How to Use AI on a Goodwill Impairment Test Without Letting It Choose the Discount Rate

Of the 47,445 accounting and finance postings in our database that carry a classified role and a full job description, 260 mention impairment and 148 mention goodwill. That works out to roughly one posting in 200, as of October 2026, and my read is that it tells you something useful: impairment testing is an annual event that lives in one or two people's heads, gets rebuilt from scratch every year, and almost nobody hires for it specifically. A task like that is about the best candidate for AI help you'll find on the close calendar, as long as you're clear about which pieces you hand over and which pieces stay with you.

What does a goodwill impairment test actually require?

Under ASC 350, you test goodwill at the reporting unit level at least once a year, plus whenever a triggering event suggests the fair value may have dropped below carrying amount. You can start with a qualitative assessment and stop there if it's more likely than not that fair value exceeds carrying amount. If you can't get there, you run the quantitative test, and since ASU 2017-04 that test is a single comparison: if a reporting unit's carrying amount exceeds its fair value, you record an impairment loss equal to the excess, capped at the goodwill allocated to that unit. Deloitte's summary of the change lays out the effective dates and the removal of the old step 2. Private companies have some alternatives available as well, so check which election your entity made before you build anything.

That single comparison sounds simple, and the arithmetic is, but nearly all the pain sits in two inputs: the carrying amount by reporting unit and the fair value estimate. Those are where the hours go, and they're where AI helps in very different ways.

Which parts of the impairment test can AI handle?

Five pieces are mechanical enough that I'd hand them to an agent with a reviewer sitting behind it.

Carrying amount by reporting unit. Give the agent the trial balance, your reporting unit mapping, and last year's workpaper, and ask it to rebuild equity carrying value by unit, flagging every account that moved units since last year. Misallocated shared assets and unallocated corporate debt are the usual culprits, and a tireless reader is good at spotting the ones that quietly changed.

Forecast backtest. Auditors will ask how good your forecasts have been, so ask the agent to compare the projections you used in prior years against what actually happened and summarize the misses by unit. It takes a person a day. It takes an agent a few minutes, and you walk into the conversation already knowing where the bodies are buried.

Sensitivity grid. Discount rate up and down 100 basis points, terminal growth up and down 50, and the headroom at each combination. The agent builds the table and highlights the cells where headroom goes negative, which is the part your CFO will actually read.

Market capitalization reconciliation. If you're public, the sum of reporting unit fair values should reconcile to market cap with a believable control premium. An agent can lay out the bridge and draft the explanation for any gap. You decide whether the explanation holds up.

Memo and disclosure drafts. Feed it your policy, last year's memo and this year's numbers, and let it produce a first draft. We covered the same approach in using AI on a technical accounting memo and financial statement footnotes, and the rule carries over: the model drafts, you conclude.

What should AI stay away from in an impairment test?

The discount rate, the long-term growth rate, the forecast itself, and the decision about what counts as a reporting unit. Those are judgments, and they're the first things an auditor or valuation specialist will challenge. An AI will happily produce a WACC that looks reasonable, with a tidy build-up and confident language, and that's exactly why you shouldn't let it pick one. If you ever have to defend that number to an audit committee, "the model suggested it" is a damn weak answer.

There's a counterweight here worth holding onto. AI is useful as a challenger on those judgments even though it shouldn't make them. Ask it to argue that your discount rate is too low, or to list what a skeptical reviewer would ask about the forecast, and you'll often find the holes before somebody else does. Using it to pressure-test your assumptions gets you most of the value with none of the exposure.

How do you set this up so it holds up in an audit?

Treat it like any other control. Keep the prompts, inputs and outputs in the workpaper folder, have a named human tick and tie every number the agent produced back to source, and record who approved the key assumptions and when. Auditors are fine with tools they can follow, and the bullshit starts when nobody can show where a figure came from. Run it once alongside your manual process the first year, compare, and then decide how much to trust it. If you're hunting for the right tools, the Audit Friendly software directory is a decent place to see what close and consolidation platforms can do.

What I'd do

I'd pick one reporting unit, rebuild last year's test with an agent doing the carrying amount, backtest and sensitivity grid, and compare the output to the workpaper you already have. It'll take an afternoon, and you'll learn immediately where it's solid and where it needs a babysitter. Then I'd widen it, one piece at a time, while keeping every judgment call in a human's hands. The people who get good at this early also stand out in the market, since 2,459 of those same postings ask for technical accounting skills and very few candidates can say they've built their own AI-assisted workflow for it. You can see who's asking on the Audit Friendly job board, and if revenue recognition is next on your list, our piece on AI and ASC 606 follows the same pattern.

Frequently asked questions

Can AI perform a goodwill impairment test?

AI can build the supporting analysis: carrying amounts by reporting unit, forecast backtests, sensitivity tables, market cap reconciliations and draft memos. The conclusion, along with the discount rate and forecast assumptions behind it, should stay with a qualified person.

How often do you have to test goodwill for impairment?

At least annually at the reporting unit level, and between annual tests whenever an event or circumstance makes it more likely than not that fair value has fallen below carrying amount.

What changed with ASU 2017-04?

It removed the old step 2. Impairment is now the amount by which a reporting unit's carrying amount exceeds its fair value, limited to the goodwill allocated to that unit.

Should I let an AI choose the discount rate?

No. The rate is a judgment that auditors and valuation specialists will challenge, so use AI to challenge your assumptions and document your own reasoning.

How many accounting jobs mention goodwill impairment?

In our database, 260 of 47,445 classified postings with full descriptions mention impairment, and 148 mention goodwill, as of October 2026.

Pick the smallest reporting unit you have and run the experiment before next year's test is due, because the first time you find a stale mapping in an afternoon instead of in the audit fieldwork, you'll want to build the rest.