Hiring a Controller Who Can Sign for an Agentic Close

Where Does an Agentic Close Actually Break?

Day two of the close, and the agent has already posted nine hundred journal entries. Eight of them are flagged. One is an intercompany accrual that reversed twice and netted to something plausible. Nobody wrote it, nobody remembers approving it, and a controller has to sign the certification on Friday. That moment is the job.

The automation is real and it mostly works, which is exactly what makes the failure mode narrow and expensive. Reconciliations, accruals and flux commentary come back fast and formatted. What comes back with them is a queue of items the process could not resolve on its own, and a much larger population of items it resolved confidently and wrongly. The first queue is visible. The second one is the reason to hire carefully.

This shift already has a name in the finance staffing market. Transactional accounting roles are being described as exceptions work: the process runs, and when it hits a discrepancy it cannot settle, a person steps in ². That is a fair description of the clerk-level job. It undersells the controller's version, because the controller is accountable for the entries nobody escalated.

So the interview question that separates candidates is not how they would build an automated close. It is what they would do on the Friday. Ask which balances they would look at before signing when the system reports everything green, and why those balances. A candidate who answers with a risk-ranked list, a sampling rule and a stop condition has done this. A candidate who answers with the name of a platform has watched it.

Which Controller Traits Survive Contact With an Agent?

The traits that matter are visible in how a candidate talks about work they did not perform. Look for someone who samples rather than skims, who can name the specific control that would have caught last quarter's error, and who treats a clean agent output as a claim to be tested. Fluency with the tool is common now. Suspicion held in the right places is not.

Three tells separate the real thing from the rehearsed version.

The first is specificity about a failure. Ask for one time an automated or outsourced process produced a number they did not believe. Real answers are boring and detailed: a rebate accrual built off a stale rate table, a bank feed that silently dropped a currency, a duplicate vendor that survived a dedupe rule. Performed answers stay at the level of "you always have to check the output," which is a sentiment rather than a memory.

The second is whether the candidate reasons about coverage rather than volume. A controller who reviews more entries is not doing better work than one who reviews the right forty. Ask how they would decide what to test in a close where every account reconciles on schedule. The strong answer talks about which balances carry judgment, which ones changed behavior versus last quarter, and which ones an agent has been observed to get wrong before.

The third is where they stop delegating. Everyone will say estimates and judgment calls. Push for the boundary: who writes the reserve memo, who decides the materiality threshold, who talks to the auditor about a scope change. A candidate with a real line can draw it in one sentence and defend it.

The practice behind the skill shows up when you ask how they got good. The best answers describe deliberate testing rather than usage. Someone ran the agent against a prior period that was already closed and audited, then compared entry by entry to see where it diverged. Someone kept a private file of known-bad transactions and re-ran them after every model or configuration change. Someone insisted on reading the prompt or rule set the finance systems team had written, found it assumed a single functional currency, and had it fixed. Quality control of AI output is now among the skills workers themselves name as rising in importance, and most AI users say they treat model output as a starting point rather than a final answer ¹. Plenty of candidates will say that sentence. Fewer can show the file.

One exercise belongs in the interview loop rather than the offer letter. Give the candidate a real exception from a real close, with the agent's output and its stated reasoning attached, and ask them to decide. The intercompany accrual that reversed twice is a good one, because nothing in the file settles it. What they ask for before deciding tells you more than anything they say about themselves.

Which Backgrounds Produce a Controller for the Agentic Close?

Three benches produce this person. External audit, where sampling and evidence are the daily craft. Systems accounting and ERP implementation, where someone has already written a control matrix against software behavior. And internal audit or SOX testing, where evaluating someone else's process is the entire job. Traditional staff-accountant progression produces excellent closers who have never designed a control.

The audit background is the most competed for, and what makes it work is not the technical accounting. An auditor has spent years forming an opinion about work performed by other people, on evidence rather than on trust, and writing down why. That is structurally the same problem the agentic close creates, with the population changed from a client's staff to a software process.

The unexpected backgrounds are worth more attention because the market prices them lower.

IT audit and application controls, often the least glamorous seat in a Big Four practice, produces people who already think in terms of access, change management, completeness of a data feed and whether a log can be trusted. That vocabulary transfers directly, and the overlap with what an AI governance and policy analyst does for models is close enough that the two roles frequently end up writing each other's documentation.

