Skip to main content
DoneThat

AI Adoption GuideHRReward

Merit cycle agent

Drafts manager-level compensation recommendations within budget, equity, and band guardrails, with explanations.

HR processPlanSourceSelectHireOnboardDevelopRewardExit

By Don, DoneThat’s AI coach · updated

A draft is only usable when budget, band, and equity are cited

A merit recommendation is usable when every number on it cites the cycle budget, the pay band, and the equity file. If a cite is missing, that field stays empty. Comp still publishes the cycle. Managers still decide. The agent does not auto-post pay and does not invent a merit percent to complete a row.

The draft is a working paper for the manager, not a payroll instruction. A number without a cite is an incomplete draft. Send it back or leave it blank.

You will still run the cycle in the systems you already use. Workday, CompAnalyst, Lattice, and Visier, and the rest of that class, remain where envelopes, bands, worksheets, and workforce files live. The agent does not replace those records. It names the file it used. It stops when the named file is not there.

Load budget, band, and equity before anyone sees a recommendation

Load the three files before a manager sees a draft. The draft may only speak from what you loaded.

Load the cycle budget first. Use the Comp-published envelope for this manager's node: allocated dollars, any holdback Comp reserved, and any rule about leftover funds. Name the extract in the cite (cycle, node, version Comp locked). If the budget file is missing, or the manager's node is not in it, leave spend fields empty. Do not borrow another node's leftover.

Load the band file second. Use the range Comp published for this cycle: minimum, midpoint, maximum, and the in-range versus promotional rules that already apply. Name the band table in the cite (structure version, job code, geography). A recommendation with no band cite is a failure, not a draft. If the person is unmapped, dual-banded, or still on a range Comp has not refreshed, leave in-range and to-midpoint fields empty. Do not substitute last year's range. Do not substitute a neighboring job code.

Load the equity file third. Use the picture Comp will stand behind for this population: flagged clusters, holds Comp already attached, and any freeze Comp wrote before the cycle opened. Name that extract in the cite. When flags originated in continuous pay-equity monitoring, point at the cycle equity file Comp cut from that work, not at a live dashboard the manager cannot replay. If this person is not in the equity extract, leave equity commentary blank. Do not invent a gap. Do not invent a closing percent.

Only then attach optional context Comp has already scoped into the cycle. Live market benchmarking can sit next to the band file when Comp named the market cut this cycle will use. Compensation flight-risk model output can sit in a notes field when Comp said retention risk is in scope for this manager. Neither file is a fourth required cite. Neither file authorizes a merit percent on its own.

Write the rec with cites, then leave every uncited field empty

Draft one recommendation per person, in the manager's worksheet shape, with the cite next to the field it supports.

Budget fields cite the budget file: envelope, dollars already allocated to other drafts in the same node, and any Comp holdback. Band fields cite the band file: current position in range, proposed position in range, promotional path versus in-range path. Equity fields cite the equity file: whether this person sits in a flagged cluster, whether Comp already attached a hold. If a field cannot point at one of those files, leave it empty. Empty stays empty.

Do not invent a merit percent. If the draft cannot cite budget, band, and equity together, it does not mint a percent as a stand-in. A blank percent with two good cites is still a usable draft: the manager can see what is known and what Comp still has to fix. A filled percent with a missing band cite is not usable. That is the first failure mode, a rec with no band cite, and a narrative explanation does not repair it.

Keep the explanation checkable. The manager should be able to open the named files and see the same facts. If not, delete the explanation and clear the field.

Here is the illustration. A manager has twelve people. Eleven have a budget line, a current band, and an equity-file row. The twelfth is a recent transfer still on the old job family in the band file, and missing from this cycle's equity extract. The eleven can carry cited budget, band, and equity language. The twelfth shows the budget cite and leaves band position, equity note, and merit percent blank. The manager still holds the conversation. Comp still decides whether to remap before publish. Nobody fills the blank percent to complete the row.

That blank is the quality signal. Filling it would recreate the failure: a rec with no band cite.

Managers decide; Comp publishes; pay is not posted from the draft

The manager reads the draft, including the blanks, and decides. The decision can accept a cited recommendation, change it, or send the person back to Comp for a remap. The agent does not close the worksheet, generate a letter, or push a new base pay into the HRIS.

Comp publishes the cycle. Publication turns manager decisions into the official worksheet: locked budget, locked bands, locked equity treatment, and the percents Comp will stand behind. Until Comp publishes, the draft is not posted pay.

Treating a draft as posted pay is the second failure mode. It shows up when a manager quotes an agent percent in a 1:1 before Comp has locked the file, when finance accrues to a worksheet Comp has not published, or when payroll is handed a file that still contains blanks. The only pay that moves is the pay Comp published.

The third failure mode is inventing a merit percent so the row looks complete. Completeness is Comp's job at publish, not the agent's job at draft. If Comp cannot publish a person because band or equity is still missing, that person stays off the posted cycle or goes out with a Comp-authored exception. The agent does not author the exception.

Before publish, confirm every posted percent has a budget cite, a band cite, and an equity cite, or a Comp-signed exception that names which cite is waived. Confirm no row still shows an agent draft flag.

Calibration still belongs to humans. If you inspect the manager's decisions for pattern, inspect the decision set, not the draft. Point that work at a calibration bias detector after managers have acted.

Keep market, equity, and calibration work in their own files

Merit drafting is one step in reward, not the whole stack. Market cuts, equity monitoring, flight-risk scores, and calibration checks each have owners and files. Pull them in as named context when Comp has already scoped them into this cycle. Do not collapse them into a single invented percent.

When the band file is stale relative to the market cut Comp approved, fix the band file. Do not ask the agent to split the difference. When equity monitoring and the cycle equity file disagree, Comp reconciles them before draft, not in the manager's 1:1. When flight-risk output would push a person through the top of the band, the draft stays inside the band cite or stays blank until Comp writes an exception.

The usable output is a manager-level draft that cites budget, band, and equity, with empty fields where a cite is missing. Comp still publishes. Managers still decide. Pay is posted only after that.

Is this worth automating for you?

Whether this pays back depends on how much time it takes your team today. Most teams estimate that from memory, and the estimate is usually wrong in one direction or the other.

DoneThat reconstructs where the time actually went, with no timers to forget, so you can measure the baseline before committing to a project and check the gain afterward.

Measure the baseline first