Read AI Summary
For years, the economics of the technology workaround made sense.
A core system could not support a new experience, so a portal was added. Systems could not communicate, so an integration layer connected them. A process was too manual, so workflow and automation were introduced.
None of these decisions was wrong. Each solved a real business problem while protecting the policies, operations and institutional knowledge already in place. Technology also moved slowly enough that these solutions stayed useful for several years. The business captured value. Complexity accumulated gradually and, for the most part, stayed manageable.
That equation is changing.
AI capabilities are advancing faster than the technology cycles insurers have traditionally managed. New use cases appear before many carriers have moved their first experiments into production. Customer and distributor expectations are moving at the same pace.
The industry’s response looks familiar. Put an AI copilot above the legacy platform. Add orchestration to navigate across systems. Use agents to interpret requests, call the right applications and coordinate the work.
There is real value in this. An agent can help a service representative understand a contract or identify the next action, and orchestration can connect processes that span several systems. But it is worth being precise about what has actually changed. If products remain hard-coded, they are still hard-coded. If the same rule exists in four systems, an orchestration layer does not turn it into one rule.
Most of the AI conversation has been about intelligence. Can a model understand intent, interpret a policy, recommend the next action, design a more relevant journey? Increasingly, the answer is yes.
The harder question starts after the recommendation. Can the insurer act on it safely? Does the product capability exist, is the rule consistent across administration, distribution and service, and can the change be made through configuration rather than development?
The recommendation takes seconds. Acting on it can still take weeks or months.
The constraint is shifting from whether AI can identify the right action to whether the insurer can execute it safely and at scale.
Executability can be measured, and the measure is uncomfortable. Count how many parties have to agree before one product, one regulatory change or one AI use case can go live. If the answer is four stakeholders and three release windows, that number is the ceiling on what any AI investment can return, however good the recommendation is.
The gap is easiest to see at the working level. A product manager is asked whether a rider can be added to a product in two states before the end of the quarter. Answering that requires knowing which version of the product each state is on, which rules the rider touches, what else depends on those rules and which systems have to change. Assembling the answer takes days, and the answer is usually no. An AI assistant that can draft the rider does not shorten that week.
This is not only an architectural concern. It shows up in three places a chief executive already watches.
The first is premium. A rate you cannot publish is business a competitor writes. When repricing takes a quarter, the market has moved before the change is live, and knowing what to do sooner is worth nothing.
The second is the cost of regulatory change. When the same rule lives in several systems owned by several vendors, one rule change becomes several projects, each with its own interpretation, testing and release schedule. The expensive part is not the rule. It is the coordination.
The third is what each AI layer adds to the compliance burden. Every AI capability attached to a different system adds another surface that must be tested, monitored and defended at examination. As use cases multiply across a fragmented estate, the governance burden can multiply with them, and that cost recurs every year rather than once.
Insurance technology evolved in layers because replacing systems that administer long-duration promises is genuinely difficult and risky. That has not changed. What has changed is the speed at which new capability can now be created.
AI makes it easier to build workflows, interfaces, integrations and experiences. Without a configurable and governed foundation, it also makes it easier to create fragmentation. An agent working across inconsistent rules does not correct them. It applies them to more transactions and more channels. A workflow built around one product becomes the next project’s dependency.
Yesterday’s workaround usually had years to produce value before it became a constraint. An AI workaround may have a much shorter useful life. That changes the arithmetic. Insurers can now accumulate complexity at the same speed at which AI promises to create value.
So the question is not whether a particular AI use case works. Many will. The question is whether each implementation makes the next one easier, or gives the next one one more thing to work around.
This is where the two approaches separate, and it is not visible in the first project.
If each use case is built as its own workflow and integration, the tenth costs more than the first. There is more to connect, more to reconcile, more to certify and more to regression test. Cost rises with adoption, which is the opposite of what a business investment should do.
If each use case instead adds a capability others can reuse, the second is cheaper than the first and the tenth is close to free. The rule already exists in one place. The product component is already approved. The governance is already there.
Most AI business cases are written for the first use case. The decision that matters is what happens to the twentieth.
Orchestration matters. It coordinates work across systems, routes requests, invokes trusted services and brings people into a process when judgment is required. We rely on it, and so will every carrier running AI at any scale.
The point is not that orchestration is the wrong tool. It works with the capabilities it is given, and the capabilities underneath determine the result. Over hard-coded products, duplicated rules and overnight cycles, it coordinates existing constraints more elegantly. Over versioned, configurable capabilities, its value compounds because every step it calls is something the next process can call too.
That distinction matters because the first kind can look like modernization. The experience becomes smoother while the cost and difficulty of change underneath stay much the same. That can be a reasonable interim choice, and extending the useful life of a legacy platform is often pragmatic. The problem starts when an interim accommodation quietly becomes the long-term architecture.
Systems of record remain essential. They hold contractual truth, maintain policy history and execute transactions that have to be accurate and repeatable. They were not designed for the pace and breadth of change now being asked of them.
AI adds a second kind of system, a system of intelligence, which interprets information, recognizes intent and recommends action.
Between the two there is a missing capability. Insurers need something that turns intent into governed, executable change. Products assembled from approved and reusable components rather than written as code. Rules, rates and agreements versioned, so it is always clear which one applied to which policy. A change published once and consistently across administration, distribution and service. Governance built into how a change is created rather than added after it ships.
This is a different job from orchestration. Orchestration coordinates an action. A system of change makes the underlying business capability adaptable.
AI and orchestration will extend the life of many legacy platforms, and in some cases that is the right decision. But leaders should separate extending the life of a system from extending the life of its limitations.
Before approving the next AI layer, it is worth asking four questions. Does this make the underlying business capability more adaptable, or only easier to reach? Will the next product, regulatory change or customer journey be easier to deliver because of what we build now? Are we reducing dependencies or adding one? Can the outcome be governed consistently across products, channels and policy generations?
If the answers are no, the organization may be adding another workaround rather than building the capacity to change.
AI is exposing the distance between what insurers can imagine and what their operating environments can execute. Orchestration can help navigate that distance. It cannot close it on its own. At some point the industry has to stop making constrained systems easier to work around and start making the business easier to change.
.png)