Publicité
ERP IMPLEMENTATION
🇫🇷 Lire en français →

What Does One Hour of ERP Downtime Cost? How to Calculate It and Set the Right SLA, RTO and RPO

Calculate what one hour of ERP downtime costs, then set your SLA, RTO and RPO from that figure. Cost model, uptime table and a worked mid-market example.

What Does One Hour of ERP Downtime Cost? How to Calculate It and Set the Right SLA, RTO and RPO

Ask your leadership team how much one hour of ERP downtime costs. Most of the time you will get silence, or an answer along the lines of “a lot”. Yet this is the figure that should set the availability level you require, the resilience budget and how hard you push in negotiations with your vendor or hosting provider.

This article describes a method you can reproduce in a spreadsheet in half a day: put a price on one hour of downtime, vary it by period and module, translate it into RTO, RPO and an availability percentage, then test the result against a SaaS vendor’s SLA. All amounts in the example are fictitious and illustrative; the only hard numbers are arithmetic. For building the recovery plan itself, see our guide to business continuity and disaster recovery for your ERP, which we do not repeat here.

Why most companies don’t know what ERP downtime costs them

Three reasons explain this blind spot.

The cost is scattered. An ERP outage does not produce one identifiable expense line. It is spread across production (blocked work orders), sales (orders that cannot be entered), idle payroll for people waiting, next-day overtime, late-delivery penalties and sometimes compliance fines. No single department sees the whole picture: IT sees the incident, production sees idle machines, finance sees the effect a week later.

Downtime is seen as a technical issue. In many organisations the topic stays within infrastructure. The business was never asked what it can tolerate. The result: availability targets are whatever the hosting provider offers by default, not what operations actually require.

The contract settles the question before the calculation exists. The SLA is signed during commercial negotiation, when attention is on licences, per-user pricing and the project plan. The service level shows up in an appendix with a percentage that sounds high (99.5%, 99.9%) and nobody has converted it into minutes.

Putting a number on downtime, even a rough one, flips the conversation. You stop asking “what SLA does the vendor offer?” and start asking “what SLA does our business require, and at what price?”

The calculation model: three families of cost

The cost of one hour of downtime breaks down into three blocks. The first two can be measured with company data; the third is closer to scenario-based estimation.

Direct costs: production, lost sales, idle labour

  • Idle labour. Number of people who depend on the ERP × average fully loaded hourly cost × actual idle rate. The idle rate is rarely 100%: some teams switch to off-system tasks (tidying, customer calls, paper-based work). Measure it with department heads rather than assuming it.
  • Production and shipping. If manufacturing orders, pick lists or delivery notes come out of the ERP, every hour of downtime delays volumes. What matters is not the hour’s revenue but the contribution margin (revenue minus variable costs) that is not recovered afterwards.
  • Lost sales. For e-commerce or point-of-sale connected to the ERP, some customers will not come back. For contract-based B2B, the order is usually postponed rather than lost: the permanent loss rate is low, but not zero on substitutable products.

The distinction between “postponed” and “lost” is the most sensitive variable in the model. Never set it alone: have it validated by sales and operations leadership.

Catch-up costs: overtime and re-keying

Once the system is back, the company pays for the accumulated backlog:

  • Overtime to clear the queue (premium included), or temp agency fees;
  • Re-keying of whatever was noted on paper or in spreadsheets during the outage, with a risk of errors that will cost you later (stock discrepancies, incorrect invoices);
  • Express freight to meet the delivery dates promised to priority customers.

These costs are often forgotten because they land the next day, in other budget lines. Yet they can amount to a share comparable to the direct costs.

Indirect costs: penalties, reputation, compliance, e-invoicing

  • Customer penalties. Review your sales contracts: guaranteed delivery times, per-day late penalties, service clauses in retail or automotive. Use an expected value per hour (probability × amount), not the worst case.
  • Reputation and loyalty. Hard to quantify. Don’t invent a number: flag it as a qualitative line in the spreadsheet, with zero in the base calculation and a separate sensitivity analysis.
  • Compliance. A month-end close or statutory filing that fails because of an outage can trigger late-filing penalties. The same goes for traceability obligations (food, healthcare).
  • E-invoicing. With the rollout of e-invoicing mandates in Europe, issuing and receiving invoices increasingly goes through a certified platform connected to the ERP. If the ERP or the platform is down, invoices pile up and cash collection slips. Our analysis of the French e-invoicing platform outages in September 2026 shows that this external dependency belongs in the calculation, because the failure can come from a link you do not control.

Formula and spreadsheet to reproduce

The basic formula, per hour of downtime:

Hourly cost = idle labour + unrecovered margin + catch-up + expected penalties

where:

  • idle labour = dependent headcount × loaded hourly cost × idle rate;
  • unrecovered margin = (hourly revenue × contribution margin rate) × share permanently lost;
  • catch-up = premium-rate overtime + re-keying + exceptional costs;
  • expected penalties = probability of triggering × average amount.

