limit

During an early build phase, it makes sense to isolate online sales, app downloads and digital acquisition. Management needs to know whether the new capability is gaining traction.

The problem begins when the reporting structure becomes the strategy. An online team can improve conversion while the business creates more delivery cost. A store can lose a sale after helping a customer choose the product. A promotion can lift digital revenue by shifting an existing customer to a more expensive fulfilment route.

None of those outcomes means the channel is failing. They mean channel-level numbers cannot answer the commercial question on their own.

Once customers move between search, store, app, delivery and collection, the useful unit of analysis becomes the customer and the complete order. What did it cost to win, serve and retain the relationship? Which part of the experience created the value, even if another part recorded the sale?

Make the economics travel with the order

A digital order has costs that can sit in different places: picking, packaging, payment, fraud prevention, delivery, returns and customer support. Stores may provide inventory or labour. Partners may perform fulfilment. A loyalty benefit may be funded by another commercial arrangement.

If each team optimises only its own line, the business can grow turnover while weakening the economics of the complete transaction.

Management does not need perfect attribution before making decisions. It does need a consistent view of contribution after the costs that change with the order. That view should include failure: repeat delivery, refund handling, substitution and the support required when the promise is not kept.

The same discipline applies to customer value. A shopper who buys in store this week and online next week is not two customers. The business should be able to see whether convenience increases frequency, expands the basket or simply moves an existing purchase into a more costly route.

Design one proposition, with honest choices

A joined-up model does not require identical pricing, range or service everywhere. Different channels have different economics. Pretending otherwise can create a promise the operation cannot support.

The better approach is to give customers clear choices. Collection may be cheaper than delivery. A store may offer immediate advice and availability. A subscription may make sense for households that order frequently, while an occasional shopper may prefer to pay per delivery.

The proposition should explain the trade-off without making the customer decode the organisation. One brand can offer several fulfilment choices, provided the prices, timing and recovery path are understandable.

Loyalty should connect the system

At this scale, a loyalty programme should do more than attach a discount to each channel. It can help the business recognise a customer across journeys, reward valuable behaviour and make the next interaction easier.

That does not mean rewarding every online order. The commercial value may lie in collection, a consolidated basket, a return completed in store or a partner benefit that reduces the company’s direct cost. A good programme makes those choices useful to the customer as well as sustainable for the business.

The report identifies subscriptions and delivery economics as important competitive tests. The attraction is understandable: a subscription can reduce the customer’s hesitation to order again. But free delivery is never free to fulfil. The economics depend on frequency, basket, density and what the membership changes.

AI should improve a decision, not decorate a channel

Online retail creates more opportunities to use AI in search, recommendations, service and forecasting. The same commercial test applies. Which decision improves, for whom, and what happens elsewhere in the operation?

A more relevant recommendation can lift conversion. If the item is unavailable or the fulfilment promise fails, the system has made the front of the journey better without improving the customer outcome. An assistant that reduces service time is useful only if the answer is correct and the issue is resolved.

AI projects should therefore be measured across the complete job. Faster generation, higher clicks or more personalised offers are inputs. The business result includes fulfilment, margin, repeat use and the work required to correct mistakes.

The physical business still matters

The strongest objection is simple: 90% of retail turnover remains outside online commerce. A retailer can damage a strong store business by chasing a digital growth statistic.

That objection is right. The 10% milestone is not a command to become digital-first. It is evidence that digital can no longer be treated as separate.

Stores may remain the most valuable part of the proposition. They can provide discovery, advice, immediate possession, returns and trust. The management task is to understand those contributions and design the network around what customers value—not to force every transaction through the newest channel.

A better board discussion

Boards should ask for one view of growth across the customer system. Four questions are a useful start.

Which customers are growing, not only which channels? What is the contribution after the costs of serving the complete order? Which hand-offs create friction or failure? What investment would improve the proposition across more than one channel?

The answers may justify more fulfilment capacity, better stock visibility, a revised loyalty benefit, a different partnership or fewer promises in places where the economics do not work.

At 10%, online retail has earned more than a line in the channel report. It has earned a place in the operating model. The businesses that create lasting value will not be those that declare stores finished or digital victorious. They will be those that make the whole system work for the customer—and know what that costs.

Source: Online Retail in South Africa 2026 findings, published by Mastercard. Forecasts and estimates are presented as reported, not as audited outcomes.

South African loyalty programmes can produce impressive numbers. Sasol said in a 16 September release that Sasol Rewards had grown to more than 2.4 million members, that members accounted for 65% of sales, and that the programme had returned more than R251 million in value to customers over the preceding year. It also reported an 88.6% redemption rate.

