Revenue recognition is one of the most judgment-heavy areas of accounting. Here's how AI can help — and where it can't.
ASC 606 is one of the most consequential standards ever issued by the FASB. It replaced decades of industry-specific revenue guidance with a single, principles-based framework. And in doing so, it put a significant amount of interpretive burden on accountants.
The five-step model sounds clean. Identify the contract. Identify the performance obligations. Determine the transaction price. Allocate it. Recognize revenue. Five steps.
But anyone who has actually worked through a complex revenue recognition question knows that the difficulty isn't in understanding the steps — it's in applying them. What qualifies as a distinct performance obligation? When is variable consideration "probable" enough to include? How do you handle a contract modification that changes both the scope and the price?
These are judgment calls. They require reading the guidance carefully, understanding the principles behind it, and reasoning through the specific facts of your transaction. There's no algorithm that spits out the answer.
That's also why this is an area where AI can genuinely help — if you use it right. I spent years working in revenue recognition, including time at a company specifically focused on ASC 606 software and implementations. Here's what I've learned about where AI earns its keep and where you have to be careful.
Where ASC 606 gets hard
Before talking about AI, it's worth being specific about where 606 tends to generate the hardest questions. These are the areas where the guidance requires the most interpretation, and where the stakes of getting it wrong are highest.
Performance obligation identification. Whether two promises in a contract are a single performance obligation or two distinct ones often comes down to specific facts about whether the goods or services are "capable of being distinct" and "distinct within the context of the contract." These are highly fact-specific determinations that depend on the nature of what you're selling and how your contracts are structured.
Variable consideration. The standard requires you to estimate variable consideration — discounts, rebates, refunds, performance bonuses — and include it in the transaction price to the extent it's "probable" you won't have a significant reversal. Determining what's probable is a judgment call that requires understanding your historical experience, your customer relationships, and the specific terms of the arrangement.
Contract modifications. When a contract changes, 606 gives you three ways to account for it, and which method applies depends on whether the modification adds distinct goods or services and whether the pricing is at standalone selling price. Getting this wrong can result in revenue being recognized in the wrong period — which is a material error.
Principal vs. agent. If you're involved in a transaction where someone else also plays a role in fulfilling the obligation, you need to determine whether you're the principal (recognize gross revenue) or an agent (recognize net). This analysis depends on which party controls the goods or services before they're transferred, which is often not obvious.
These are the kinds of questions where technical accounting research matters most — and where AI has the most to offer.
How AI helps with ASC 606 research
Getting oriented on complex questions
When you're facing a new or unusual revenue recognition question, one of the hardest parts is just figuring out where in the codification to focus. ASC 606 is extensive, and there's also ASC 340-40 (costs to obtain and fulfill contracts) and various implementation guidance and illustrative examples to consider.
AI is good at orientation. Describe your fact pattern and ask what aspects of 606 are most relevant — what sections you need to read, what the key judgment calls are, what the illustrative examples address. That map is faster to get from AI than from navigating the codification cold.
Working through the five-step model
One practical use of AI in 606 work is running through the five-step model systematically against your specific facts. This sounds mechanical, but it's easy to skip a step or give a step less attention than it deserves when you're moving quickly.
A well-designed AI can walk through each step against your transaction, flag where the analysis is straightforward and where it requires more work, and help you make sure you haven't missed anything before you start writing the memo.
Analyzing specific judgment areas
The areas I listed above — performance obligations, variable consideration, contract modifications, principal vs. agent — each have substantial codification guidance and illustrative examples. AI that's been built with accounting knowledge can help you work through the relevant guidance systematically, surface the examples that are most analogous to your situation, and help you think through the analysis.
This isn't the AI making the call. It's the AI helping you structure the thinking before you make the call.
Drafting documentation
Revenue recognition memos tend to follow a consistent structure, and AI is good at generating that structure. Once you have your analysis worked out, AI can help you document it in a way that's organized, clearly reasoned, and complete — covering the question, the relevant guidance, the analysis, and the conclusion.
Where AI falls short on ASC 606
Novel contract structures
ASC 606 is a principles-based standard, which means the guidance doesn't have a specific answer for every possible contract structure. When you're dealing with a genuinely novel arrangement — a new business model, an unusual combination of goods and services, something your industry hasn't encountered before — AI is less reliable.
The model is trained on patterns. If your transaction doesn't fit those patterns, you're more likely to get a response that sounds right but misses the specific complexity of your situation. This is exactly when you need to slow down and rely on your own judgment, potentially with input from a technical accounting specialist.
Industry-specific nuance
Some industries have significant industry-specific guidance that overlays the general 606 framework — software, real estate, financial services, construction. If your revenue recognition question implicates industry-specific rules, make sure the AI you're using is aware of them. Generic AI tools often miss industry nuance.
The final call on significant estimates
Variable consideration estimates, standalone selling price allocations, probability assessments — these require judgment that goes beyond applying the standard. AI can help you structure the analysis, but the estimate itself has to be yours, grounded in your knowledge of the business and the facts.
A practical workflow for ASC 606 research
Here's how I'd approach a complex 606 question using AI:
Start with the facts. Write out the key terms of the arrangement before you do anything else. Who are the parties? What's being promised? How is pricing structured? Are there options, modifications, or variable elements? The quality of your AI-assisted research depends heavily on how well you've articulated the facts.
Ask AI to map the relevant guidance. Based on your facts, what aspects of 606 are most relevant? What sections should you read? What illustrative examples are analogous? Use this to build your research agenda.
Read the actual codification. Don't skip this step. The AI gave you a map — now go read the territory. For any significant judgment call, you need to read the actual guidance, not just the AI's description of it.
Work through each step of the five-step model explicitly. Even if some steps are simple, be explicit about them. Document your analysis at each step, not just at the conclusion.
Draft the memo. Use AI to help structure the documentation. Review it carefully — the application to your specific facts is where AI output is most likely to need work.
Check your conclusion against the facts. Before you finalize anything, re-read your statement of facts and make sure your conclusion actually follows from them. This sounds obvious, but it's the step that catches the most errors.
Revenue recognition is hard. That's not going to change. But the research and documentation work around it doesn't have to be as slow and manual as it has been. Used carefully, AI can make a real difference in how quickly you get to the right answer — and how well you document it.
Gaapio was built with ASC 606 as one of its core areas of focus — it handles the five-step model, surfaces the relevant illustrative examples, and structures output for memo documentation. If revenue recognition comes up in your work, it's worth seeing how it handles a real question.