In the spreadsheet, build an “assumptions” tab (one cell per parameter, sourced internally with the validator’s name), a “calculation” tab containing only formulas, and a “scenarios” tab (normal day, month-end close, sales peak). Each assumption carries a date and an owner: the figure becomes a governance object, not a one-off exercise.

Variation by module and by period

The cost of an hour is not constant. Three factors make it vary widely.

The period. An outage on the 28th of the month during close, on a peak order day or the day before a major shipment does not cost what the same outage costs on a Tuesday in February. In the example below we apply a 1.5 multiplier for month-end close; calibrate your own with the teams concerned.

The module affected. An accounts payable outage on a Thursday can be recovered; a warehouse inventory outage physically blocks the trucks. Rank processes by criticality:

ProcessTypical downtime toleranceReason
Shipping, picking, point of saleVery low (minutes to 1 h)Physical blockage, customers present
Production (scheduling, MES)Low (1 to 4 h)Stopped lines, scrap
Order entry, e-commerceLow (1 to 4 h)Lost sales
Invoicing, cash collectionMedium (4 to 24 h)Delayed rather than lost, but watch e-invoicing
Accounting, closeMedium in general, low during closeStatutory deadlines
Reporting, BI, off-cycle payrollHigh (24 h and more)Recoverable

These values are starting orders of magnitude, to be tested against your own business.

The dependencies. The ERP is rarely alone. If DNS, the authentication directory (SSO), the VPN, the site network or the e-invoicing platform goes down, the ERP is “available” from the vendor’s point of view and unusable from the user’s. Map these dependencies and ask, for each, who owns the SLA.

Turning the cost into targets: RTO, RPO and availability by process

Once the hourly cost is known, three indicators follow.

  • RTO (Recovery Time Objective): the maximum acceptable interruption before recovery. It is the threshold beyond which the cumulative cost becomes unbearable for the process.
  • RPO (Recovery Point Objective): the maximum amount of data you accept to lose, expressed as time (“15 minutes of transactions”). For a transactional ERP, an RPO of a few minutes requires continuous replication; a 24-hour RPO corresponds to a simple daily backup.
  • Availability: the percentage of time the service is usable over a given period.

RTO is reasoned about in cumulative cost: if an hour costs €15,000, a full day of downtime costs far more than 24 times that amount, because catch-up and penalty costs grow non-linearly. RPO is reasoned about in reconstruction cost: how many person-hours does it take to re-key one hour of lost transactions?

The conversion table below is simple arithmetic: a 365-day year has 525,600 minutes (8,760 hours), a 30-day month 43,200 minutes.

AvailabilityMax downtime per yearMax downtime per month (30 days)
99%5,256 min (87.6 h, or 3.65 days)432 min (7.2 h)
99.5%2,628 min (43.8 h)216 min (3.6 h)
99.9%525.6 min (8.76 h)43.2 min
99.95%262.8 min (4.38 h)21.6 min
99.99%52.56 min4.32 min
99.999%5.26 min0.43 min (about 26 seconds)

Two takeaways. First, the difference between 99.5% and 99.9% is a factor of five on tolerated downtime: a few tenths of a point is not cosmetic. Second, an annual percentage hides concentration: 99.9% allows a single outage of nearly 9 hours in one go, right in the middle of the close. So also demand a cap on duration per incident and monthly measurement.

Then set targets per process, not for the whole ERP. Paying for a 99.99% architecture for reporting modules is waste; accepting 99.5% on a distributor’s shipping is a mistake.

Reading a SaaS vendor SLA

A SaaS SLA reads like an insurance policy: the exclusions are what matter. Five points to check.

  1. Scope. Which service is covered: the application, the API, the connectors, the test environment? An SLA that covers only “the platform” may exclude the add-on modules or integrations you depend on. Do secondary or regional production environments carry the same commitment?
  2. Exclusions. Planned maintenance is almost always excluded from the calculation, sometimes within a generous window (several hours per month). Do the maths: if 4 hours of monthly maintenance are excluded, the “contractual” 99.9% hides a much higher real unavailability for your users. Also look at exclusions for force majeure, third-party supplier failure (the underlying cloud) and “abnormal use”.
  3. Measurement method. Who measures, where and how often? A vendor’s internal probe testing a health page says nothing about your experience. Ask for access to the measurement data and the ability to compare it with your own external monitoring. Also specify what counts as “unavailability”: a service entirely unreachable, or performance degraded to the point of being unusable?
  4. Calculation period. Monthly, quarterly or annual: the longer the period, the more an isolated outage is diluted.
  5. Remedies. Almost all SLAs provide service credits, i.e. a credit note on the next invoice, often capped at a percentage of the monthly fee and triggered on the customer’s request within a short deadline.

The key point: service credits are rarely proportional to the actual loss. They are calculated on the fee, not on your operating loss. An outage that blocks your shipping costs you tens of thousands of euros; the credit is a fraction of one monthly subscription. In most contracts the credit is also presented as the sole and exclusive remedy, which closes the door on a damages claim, except for gross negligence or wilful misconduct depending on the applicable law. This is one to have a lawyer review (see our guide to the clauses to negotiate before signing).

