What The Stakes Of AI Look Like In Regulated Environments

What The Stakes Of AI Look Like In Regulated Environments

Ryan Padget, President of IGEN. Ryan helps CFOs transform excise tax compliance from a back-office cost into a financial control advantage.

getty

Most conversations about building versus buying AI capabilities treat the topic as a technology strategy question. What’s the best way to implement AI, and how can it improve efficiency?

But in regulated industries, AI is something different: a risk decision with financial consequences that often don’t announce themselves until they’re already compounding. The calculus changes significantly when the environment you’re operating in has no tolerance for ungoverned outputs, when the decisions your AI tools make must be explainable to a regulator and when a compliance failure shows up on the balance sheet.

Based on my experience working with organizations to evaluate AI builds, here’s what companies and financial leaders in regulated environments should consider before implementing AI into their workflow.

The Accountability Gap Nobody Is Pricing In

There is a version of AI adoption that works well for a period of time—and then doesn’t. A team identifies a problem, builds something quickly using available AI tools and deploys it into a workflow. It handles the task, but nobody formally owns it, and nobody has documented how it makes decisions. Then something changes within the model, the data or the regulatory requirement, and the build fails.

In most environments, that’s an inconvenience. In regulated environments like financial services, it’s a real problem. When material data lives inside an ungoverned application with no clear owner and that application makes decisions that affect compliance obligations, the exposure could become financial and, in serious cases, personal.

As AI takes on more consequential roles in compliance workflows, knowing who is accountable for what those tools produce is no longer a nice-to-have. I’ve found that in most cases, the answer falls on leadership regardless of how well they understand the system producing the output. As a result, the question becomes less about AI’s ability to function in these environments (it can) and more about whether the organization has the domain expertise, the governance structure and the accountability framework to sustain them when something goes wrong.

In my experience, most builds that skip those steps don’t fail immediately; they fail later, when the options for managing the damage are already narrowed.

The Domain Expertise Gap

In excise tax alone, managing compliance across motor fuel and tobacco means navigating more than 10,000 rates across federal, state and local jurisdictions, each with different product categories, license requirements and filing obligations. The legislation governing those obligations is not plain English. Interpretation matters. Judgment calls built up over years of direct exposure to audits and regulatory decisions matter.

AI built without that domain expertise embedded in it will most likely miss the same nuances that spreadsheets also miss, just faster and at greater scale. In other words, leadership may be seeing efficiency gains with an amplified risk from whatever gaps exist in the model’s underlying knowledge.

This is where the build versus buy decision in regulated environments gets its specific character. It is not enough to ask whether an AI tool can automate a workflow; you also need to ask whether the domain knowledge required to make that workflow accurate and defensible is embedded in the system or sitting in the heads of people who may not always be there. When it’s the latter, the build is carrying a hidden dependency that doesn’t show up in the project budget.

Regulators On The Move

As of earlier this year, the IRS has already started laying the groundwork for regulating more pervasive AI use, per the IRS Policy for AI Governance. From what I have seen, use cases like continuous monitoring, automated data-matching and AI-assisted anomaly detection are not far behind, if they aren’t already being used. The window between when a compliance problem occurs and when a regulator identifies it is closing.

Organizations that have built AI-powered compliance workflows without governance in place may soon find that the environment they built for has changed. The filings still go out. The question is whether the logic behind them will be documented, auditable and defensible when scrutiny arrives. In the current regulatory direction, that scrutiny is becoming more likely, not less.

Asking The Right Questions To Drive Your Decision

For regulated environments, I recommend starting the build versus buy decision with a different set of questions than what might be asked in general enterprise contexts:

1. What happens to AI output accountability if the person who built the LLM or the person who understands it leaves? I have seen organizations discover years after the fact that a filing obligation was missed because it existed only in a retiring employee’s institutional memory. When that same dynamic applies to an AI build, the exposure compounds faster and can surface in ways that are harder to attribute.

2. Can the decisions this tool makes be explained to a regulator? Your organization should be able to reconstruct why a specific decision was made, why a particular filing position was taken or why a transaction was treated the way it was, with documentation that holds up under audit. If your systems rely on an AI model whose logic isn’t documented, or on a build whose original developer is no longer available, that’s a liability.

3. What will it cost over time to maintain this AI system as underlying models, the regulatory requirements and the business itself change? From my observations, the cost models being used today typically reflect pricing that is actively subsidized by the platforms providing it, and such pricing will not hold indefinitely as the market matures.

4. Where does the domain expertise live? For example, in excise tax, you should have someone who knows which exemptions apply in a given jurisdiction, how a regulator has historically interpreted an ambiguous product classification or when a filing position needs additional documentation. That knowledge, like other nuanced compliance knowledge, accumulates through years of direct exposure and isn’t easily replicated by AI.

Ultimately, I’ve found that the best foundation for enterprises wanting to adopt AI includes knowing who owns every build, what it costs over its full life and whether its outputs can be defended when the environment demands it. These are the questions that can help determine whether AI creates advantage or quietly accumulates risk.


Forbes Business Council is the foremost growth and networking organization for business owners and leaders. Do I qualify?


Read More

Zaļā Josta - Reklāma