Those figures show reach, participation and customer use. They deserve attention. But they do not answer the question a board ultimately has to ask: what changed because the programme exists?

That is not a criticism of Sasol Rewards, which was recognised at the 2026 South African Loyalty Awards. It is a distinction between evidence of engagement and evidence of incremental value. The distinction matters because a large programme can be popular with customers while still rewarding behaviour that would have happened anyway.

The counterfactual is the commercial question

A loyalty dashboard usually makes activity easy to see: members, active members, points issued, points redeemed, app usage and sales linked to a card. These are useful operating measures. None of them, on its own, establishes causation.

Members are often a company’s best customers before they join. They buy more frequently, spend more or prefer the brand already. If their full spend is credited to the programme, the programme receives credit for the relationship rather than for the behaviour it changed.

This is not merely a theoretical concern. A study of seven grocery loyalty programmes found a positive effect on share of wallet after correcting for customers who self-selected into membership, but the effect was seven times smaller than a naïve comparison between members and non-members suggested.

The right commercial question is therefore a counterfactual: what would this customer probably have done without the reward, offer or partner benefit?

No business can observe that alternative perfectly. It can estimate it more honestly.

Measure behaviour, then subtract the full cost

The first test is behavioural. Did the programme increase purchase frequency, move more of the customer’s category spend to the business, reduce lapse, expand the basket or change the mix towards something more valuable?

The second test is economic. What did the change cost after rewards, discounts, technology, communication, service, fraud controls, partner fees and the administration required to run the programme?

That calculation should happen by customer group and behaviour, not only at programme level. A reward may change the habits of an occasional customer while simply subsidising a heavy buyer. A partner benefit may attract a useful new audience while adding cost for customers who were already loyal.

Research on paid free-shipping memberships illustrates the point. One field study found that the programme increased net customer revenue over time, but the economics varied across segments; some heavy buyers generated no significant increase and could become less profitable when they split orders to use the benefit. A single average can hide the customers for whom a proposition works and those for whom it does not.

The board does not need to run the analysis. It does need to insist that management can explain where the incremental value comes from.

Redemption is neither automatically good nor bad

A high redemption rate can indicate that rewards are understandable and useful. That is good customer design. It also means the economic cost is real rather than theoretical.

A low redemption rate can reduce short-term cost, but may signal that the benefit is difficult to access or not worth the effort. It can also leave customers with a growing balance they do not value.

The point is not to maximise or minimise redemption in isolation. It is to connect redemption to the behaviour that follows. Do customers return sooner? Do they buy in a different category? Does the benefit make the wider proposition more useful? Does it create another support problem?

The same discipline applies to partnerships. A partner can make a programme more relevant by extending where value can be earned or used. It can also add rules, dilute the proposition and introduce an economic dependency the customer never sees. The partnership should be judged by the customer behaviour it improves and the value left after both parties are paid.

The objection: loyalty is a long-term asset

The strongest counterargument is that loyalty cannot be reduced to the next transaction. A good programme can improve recognition, trust and the customer’s willingness to choose the brand over time. Demanding perfect attribution may encourage management to cut a valuable relationship-building capability because its benefits are difficult to isolate.

That objection is fair. The answer is not to pretend every benefit can be measured precisely. It is to separate what is known from what is believed.

Matched customer groups, phased rollouts, offer tests and changes in behaviour before and after enrolment can all strengthen the evidence. Qualitative feedback can explain why customers use a benefit. Brand measures can capture part of the longer-term effect. None is perfect, but together they are better than treating membership as proof.

A board can accept uncertainty. It should not accept a category error.

Five questions worth asking

When a loyalty programme is presented to a board, five questions bring the economics into view:

1. Which customer behaviour are we trying to change?
2. What evidence suggests the programme caused that change?
3. Which customer groups create the incremental value, and which are mainly being subsidised?
4. What is the complete cost of delivering and administering the benefit?
5. If a major partner or reward disappeared, would the customer proposition still make sense?

Membership matters. Redemption matters. Awards matter. They signal that customers have noticed and used what the business built.

But the board’s job is to go one step further. It must ask whether the programme created a stronger customer relationship at a cost the business can sustain.

The number worth celebrating is not how much existing value passed through the programme. It is how much better the relationship became because of it.

Sources

• Sasol, “Sasol Rewards celebrated at 2026 South African Loyalty Awards,” 16 September 2026: Read the source
• Meyer-Waarden and Benavent, “Do loyalty programs really enhance behavioral loyalty?” International Journal of Research in Marketing: Read the source
• American Marketing Association, “The Power of Free Shipping”: Read the source

← Back to writing