AI is increasingly being used within varied business processes, from human resources to finance and accounting, and even strategic planning. However, a large proportion of companies are applying AI incrementally to existing processes. For these companies, AI’s real value is in speeding up and improving existing ways of working and delivering minor efficiency gains.
Supply Chains Don’t Fail At The Moment Of Crisis
Predictive analytics changes the game for everyone. It means no more surprise truck failures, because you knew that one was due for maintenance before it left the depot. Or no more “acts of God” holding up a shipment. They fail weeks earlier, when the signals were there but nobody was reading them fast enough.
We like to say that AI puts a logistics manager in the cab of every truck, overseeing every driver, rerouting any that are heading for a backup, and knowing days in advance if any likely is. That sounds expensive, but it’s not. Not compared to the cost of a late shipment. Or the costliest cab of all, the empty truck.
Document-Heavy Departments Are Sitting On Unextracted Value
Legal and finance teams do a ton of work that AI can do quicker and more accurately. Contract review, compliance checks, invoice reconciliation – these are all essentially text analysis problems, and natural language processing is very good at them.
But it’s not just about being faster. Anybody can see that something a thousand contracts uncover is invisible to the human eye reviewing just a few dozen times. Clause risk across your entire vendor portfolio. Small fee structures that wind up compounding their way through a hundred agreements. Obligations that had a technical status of active simply because they were impossible to identify amongst operational systems.
Around 40% of firms say they’ll increase their use of AI because of generative advances in AI. The companies doing that aren’t re-investing in their customer-facing apps. They’re re-investing in their high-complexity, still-only-human machines.
The Next Efficiency Layer Is Quantum-Adjacent
This is the point where how enterprises operate starts to look different. AI isn’t something you plug in to get better results in an existing business process – it changes the fundamental threat model you have to work with. Most enterprises a little way down the AI journey quickly realize their current security stance is inadequate. It doesn’t have to be a showstopper, but it does mean rethinking some fundamental assumptions about your software stack.
The ripple effects go further than security though. The same shift in computing power that makes your infrastructure vulnerable is also unlocking entirely new categories of problem-solving. SandboxAQ is a good example of what that looks like in practice — applying AI grounded in the laws of physics to challenges that standard enterprise software has never been able to touch. From post-quantum cryptography to accelerating drug discovery, modelling chemicals at the atomic level, and precision navigation in GPS-denied environments, it’s a glimpse of where AI meets the physical world in ways that go well beyond the chatbot.
Predictive Maintenance Changes The Industrial Cost Model
Unplanned downtimes are extremely costly for manufacturing and heavy industry. When equipment goes down, it often means a series of production shortfalls and missed delivery schedules. Equipment rarely fails catastrophically without warning, however. It gives off signals of the impending failure that could be used to take remedial action.
AI sensor networks that constantly monitor equipment and build a performance baseline can do that math and give plenty of warning that a critical piece of kit is about to fail. Maintenance can be scheduled based on the warning rather than as an immediate reaction to a critical breakage.
Real-time processing of data at the edge (i.e. on a local server close to the action) is necessary for the same reason. The latency involved in sending data to the cloud, processing, then sending back the instructions is often too long a wait if you want to prevent disaster on the factory floor.
The Human-In-The-Loop Design Question
There are two ways of using AI to enhance efficiency. The first one is by excluding people from making decisions while the second one aims at assisting people in making better decisions. The organizations that succeed in convincing their workforces that AI and its automation can improve their decision-making capabilities embrace the second approach. However, this is not done for cultural reasons but for operational reasons.
Top-level management decisions consider context, interrelations, and risks that can be influenced by AI but must ultimately be handled by humans. If your tech solution was programmed to make sure that human control is part of the decision process – that decisions are monitored but not obstructed by AI – then results are not only less disputable, but they are mostly even better. People and algorithms have a tendency to make an opposite kind of error. As people get better with more data and help with different viewpoints, same goes for algorithms in the context of improving their insights with significant human supervision.
In practical terms that means that you will not only be buying a few AI models in the future, you will need to be thinking about the whole workflow that they will fit into. Who will be monitoring the output? What is their authority level? Where in your existing organization will the output actually fit into processes? These are not IT questions, they are organizational questions, though, and the final answers will decide whether the ROI will make any sense but also whether you will be able to successfully incorporate your new AI into your organization.
