Contract Review to Accounting Implications

ASC 606, the FASB's Revenue from Contracts with Customers (Topic 606), organizes revenue recognition into a five-step model: (1) identify the contract with the customer, (2) identify the performance obligations in the contract, (3) determine the transaction price, (4) allocate that price to the performance obligations, and (5) recognize revenue when, or as, each obligation is satisfied. A good read walks each contract through those five steps and notes, at each one, the clause that drives it and the question it raises. Used this way, the model is a lens for spotting issues, not a machine for settling them.

Step two carries much of the work in bundled deals. Under ASC 606-10-25-14, an entity identifies as a performance obligation each promise to transfer a good or service that is distinct, or a series of distinct goods or services that are substantially the same. A performance obligation is that distinct promise, and an arrangement that bundles equipment, installation, and a maintenance plan can hold several. Whether a promise is distinct, meaning the customer can benefit from it on its own or with readily available resources and it is separately identifiable within the context of the contract, changes how the price is split and whether revenue lands at a point in time or over time. Miscounting the obligations is one of the more common ways a read goes wrong, so the count gets checked early.

The short overview below walks through the same five steps.

Step three, the transaction price, is where the second recurring risk sits. Variable consideration is any part of the price that is not fixed: usage fees, royalties, rebates, refunds, service credits, price concessions, performance bonuses, penalties, and similar terms. ASC 606-10-32-5 through 32-13 set out how to handle it. The amount is estimated, using either an expected-value or a most-likely-amount approach depending on which better predicts the consideration the entity expects, rather than booked at a guess or assumed at its optimistic peak.

The estimate is then held back by the constraint. Variable consideration is included in the transaction price only to the extent it is probable that a significant reversal in the cumulative revenue recognized will not occur once the underlying uncertainty is resolved. The point of the constraint is to keep a firm from recognizing revenue it may later have to reverse, which is why a usage fee or a royalty is estimated cautiously. A sales-based or usage-based royalty tied to a license of intellectual property can carry its own recognition timing linked to the underlying sale or usage, so it is worth flagging on sight. Spotting variable consideration and framing it as an estimate-and-constrain question, rather than a fixed figure, is a large part of a careful read.

The Big 4 revenue handbooks (KPMG's Handbook: Revenue recognition, PwC's Revenue from contracts with customers guide, and Deloitte's Roadmap: Revenue Recognition) are where the working discipline is spelled out, and a few habits recur across them. Read the whole arrangement, not the base contract alone, because a side letter, an amendment, an order form, or an exhibit can change a term the base document appears to settle. Anchor each accounting point to the specific contract language, quoted rather than paraphrased, so a reviewer can trace the conclusion back to the words on the page. And treat a first-pass read as a set of issues to resolve, not conclusions to file, since the recognition judgment belongs to the people who sign the memo.

Only after the work is understood does the AI question arise, and here the answer leans favorable for one reason: the bulk of this task is language, not arithmetic. Pulling the clauses that drive recognition out of dense prose and organizing them into the five-step frame is extraction, exactly the kind of reading a model does quickly. The judgment, deciding the actual treatment, stays with the accounting team. That split, fast extraction on one side and human judgment on the other, is the design the rest of the module builds on.

The most useful thing you can hand the model is an example of the destination. A prior clause-and-implication read the accounting team liked is exactly that: it shows the table shape (clause, verbatim quote, ASC 606 step, the issue it raises, a confidence rating, and any open questions), the level of detail, and the restraint of framing issues rather than conclusions. With that example in the folder, the model reproduces a known shape rather than inventing a format.

The journey is clear at both ends. You start with contracts you have not yet mapped to accounting terms, and you are going to a read the team can turn into memos. The worked example anchors the far end so the output arrives in the form the team already reviews.

Two parts of this work are repeatable, and both can come off the judgment path. Pulling clauses and organizing them into the five-step frame is the extraction, and a model handles it quickly. The small arithmetic that sizes a variable term, such as the dollars a usage tier produces at a given volume, is deterministic, and it belongs in a formula or a line of code rather than in prose a model composes from memory. Deciding the accounting treatment is the judgment part, and it stays with the accounting team. The split is what keeps the workflow honest: the model drafts issues and evidence, a formula does the arithmetic, and the team owns the answer.

Inside the extraction, one habit does the heavy lifting: quote the clause verbatim. A paraphrase can soften a firm acceptance condition into a general one, or drop a word that changes the timing, and the accounting can shift with it. Quoting the exact language keeps each implication anchored to the contract, so a reviewer can check the read against the source in seconds. Each item also gets a confidence rating, which tells the team where to spend their attention.

Take the pattern through a single clause from the Northwind Robotics SaaS contract so the abstract shape becomes concrete. The usage clause reads, verbatim: "For each calendar month in which Customer's API calls exceed 1,000,000, Customer shall pay $0.002 for each additional call." Copying those exact words, rather than summarizing them, is step one, so a reviewer can check the read against the page in seconds.

Step two maps the clause to the five-step model. The charge sits in step three, determine the transaction price, and it is variable consideration, because the amount owed depends on usage that has not happened yet. The read frames the implication as an issue, not a conclusion: how should the accounting team estimate the monthly usage, and does the constraint hold that estimate back? The read surfaces the question rather than answering it.

A number shows why the framing matters. In a month that runs 1,600,000 API calls, the charge applies only to the 600,000 calls above the threshold: 600,000 times $0.002 is $1,200 of usage revenue, on top of the fixed $12,000 platform fee. Shift the usage forecast and that $1,200 moves with it, which is exactly why the read flags an estimate instead of booking a fixed figure. The last step is the confidence rating: medium, since the mapping to variable consideration is clear while the estimate itself rests on a usage forecast the accounting team should own. One clause, quoted verbatim, mapped to its ASC 606 issue, sized, and rated: that is the whole pattern in miniature.

