Setting Credit Caps and Exclusions for a Vendor Remedy Table

The first decision is the denominator. A cap expressed as four months' fees per year is common in published schedules, and it is far too loose for a service that matters. A sample clause limits aggregated annual credits to four months of monthly service fees and caps any single billing month at 50% of that month's fees (Law Insider's sample credit calculation). That structure protects the vendor against a catastrophic quarter far more than it protects the buyer.

Work from what the buyer pays, not from the annual contract value as a headline. Annual contract value is the annualized value of a multi-year contract, typically calculated by subtracting one-time fees from total contract value and dividing by contract length (Growth Equity Interview Guide on ACV). That is a sales and finance metric. It is useful for budgeting a renewal, and it is the wrong basis for a credit cap, because credits are applied against invoices and invoices are issued monthly or quarterly.

Here is a worked example that shows why the denominator matters.

Take a contract with total contract value of $180,000 over three years, plus $30,000 of one-time onboarding. Annual contract value is ($180,000 − $30,000) ÷ 3 = $50,000 (Wall Street Prep's ACV formula). Quarterly recurring fees are $50,000 ÷ 4 = $12,500.

Now suppose the availability target is 99.9% monthly, and the credit schedule pays 10% of the affected monthly fee at 99.0% to 99.9%, and 25% below 99.0%. In a bad quarter the buyer lands at 98.6%, 99.4% and 97.9%. Applying the schedule to the monthly recurring charge of $4,167 gives credits of roughly $1,042, $417 and $1,042, about $2,500 in total.

A 25% quarterly cap is the pragmatic middle. It binds only when the quarter is genuinely bad, it is easy to forecast for both sides, and it keeps the maximum remedy proportional to the fees paid for those three months.

The second decision is whether the cap limits credits or limits liability. Keep them separate. Credits are an offset against future invoices, not a payment. Damages for data loss, security breach or chronic failure should sit outside the credit cap, and mature clauses carve those out of the sole-and-exclusive-remedy language (ContractKen).

A rule of thumb for the cap level: set it so a single incident cannot exhaust the quarter, but a systemic failure can. If your second-tier credit is 25% of a month's fee, a 25% quarterly cap allows exactly one such incident before further credits stop. Run that arithmetic before you agree the tier table.

You can settle most of the cap argument in a spreadsheet. This snippet takes the recurring monthly fee and a list of availability figures, applies a two-tier schedule, and reports the uncapped credit and the credit after a quarterly cap.

def quarterly_credit(monthly_fee, availabilities, cap_pct=0.25):
    total = 0.0
    for a in availabilities:
        if a < 0.990:
            total += monthly_fee * 0.25
        elif a < 0.999:
            total += monthly_fee * 0.10
    cap = monthly_fee * 3 * cap_pct
    return {"uncapped": round(total, 2), "cap": round(cap, 2),
            "payable": round(min(total, cap), 2)}

print(quarterly_credit(4167, [0.986, 0.994, 0.979]))
# {'uncapped': 2500.8, 'cap': 3125.25, 'payable': 2500.8}

Set all three months to 0.985 and the uncapped total reaches $3,125.25, which is where the cap starts to bind. That threshold is the number worth negotiating over, because it tells both parties exactly how bad a quarter has to be before the cap does any work. A vendor that offers a high cap and a low per-incident percentage has conceded very little, since the cap never engages.

Scheduled maintenance is a legitimate exclusion. An exclusion that is not defined, not notified and not time-limited is an escape hatch. Both sides should be able to answer four questions before the quarter starts: when the window is, how long it lasts, how much notice the buyer gets, and who can move it.

Published SLA templates treat notification of scheduled maintenance as a provider obligation alongside response times (By Design Law's SLA template). The clause should state the notice period in business days, the maximum duration of a window, the maximum number of windows per quarter, and a requirement that the schedule live in a shared calendar the buyer's operations team can subscribe to. Then decide what happens when the vendor misses its own notice. The sensible rule is that an unannounced window is not an exclusion; it counts as downtime and is credit-eligible.

