AI is easy to start. The harder question is whether the business is ready to benefit from it

By Jeandré du Randt, Chief Business Environment Officer, 4Sight.
Johannesburg, 19 Nov 2026
4Sight's Jeandré du Randt at the AI Vision event.
4Sight's Jeandré du Randt at the AI Vision event.

AI may be one of the easiest major technologies businesses have ever been able to experiment with. But experimenting with AI is easy. Changing how a business performs because of AI is much harder.

That was one of the strongest themes for me coming out of our recent From AI Vision to Frontier Business event.

An excellent use case can be undermined by poor data, great data by an inefficient process, and a capable agent by systems it cannot safely reach or by people who never embed it in their work.

The real AI challenge is therefore not simply technological. It is organisational.

We should become much harder on AI ideas

Businesses are not short of AI ideas. The harder discipline is choosing.

The event's value architecture session made the point that each idea should pass three tests: a credible business case, an organisation ready enough to execute it, and somebody who owns the resulting benefit. Funding should then be staged – prove, pilot, scale – with clear stop points, so that an idea that does not deliver is stopped before it becomes an expensive programme.

I would rather have three AI initiatives where we can clearly say: “This is the problem, this is the baseline, this is what must improve and this person owns the outcome,” than 30 interesting pilots looking for a business case.

At 4Sight, the starting point is not a particular piece of technology but understanding where value is being lost or where new value could be created – slow turnaround, skilled people stuck on routine work, decisions delayed by scattered information, operational issues spotted too late.

Only then do we ask what combination of data, process redesign, intelligence, automation, integration and change is required to improve that outcome.

Intelligence needs something trustworthy to work with

AI can make information easier to access, interpret and act on. But it does not magically make the underlying information correct.

A simple example makes the point. An executive asks an AI assistant how sales are going and is told they are up 13%. A salesperson then points out that the 13% is the annual price increase: there are no new deals in the CRM. The data was there; what it meant had never been checked. Available data is not necessarily AI-ready data.

This is where the foundational work sits: bringing information together across systems, with common definitions, ownership, quality controls, appropriate access and clear governance.

That does not mean cleaning every piece of data before doing anything with AI. Start with the data a meaningful use case needs, fix what prevents the outcome, put the controls in place and prove the value. We call this a “thin slice”: choose, assess, build and prove one meaningful outcome within 90 days.

Solve enough of the foundation to deliver the outcome, then allow each successful outcome to strengthen the foundation for the next one.

Start where your workloads are

The same applies to infrastructure. Organisations do not have to finish their cloud journey before they can start with AI. Many still run important systems and data on-premises, and regulated industries such as banking and insurance must keep certain data and models there. The practical question is the right destination for each workload – retain, connect, modernise, move or retire – and then a governed bridge between them: identity, interfaces, monitoring, security and human review. Hybrid is not a compromise. For many organisations it is the fastest responsible route to a first outcome.

AI becomes more valuable when it moves from answering to doing

The first wave of AI helped us create and understand things faster. The next wave is about getting work done.

An AI capability might identify an exception, interpret information, prepare a recommendation or determine the next appropriate step. But business value often only occurs when that insight connects into an approval, transaction, system update or workflow.

That is where AI and automation meet: agents can reason, automation can execute, and people remain accountable for the decisions that matter.

Think about supplier onboarding that continuously checks for missing information. Finance processes where exceptions arrive ready for review rather than someone first having to find them. Maintenance information that moves from identifying a potential problem into the workflow required to act on it.

These are not AI features; they are redesigned business processes, and that is where businesses should be aiming.

The biggest opportunity sits where business systems meet operations. On a plant, a mine or a refinery, an insight is only valuable if it reaches the people and systems that run the operation. That is why we think of our role as building a bridge between IT and OT rather than talking about convergence: each side keeps its own disciplines, safety requirements and owners, while data and decisions flow across in a governed way. In industrial businesses, much of the value of AI will be won or lost on that bridge.

We are redesigning work, not simply introducing technology

There is a tendency to frame AI adoption around how many people have licences, how frequently they log in or how many tasks they complete with AI.

Those measures tell us something, but the more important question is: has the way work happens actually changed?

The objective is not to automate as much human work as possible; it is to create more capacity for human value.

That requires organisations to redesign roles and processes deliberately. Simply putting new technology beside an old way of working rarely produces transformation.

Adoption is a business measure

At the 4Sight event, we described adoption as a progression from available to relevant to repeated to valuable. The important stages are the middle ones: does the capability make sense to the role and has its use genuinely become part of the workflow?

That is why we talk about verified adoption.

Establish the baseline before you begin, understand which behaviours should change, enable people in the context of their actual work, and then show evidence that the new behaviour is sustained and the business outcome is moving.

Licence counts cannot tell you that and neither can a successful demonstration.

Ultimately, if AI was introduced to improve turnaround time, reduce rework, create additional capacity or improve decision-making, we should be able to measure whether that happened.

The Frontier Business will not be built in one transformation programme

The businesses that get the most value from AI will not necessarily be those that move first or buy the most.

They will be the ones that become good at turning business problems into repeatable, measurable AI-enabled outcomes.

That means architecting value before investing, building trusted data foundations, choosing the right home for each workload, connecting intelligence into the processes – and the operations – where work happens, governing agents so they act without removing accountability, and preparing people to work alongside them.

The same discipline is applied inside 4Sight: what it builds for customers is first put to work on its own processes.

A Frontier Business is not a destination where the organisation suddenly becomes “AI-powered”. It is a business with the discipline to move repeatedly from problem to intelligence to action to measurable value.

So the question leadership teams should ask is not: “How much AI are we using?” but: “Where has AI measurably changed the way our business performs?”

If there is a good answer to that question, you have something worth scaling.

If there is not, the next step probably isn’t more AI.

It is getting the foundations right.

Share

Editorial contacts

Jacqui Scorgie
Group Manager: Marketing
(+27) 83 632 2209
Jacqui.Scorgie@4sight.cloud