AI is making bespoke systems easier to build, just as the way those systems are built is beginning to change.
AI is making it possible to develop supply chain systems that would previously have required considerably more specialist software development. Forecasting models, inventory tools, planning applications and decision-support systems can increasingly be developed around a particular business requirement, with AI assisting with everything from calculations and business logic to coding and interfaces. For supply chain teams that have spent years adapting general software to quite specific operational requirements, the attraction is obvious.
There is, however, a complication. The same technology making these systems easier to build is developing so quickly that the way they are built may change significantly during their useful life. A business can therefore make a perfectly reasonable decision to build an AI-enabled system today and still find that much of the technology surrounding its underlying methodology becomes unnecessary surprisingly quickly.
That does not make bespoke development a mistake. It does suggest that businesses should be careful about how much technology they build around a requirement when the mechanisms available for delivering that requirement are still changing.
Building has become easier
The traditional decision between buying and building business software involved a reasonably familiar trade-off. Bespoke development offered the opportunity to create something closely aligned to the business, but it brought development, testing, implementation and ongoing maintenance requirements. Established software brought constraints of its own, but much of the technical development and support sat with the provider.
AI changes that balance because it can substantially reduce some of the barriers to creating bespoke capability. A supply chain team with a clearly defined requirement can increasingly contemplate developing tools that would once have needed much greater technical resource. That creates genuine opportunities, particularly where established systems do not deal particularly well with a specialist planning or operational problem.
The important distinction is that something becoming easier to build does not necessarily mean that more should be built around it. There is a risk that businesses use AI to accelerate a familiar approach to software development at precisely the point when AI itself is starting to offer different ways of delivering the same capability.
The question is no longer simply whether a bespoke system can be built economically. It is also whether all of the application surrounding the underlying requirement will still be necessary as the AI platforms develop.
Separate the capability from the system
An inventory analysis provides a useful example. The important elements are not necessarily the screens through which somebody uploads data or the interface through which the results are presented. The real value lies in understanding how demand and lead-time behaviour should be analysed, which inventory policies are appropriate, how parameters should be calculated, what assumptions need to be challenged and how the resulting recommendations should be interpreted.
Traditionally, delivering that capability repeatedly might have encouraged the development of a dedicated application. That may still be entirely appropriate, particularly where the process needs substantial integration, control or automation. Increasingly, though, AI creates the possibility of separating the valuable methodology from much of the technology used to deliver it.
A methodology might be capable of being delivered through a relatively lightweight AI capability today and through something different again as the technology develops. The interface, workflow and underlying AI may change considerably while the supply chain reasoning remains largely intact.
That makes the methodology, business rules and decision logic potentially more durable than the application built around them.
This is an important distinction for businesses starting to invest in AI-enabled supply chain systems. The question should not simply be how the requirement can be turned into software, but which parts of the solution genuinely need to be built and which can reasonably be left to an AI platform that will continue to evolve.
Today’s constraints may not last very long
Every system architecture reflects the technology available when it is designed. Development effort is often spent solving limitations that exist at that point in time, whether that involves creating interfaces, building workflows, transforming data or writing code to perform activities the underlying platform cannot manage directly.
In a relatively mature technology environment, that can be a reasonable long-term investment. AI presents a different problem because some of those limitations are moving quickly. An architectural decision that is entirely sensible today may become unnecessary as the underlying tools become more capable and adaptable.
This is where emerging approaches such as reusable AI skills are particularly interesting. Rather than assuming that every specialist requirement needs its own permanent application, a business can increasingly consider whether the knowledge, instructions, analysis and decision process can be separated from the surrounding system. The significance is not any particular AI product or feature, since those will themselves continue to change, but the broader move towards smaller, more adaptable components of capability rather than a bespoke application for every requirement.
This does not mean that applications are disappearing, or that complex supply chain requirements can simply be handed over to an AI assistant. The more important point is that businesses should avoid building permanent solutions around temporary technological limitations where a more adaptable architecture will do the job.
Invest in what is difficult to replace
For supply chain applications, the most valuable element of a bespoke solution is often not the software at all. It is the knowledge of what the system needs to do.
A forecasting system still needs an appropriate forecasting methodology. An inventory model needs to understand the relationship between demand, supply, lead time, service and stock. A replenishment tool needs suitable business rules. A planning application needs to reflect the decisions the business actually needs to make rather than simply generate mathematically interesting outputs.
AI can make it easier to turn those requirements into working technology, but it does not remove the need to define them properly. Indeed, if the technology surrounding the requirement becomes increasingly interchangeable, the quality of the underlying supply chain thinking becomes more important rather than less.
That should influence where development effort is concentrated. Time spent understanding the requirement, defining the methodology, challenging assumptions, validating calculations and establishing appropriate business rules is unlikely to be wasted simply because a new AI capability becomes available. Large amounts of bespoke code and infrastructure created only because today’s technology requires them may be less durable.
Keep the technology adaptable
This suggests a useful principle for AI-enabled supply chain systems: start with the requirement and then use only as much technology as is necessary to deliver it reliably.
For some problems, that will still mean developing a substantial bespoke system. Integration with operational systems, repeatable high-volume processing, specific controls or a particular user experience may make a dedicated application entirely appropriate. The pace of AI development is not a reason to avoid building systems that have a clear business justification.
For other requirements, however, a lighter approach may provide greater flexibility. If the underlying capability can be separated from the technology used to deliver it, the business has more freedom to take advantage of future developments without having to replace an entire system. The objective is not to predict which AI technology will eventually prevail, but to avoid unnecessary dependence on decisions made around today’s capabilities.
That changes the emphasis of AI development. Rather than asking how much can now be built, businesses should consider how much actually needs to be built and which parts of the solution need to survive the next change in technology.
AI will undoubtedly enable increasingly sophisticated supply chain systems. Some will justify considerable investment and remain important applications for many years. Others may turn out to have recreated functionality that emerging AI platforms subsequently make much simpler to deliver.
The risk is not that AI makes today’s supply chain thinking obsolete. It is that businesses build too much technology around that thinking before discovering they no longer need it.