Revenue operations and billing systems leads have usually built reconciliation logic themselves and know exactly how a matching rule fails at the edges. Tax provision specialists are used to defending a computed number to an outside reviewer with the workpaper attached. Payments and treasury operations people have lived with automated matching for a decade longer than the rest of finance and have opinions about it that are worth hearing.

One supply note that should shape expectations rather than the job description. The certified accounting pipeline has been contracting: CPA exam candidates fell about 25 percent over the past decade, and roughly three quarters of active CPAs reached retirement eligibility within the last five years ³. A search that requires a CPA, ten years of close ownership and demonstrated control design over automated work is a search for a small population. Decide which of the three is genuinely non-negotiable before posting.

Source the Agentic-Close Controller From the Audit Bench

Post the role where people who test systems already read. The strongest pool sits in external and internal audit practices at the Big Four and the national mid-tier firms, in the IT audit and SOX groups specifically. Also look at controllers who have already survived an ERP cutover, and at finance transformation consultants who want to stop advising and start signing.

The professional bodies are the honest venues. The Institute of Internal Auditors and ISACA both run local chapters and certifications aimed squarely at people who evaluate automated processes, and their members are used to being approached about industry roles. Financial Executives International chapters put controllers and CFOs in the same room monthly. The AICPA and the state societies run technology-focused sections and conferences where the attendees are, by self-selection, the accountants who care about this.

ERP and close-software user communities are the other productive channel: NetSuite, Workday, SAP and Oracle user groups. Somebody who posts a detailed answer about intercompany elimination behavior in one of those forums is telling you a lot about how they work.

Adjacent titles worth sourcing directly: Assistant Controller at a company one size larger than yours, Manager of Technical Accounting, Internal Audit Manager, SOX Program Manager, Finance Systems Manager. Feeder companies are less about brand than about condition: any organization that shortened its close by automating it in the last three years has produced people who learned this the hard way, whether the automation worked or not.

Two search notes. First, the title is not stable yet, so a keyword search on any single phrase will miss most of the market: look for the responsibility in the body of a profile rather than the words in the headline. Second, some of the best candidates are currently in a role that touches the systems rather than the ledger, closer to what an IT specialist does in a finance organization, and their resume will not read like a controller's.

What Does It Take to Close a Controller Who Signs for Agent Work?

What closes this candidate is authority rather than title. They want the control framework to be theirs to design, a named path to the certification they will sign, and a CFO who will back an override that costs the company a day. Offers die when the candidate learns that the agent rollout is already decided, the vendor is already chosen, and the controller's job is to bless it.

The second offer-killer is unclear accountability. If the finance systems team owns the configuration, the automation vendor owns the logic, and the controller owns the signature, say so in the interview and say who breaks a tie. A candidate who is good at this will ask. Getting a vague answer is the most common reason a strong finalist takes the other job, and it is usually a real signal about the organization rather than a negotiating posture. Where the control framework spans finance, IT and the model itself, some companies bring in an AI governance consultant to write the first version. That is a reasonable bridge, and it is not a substitute for an owner.

Anchor the Offer on the Corporate Controller Band, Then Name the On-Site Weeks

No published salary series exists for the agentic-close variant of this title, so anchor on the established one and expect a premium at the top of the band. Robert Half's 2026 Salary Guide, one vendor guide rather than a wage survey, puts the corporate controller range at roughly 152,000 dollars at the low end, 185,000 at the midpoint and 213,250 at the high end in the United States ⁴.

As of mid-2026, candidates who can genuinely evidence control design over automated close work are negotiating at or above that high figure, and in high-cost metros and at public filers above it. Treat those numbers as a base to adjust for geography, company size and public-company reporting duties rather than as an offer, and say which of the three you are adjusting for when you present it.

Remote norms are more permissive than the finance function's reputation suggests, with one recurring exception. The review work itself is screen work: sampling entries, reading logs, writing memos, all of which travel. Most postings for this profile are hybrid, typically two or three days on site, and fully remote arrangements are common at companies whose finance team was already distributed. The exception is the close itself. Many teams want the controller physically present for the first two or three closes after an automation change, and for the audit fieldwork weeks, because the conversations that resolve an exception fastest are the ones nobody scheduled. Say which weeks are on site when you make the offer. A candidate who finds out in month two treats it as a change in terms.

See how it works >