AI for Revenue Recognition (ASC 606)

ASC 606, Revenue from Contracts with Customers, uses a five-step model: Identify the contract; Identify the performance obligations; Determine the transaction price; Allocate the price to the performance obligations; and Recognize revenue when, or as, each obligation is satisfied. Public business entities adopted it for annual periods beginning after December 15, 2017, with other entities following later. AI is most useful in steps one to three, where the work is reading documents. A promised good or service is a separate performance obligation if it is distinct. That means it is capable of being distinct, so the customer can benefit from it on its own or with readily available resources, and it is distinct in the context of the contract, so it is not highly interdependent with other promises. A model can find promises spread across master agreements, statements of work and order forms. It can also flag terms that need judgment: acceptance clauses; termination rights; options for discounted renewals, which may be material rights; and variable consideration such as rebates, penalties or usage fees. Variable consideration goes into the transaction price only to the extent it is probable that a significant reversal of recognized revenue will not occur. Step four allocates the price in proportion to relative standalone selling prices. Step five decides whether control transfers over time or at a point in time. Both are best handled by rules in a revenue subledger, with AI supplying labeled inputs. The biggest misconception is that the AI "does the accounting." Deciding whether an obligation is distinct, estimating standalone selling prices and applying the constraint remain management's judgments, and auditors will test them. A second limit is completeness. A model cannot find a side letter, email concession or verbal amendment that was never put in the contract repository, and those are exactly the items that often change the accounting.