Can AI Help With ASC 450 Loss Contingencies? It Can Track the Matters, and It Can't Call the Odds

Across the 9,615 live staff accountant, senior accountant, accounting manager, director of accounting and controller postings in our database, 4,288 mention month-end and 10 mention loss contingencies, ASC 450 or a legal reserve. That's roughly 45% against 0.1%, and eight of those ten are senior accountant roles, which tells me this work is landing on mid-level people who were mostly never hired for it.

I think that gap is a real opportunity, because loss contingencies are one of the few places on the close checklist where a missed item can turn into a restatement-flavored conversation with your auditor, and most teams track them in an email folder and a prayer. AI is pretty good at the tracking half of that problem, and it has no business making the other half.

What does ASC 450 actually ask you to decide?

You accrue a loss when it's probable that a liability has been incurred and you can reasonably estimate the amount, you disclose when it's reasonably possible, and you do neither when it's remote. Deloitte's roadmap on ASC 450 recognition lays out the two accrual criteria and defines the three terms, and it notes the guidance gives no numeric thresholds, so the line between probable and reasonably possible is a judgment you document and defend.

People often read probable as something like 70% and remote as 10% or less because that's common practice, but the codification doesn't say so, and an auditor who disagrees with your read won't care what the rule of thumb was. Every pending matter on your list gets sorted into one of three buckets, and the sorting is the job.

Why can't AI call probable, reasonably possible or remote?

Because the facts it would need live in a lawyer's head and a courtroom, and the lawyer isn't going to give them to you the way you'd like. The ABA's Statement of Policy on audit responses, as WilmerHale summarizes it, says lawyers will generally assess the likely outcome only when they conclude an unfavorable result is probable or that it's remote. The big middle, where the outcome is reasonably possible, is exactly where your disclosure decision sits, and counsel will mostly decline to give you a number there.

So the person holding the pen on that footnote is you, working from a letter that may say very little. A model can't conjure a settlement probability out of that, and if it hands you one with a straight face, treat it like the bullshit it is, because a confident percentage built on a vague legal letter is worse than an honest blank.

What can AI do on a loss contingency?

It can run the inventory, the roll-forward and the first draft of the disclosure, which is most of the hours. Start with the matter list. Give it your current list, the prior-year legal letter responses, the last two quarters of board minutes and the outside counsel invoices from AP, and have it flag every matter that shows up in one place and not another. A law firm vendor with fresh invoices and no matching line on your list is exactly the kind of loose thread that blows up in fieldwork.

Next is the accrual roll-forward: beginning balance, new accruals, payments, releases, ending balance, tied to the GL account with every break flagged. It's the same pattern we covered in rolling reconciliations and the AI-assisted close, pointed at a reserve instead of cash. Then the legal letters, where AI can compare this year's responses to last year's and highlight the sentences that changed, since a new phrase like unable to evaluate is the thing you'd otherwise find at 11pm.

Last, the disclosure skeleton. Give it your policy language and the matters you've already classified and it'll draft the nature of the contingency, the range of possible loss where you've estimated one, and the statement when you can't, so you edit rather than start from a blank page. If you've already built a workflow like the one in our audit prep and PBC list piece, the matter list slots in as one more schedule.

Where does AI get contingencies wrong?

It treats absence of evidence as evidence of absence. If a claim hasn't been asserted yet, it won't be in your invoices, your letters or your minutes, and a model working from those documents will tell you the list is complete when it isn't. It also blurs the line between a pending claim and an unasserted one, which matters because counsel's obligations around unasserted claims are narrower and the disclosure analysis changes.

It will also happily average a range, and the accounting under ASC 450 doesn't work that way when one number in the range is a better estimate than the others. That's a call for you and the people who understand the underlying facts, and hell, sometimes it's a call for a phone conversation with the general counsel that no document can replace.

What I'd do

I'd build the matter inventory first and leave the classification alone for a quarter. Get the AI cross-check running against AP, legal letters and minutes, review every flag by hand, and see how many matters you had been missing. If you're a senior accountant who ends up owning this, that's the kind of project that gets noticed, and remote senior accountant roles are where we see most of those rare ASC 450 mentions.

Then write down your own threshold language for probable, reasonably possible and remote, get your controller to sign off on it, and have the AI apply your wording consistently while you review the exceptions. For the written-up version of that judgment, how to use AI on a technical accounting memo covers keeping the conclusion yours. And if you're still picking tools to run all this on, the accounting software directory is a decent place to start.

Frequently asked questions

Can AI determine whether a loss is probable under ASC 450?

No, and it shouldn't try. Probable, reasonably possible and remote are judgments with no numeric thresholds in the guidance, and they depend on legal facts that counsel mostly won't quantify in the middle range. AI can organize the evidence, and you make the call.

What is the difference between probable, reasonably possible and remote?

Probable means the future event is likely to occur, reasonably possible means the chance is more than remote but less than likely, and remote means the chance is slight. You accrue when a loss is probable and estimable, disclose when it's reasonably possible, and generally do neither when it's remote.

What should AI do with legal letters from outside counsel?

Compare them to last year's responses and to your matter list, then flag changed language and matters that appear in one source but not the others. It should not summarize them into a conclusion about likelihood.

How many accounting jobs mention loss contingencies?

In our data, 10 of 9,615 staff accountant through controller postings mention loss contingencies, ASC 450 or legal reserves, about 0.1%, as of November 2026. Month-end appears in roughly 45% of the same postings.

Is it safe to put legal matter details into an AI tool?

Check your company policy and your counsel's view first, because privileged material can lose protection if it's shared with the wrong vendor. Start with non-privileged inputs like vendor names, invoice data and your own roll-forward, and keep the legal letters inside tools your company has approved.

Pick one quarter, build the matter cross-check, and see what it finds that your email folder didn't, because that finding is the proof you'll want when you pitch doing more of this with AI.