The lab folder stays deliberately small: the three contracts, a short file of the ASC 606 step definitions to apply consistently, and one worked example of a finished read. Nothing else. A larger pile of side agreements or unrelated schedules would bury the clauses that actually drive recognition and raise the chance the model latches onto something irrelevant. Minimum context is the fuel here, and it doubles as a governance control, since fewer files in the folder means less sensitive material in the tool.

The setup is worth saving so the next batch of contracts does not start from scratch. Two scaffolds carry most of the reuse. The first is a reusable prompt, or a packaged skill in a tool such as Claude, that states the destination once: extract the recognition-relevant clauses, quote each verbatim, map it to its ASC 606 step, frame the implication as an issue, size any variable term with a formula, and rate confidence. Written once and reused, it holds the read to the same shape each time and keeps the "issue, not conclusion" discipline in the instructions rather than in your memory.

The second is a short review checklist for the human close, run before anything reaches the accounting team: does each quote match the contract word for word; is each implication framed as an issue rather than a conclusion; is each variable-consideration term flagged as an estimate-and-constrain question; and does a confidence rating sit on each item. The checklist stays deliberately mechanical, because the value of the human step is catching the quote that drifted or the clause the extraction missed, not re-deriving the read from scratch.

Download the folder and run the extract-and-rate pattern in whatever AI you use. The folder holds three signed contracts (a SaaS subscription with Northwind Robotics, an equipment-plus-service bundle with Cascade Assembly, and a software license with Pinnacle Tooling), a short file of ASC 606 step definitions, and one worked example of a finished read. Pull the accounting-relevant clauses, quote each one verbatim, map it to its ASC 606 issue, assign a confidence rating, and gather the open questions. Then come back for validation. The companies and the figures are fictional.

The heart of validation is fidelity to the source. Take each quoted clause in the read and confirm it matches the contract text word for word; a single altered clause is enough to send the read back. Confirm, too, that each implication is framed as an issue rather than a conclusion, that each item carries a confidence rating, and that the variable-consideration terms (the usage charges, the royalties, the credits, and the refunds) are caught and flagged. Work the checklist below against your read before you would hand it to the accounting team.

The read below is what you would hand the accounting team. Each item quotes the clause verbatim, states the ASC 606 point as an issue rather than a conclusion, and carries a confidence rating that tells the team where to look first.

Northwind Robotics.

• "Customer shall pay a platform fee of $12,000 per month for access to the platform." Issue: the monthly platform fee looks like consideration for a single, ongoing access obligation recognized over the subscription term. Confidence: high.
• "For each calendar month in which Customer's API calls exceed 1,000,000, Customer shall pay $0.002 for each additional call." Issue: the usage charge is variable consideration; the question is how to estimate it and whether a usage-based approach applies. Confidence: medium.
• "Customer shall not receive access to the production environment until implementation is complete." Issue: with a $30,000 implementation fee and access gated on completion, is implementation a distinct performance obligation, or is it so bound up with platform access that it is not separable? Confidence: low.

Cascade Assembly.

• "Title to and risk of loss for the Equipment shall pass to Buyer upon shipment." Issue: the $480,000 equipment appears to be a performance obligation satisfied at a point in time, on shipment; the $60,000 installation and the $90,000 three-year maintenance look like separate obligations with different timing. Confidence: medium.
• "Buyer shall withhold 20% of the total contract price until written acceptance of the installed Equipment." Issue: the acceptance holdback raises a question about the timing of recognition and about whether acceptance is a formality or a substantive hurdle. Confidence: medium.

Pinnacle Tooling.

• "Licensee shall pay a royalty of 3% of Licensee's revenue derived from products manufactured using the Software." Issue: the royalty is variable consideration, and a sales-based royalty on a license of intellectual property may carry its own recognition timing. Confidence: medium.
• "If total royalties in any annual period exceed $200,000, Licensor shall refund 15% of the fixed license fee for that period." Issue: the conditional refund is variable consideration that reduces the transaction price and needs estimating and constraining. Confidence: medium.
• "Licensor shall provide software updates on a when-and-if-available basis for three years." Issue: are the three years of updates a separate performance obligation to allocate part of the $150,000 fee to? Confidence: low.

The lab is clean by design: three self-contained contracts, a short definitions file, and no side letters. Real contracts are messier, and the extraction breaks in a few recognizable ways.

• The overriding side letter. An amendment or side agreement can flip a term the base contract appears to settle, such as a most-favored-nation pricing clause or a termination-for-convenience right, so a read of the base document alone can be confidently stale.
• The term buried in an exhibit. Acceptance criteria, service-level credits, or a renewal-and-escalation clause often live in an order form or a schedule the model treats as reference material and skims, and losing one of them can move both the timing and the amount of revenue.
• Boilerplate mistaken for substance. A long entire-agreement or indemnity clause can pull the model's attention while a one-line auto-renewal or price-escalation term does the real revenue-recognition work, so the read over-explains the routine and under-weights the clause that matters.

The model can also miss a buried clause or over-read a routine one, which is why the verbatim check and the variable-consideration flag are worth running deliberately rather than assuming the extraction caught everything.

The workflow does not remove the accounting team's judgment; it removes the mechanical drudgery of finding and organizing the clauses so the judgment has more room. The conclusion, and the audit trail behind it, stays with the team.

Rate your own read against the rubric below. An honest score shows which parts of the workflow you have mastered and which still need practice. Your scores roll up to the workflow maturity dashboard on the course hub.