A policy is a set of operating choices
The useful way to draft a policy is to make the decisions first and generate the prose second. The generator asks for exactly the decisions the document depends on:
- Reimbursable categories. Which kinds of spend the company pays back — travel, meals, accommodation, mileage, software, home office, client entertainment. Naming categories the expense system actually uses keeps the policy enforceable; inventing prose categories the workflow cannot see guarantees drift.
- Excluded expenses. The explicit non-reimbursables: fines, personal entertainment, alcohol where policy excludes it, upgrades. An exclusion list shortens every future dispute, because “not on the list” becomes a checkable fact rather than a negotiation.
- Receipt rules. What evidence is required, above what amount, and what happens when a receipt is missing. The strictest workable rule is “receipt for everything”; most companies define a small-amount exception, and the honest version states it as a number someone chose, not a law of nature.
- Approval thresholds. Who approves at which amount. The policy states the principle; the numbers belong to your delegation of authority matrix, built from your own spend distribution.
- Reimbursement timing. How quickly employees are paid after approval, and on what cycle. This is the sentence employees care most about, and the one most policies omit.
- Exception ownership. Who may approve something outside the rules, and how it gets recorded. Every real policy has exceptions; a good one names their owner instead of pretending they will not happen.
The output marks each choice it turned into text, and flags with a review marker every default it had to assume — so the person reviewing the draft sees exactly which sentences were decisions and which are placeholders awaiting one.
The sections a working policy needs
| Section | What it must state | Why finance needs it |
|---|---|---|
| Purpose and scope | Who the policy covers, which entities, effective date | Scope disputes are the first refuge of a contested claim |
| Eligible expenses | The reimbursable categories, in the workflow’s own names | Enforceable only if the system shares the vocabulary |
| Excluded expenses | The explicit non-reimbursables | Turns disputes into lookups |
| Receipts and evidence | What proof is required, thresholds, missing-receipt procedure | The substantiation questions an audit starts with |
| Travel and mileage | How travel is booked and mileage reimbursed, at which rate and from which source | The highest-volume, highest-emotion category |
| Approval authority | Who approves, by amount band and role — or a pointer to the authority matrix | The link between the policy and the delegation of authority |
| Reimbursement timing | Submission deadlines and payment cycles, in days | The employee-facing promise; late submission rules live here too |
| Exceptions | Who may grant them, how they are recorded | An unrecorded exception is indistinguishable from a control failure |
| Records and retention | Where evidence lives and how long it is kept | Retention is jurisdiction-specific; the policy names the owner who knows |
Two writing rules keep the document usable. State rules as numbers and names, not adjectives — “reasonable” and “appropriate” move every decision into the approver’s mood; a threshold, a day count, or a named role moves it into the table. Keep jurisdiction-specific promises out unless you have a source — mileage rates, per-diem amounts, and tax treatment vary by country and year, so the generator inserts a review marker where a local rate belongs rather than inventing one.
Receipts and substantiation: anchor the rules to what auditors ask
Receipt rules are where generic templates hurt most, because the applicable standard is jurisdictional. Two reference points show the shape of what authorities expect:
- In the United States, IRS Publication 463 frames substantiation as adequate records kept timely, proving the time, place, and business purpose of the expense — explicitly including cases where a standard meal allowance is used. Its accountable-plan rules add the employer-side mechanics: expenses accounted for within a reasonable period, and excess reimbursements returned, with failures on either point changing the tax treatment of the reimbursement.
- In the United Kingdom, HMRC’s employer guidance states that an employer providing expenses or benefits to employees or directors must usually report them to HMRC and pay tax and National Insurance on them, with different rules depending on the type of expense or benefit.
The practical translation for a policy: require evidence that answers when, where, and why — not just how much; require it close in time to the spend, because late reconstruction is what “timely kept records” exists to prevent; and name the jurisdictional owner (usually the entity’s finance lead) for the rules that vary by country. The generator writes the evidence rule from your choices and marks the jurisdiction-specific slots for local review.
Thresholds belong to your data, not the template
The generator deliberately ships no default thresholds. A receipt floor or an approval band copied from a template imports another company’s spend distribution — and with it, queues that get skimmed or controls that never fire. Set the numbers from two quarters of your own claims, denominated per entity in the entity’s own currency, and record the effective date so the next reviewer knows what era they belong to.
The policy document should state the principle — spending above a band requires the named role’s approval — and delegate the numbers to the delegation of authority matrix, which is built for exactly that table and is easier to re-calibrate than a prose document. The reasoning behind percentile-based thresholds, and the failure modes of over-specified rule sets, are covered in spend controls.
Timing, exceptions, and the sentences that prevent tickets
Three short sections do disproportionate work:
Submission deadline
“Claims are submitted within N days of the expense” — with the consequence for missing it stated (routes to finance review, not silent rejection). Open-ended submission windows are how December claims arrive in March.
Payment cycle
“Approved claims are reimbursed in the next payment run; runs happen every N days.” An employee who knows the cycle does not open a ticket to ask where their money is.
Exception record
“Exceptions may be granted by [role] and are recorded with their reason.” One sentence, and every exception becomes reviewable instead of anecdotal.
From document to enforcement
A finished policy enforces nothing by itself — enforcement is the workflow’s job, and the mapping is direct: each policy rule becomes a condition the system evaluates, a response when the condition is met, and a record of both. Categories become the submission form’s options; receipt rules become required attachments; thresholds become routing; exception ownership becomes an escalation path.
Clara Global enforces the rules you configure: each expense is checked and routed against the approval rules your finance team defines — amount bands, categories, currencies, entities, requesters, cumulative spend — so the policy’s rules fire on every claim instead of depending on someone remembering the PDF. Clara Global does not generate or invent policy — the choices stay yours; the workflow applies them. How rules-based checking and routing behave in practice is covered in automated expense approvals.
The rules are configured per company and evaluated on the expense itself, so they apply to an employee in any country submitting in any currency. There is no separate market rollout to wait for.
The honest sequencing: generate the draft here, review the marked assumptions with finance, set the numbers from your own data, and only then configure the workflow to match — a policy mapped to enforcement in that order stays true; a workflow configured first and documented later never quite matches its own policy.