The scale makes this concrete. A 30-day month contains 43,200 minutes, so a 99.9% monthly target permits 43.2 minutes of downtime for the entire month, and a single incident with ordinary coordination delays can consume all of it (All Quiet's SLA guide). Against an error budget that small, an unbounded maintenance exclusion is not a margin. It is the whole remedy.

The other standard exclusions need the same treatment. Third-party network failures, customer-caused outages and force majeure are reasonable, but each needs a definition and a recording method rather than a vague exception clause (SupportGPT's support SLA guide). A vendor that cannot say which exclusion applied to which incident is asking to be believed rather than audited.

For AI-based services, add one entry. Model provider changes, tool availability and evaluation-set drift should be treated as production release events with a defined review process, not as background noise that never triggers a remedy (Wavect's AI agent SLA template). The provider's own SLA does not cover the behavior of an agent built on top of it.

The third boundary is the anchor. A credit calculated on the whole subscription when one component failed is over-generous in theory and, in practice, endlessly argued. Published schedules solve this by calculating credits on the monthly recurring charges of the affected service only, using a pro rata share of the affected component (Law Insider).

Two calculation methods dominate in sample clauses. The percentage method applies the tier directly to the affected line: if that line bills $10,000 and the tier is 20%, the credit is $2,000. The hourly method prorates against hours in the month: $10,000 multiplied by 3.5 over 720 hours gives $48.61. The same sample caps annual credits at four months' fees and monthly credits at 50% of the affected service's fees.

The anchor rule should specify five things:

1. The invoice line identifier printed on every invoice.
2. A versioned appendix mapping each service component to an invoice line.
3. The rule that the credit base is the affected line's charge for the period, exclusive of taxes and one-time fees.
4. The rule that credits apply to the next invoice for that same line, not another line.
5. The rule that unused credits survive termination and are paid or offset.

Point five is the one buyers forget. Credits applied only against future invoices are worth nothing if the contract ends first. A cash-out or final-offset provision closes that gap and costs the vendor little, because it only matters at exit.

The anchor also disciplines reporting. If every incident is recorded against an invoice line, the monthly service report becomes a reconciliation document rather than a narrative, which matches how mature templates describe reporting: recurring reports with centrally accessible data, so performance is verified rather than assumed (By Design Law). One detail deserves its own line. Where a service has many components, apply the credit pro rata. If ten units are contracted and one fails, calculate on that unit's charge, not all ten, exactly as the sample clause does for bare metal blades and branch circuits (Law Insider).

The boundary layer only works if it is self-executing. The table, the exclusions, the cap formula, the anchor mapping and the claim procedure must fit in a schedule that a procurement analyst, an account manager and a finance controller apply the same way.

The largest source of friction is the claim window. Vendors commonly require written credit requests within a short period, often 30 days from the incident, and customers who fail to track incidents forfeit credits even when the breach is uncontested (ContractKen). Do not spend the negotiation removing the window. Remove the need for the buyer to remember. Require the vendor to issue a monthly report that states, for each invoice line, measured availability, incidents and the credit applied automatically. Automatic application is the key sentence, because it turns the claim window into a fallback for disputes rather than the mechanism the remedy depends on.

Handle chronic failure separately. Credits compensate for a bad month; termination compensates for a bad pattern. Mature clauses include a chronic-failure termination right triggered by repeated breaches in a rolling period or a single outage above a defined duration (ContractKen). Keep it outside the cap and outside the sole-remedy language.

Two governance habits keep the table evergreen. Version the schedule: every change to a tier, cap, exclusion or appendix gets a version number and an effective date in the header. Then rehearse the calculation once a year using the buyer's real invoice data and a real incident from the report history. If both sides cannot reproduce the same figure from the schedule alone, the schedule is not finished.

That rehearsal is also where a review of the underlying stack pays for itself. If the remedy table sits over a service that assembles several model providers, the mapping from component to invoice line must match the architecture rather than the sales deck. Teams reconciling that across a multi-model stack often route the work through a single interface instead of a dozen vendor contracts; Quanta Inference exposes text, image, video, audio and embedding models through one OpenAI-compatible key with prepaid US dollar pricing, which simplifies the invoice line the anchor rule depends on. When the drafting needs a second pair of eyes on the commercial terms, Quanta Consulting runs a free 30-minute discovery call to map goals to a recommendation.