Negotiating: clauses, reporting, audit and exit

With your hourly cost in hand, you negotiate with numbers. Here are the requests to prioritise.

Service level clauses

  • Availability per critical process or module, rather than one global figure.
  • A cap on duration per incident, in addition to the monthly percentage.
  • Limited maintenance windows, announced in advance, never during the close and peak periods you list in the contract.
  • Contractual RTO and RPO (not just “objectives”), with associated penalties.
  • Commitments on response time and communication during an incident, with a post-incident report (root-cause analysis) within a set deadline.

Remedies

  • Progressive credits, escalating as unavailability worsens.
  • Right to terminate without penalty after repeated breaches (for example, three non-compliant months out of twelve).
  • A higher liability cap for losses linked to unavailability, or at least a distinction between service credits and damages.

Reporting and audit

  • A monthly availability report, with the raw measurement data.
  • A right to audit continuity arrangements, or at a minimum access to independent certification reports.
  • A documented restore test at least once a year, proving that restoration meets the stated RTO and RPO. A backup that has never been restored is an assumption, not a protection.

Exit

Dependencies. Ask for the list of critical subcontractors (cloud host, DNS, identity provider) and require the vendor to pass its commitments down. For your own share of the scope (SSO, network, e-invoicing platform), document your internal SLAs and those of your providers: the chain is only as strong as its weakest link, and every link must be aligned.

A complete worked example: a fictitious 300-person mid-market company

Illustrative. The company, the amounts and the SLA terms below are invented to show the method. They represent no real company or vendor.

Assumptions. A mid-market industrial and trading company: 300 employees, €90M revenue, 250 working days of 8 hours (2,000 operating hours per year).

  • Hourly revenue: 90,000,000 / 2,000 = €45,000.
  • Contribution margin: 30%, i.e. €13,500 per hour.
  • Headcount dependent on the ERP: 60% of 300 = 180 people, average loaded hourly cost €45, idle rate 70%.
  • Share of margin permanently lost: 30% (the rest is postponed).
  • Catch-up: 60 overtime hours per hour of downtime, i.e. 40 people for 1.5 hours, with a 25% premium, plus 15 hours of re-keying.
  • Expected penalties: €1,500 per hour (probability × average amount of contractual penalties).

Calculation for one hour in a normal period.

ItemCalculationAmount
Idle labour180 × €45 × 70%€5,670
Unrecovered margin€13,500 × 30%€4,050
Overtime60 h × €45 × 1.25€3,375
Re-keying15 h × €45€675
Expected penaltiesAssumption€1,500
Total per hour€15,270

With the 1.5 multiplier applied during close, the hour reaches €22,905. Brand damage stays outside the calculation, flagged qualitatively.

Translation into a downtime budget. If an outage uses up the entire allowance permitted by the SLA, the maximum annual cost follows from the availability table:

AvailabilityMax annual downtimeCost (at the normal rate of €15,270/h)
99.5%43.8 h€668,826
99.9%8.76 h€133,765
99.95%4.38 h€66,883
99.99%0.876 h€13,377

Moving from 99.5% to 99.9% cuts the maximum exposure by about €535,000 a year. That is the amount it is rational to invest (SLA premium, redundancy, premium option) to reach that level, to be compared with the surcharge the vendor asks for.

Comparison with the service credit. Assume a €240,000 annual subscription (€20,000 per month) and a fictitious SLA providing a credit of 10% of the monthly fee below 99.9%. For a 4-hour outage in a normal period:

  • estimated loss: 4 × 15,270 = €61,080;
  • service credit: 10% × 20,000 = €2,000, about 3.3% of the loss.

The practical conclusion: the credit does not compensate the loss. The SLA is mainly valuable for its indirect levers (termination right, reporting, cap on duration per incident), which should be negotiated first.

Targets set for this company (illustration).

ProcessTarget availabilityRTORPO
Shipping, logistics99.9% during working hours1 h5 min
Production99.9%2 h15 min
Order entry99.9%2 h15 min
Invoicing, accounting99.5% (99.9% during close)4 h1 h
Reporting, BI99%24 h24 h

10-point checklist

  1. Do you have a documented cost for one hour of downtime, validated by finance and operations?
  2. Is this cost broken down by period (normal, close, peak) and by critical process?
  3. Do you distinguish postponed sales from lost sales, with sales validation?
  4. Are RTO and RPO set per process and signed off by business owners?
  5. Have you converted each availability percentage into minutes per month and per year?
  6. Does the SLA specify scope, exclusions, maintenance windows and measurement method?
  7. Is there a cap on duration per incident, in addition to the monthly percentage?
  8. Have you compared the service credit with the estimated loss, and negotiated termination and reporting accordingly?
  9. Are dependencies (DNS, SSO, network, e-invoicing platform) mapped, each with an owner and an SLA?
  10. Is a documented restore test performed at least once a year, with comparison against contractual RTO and RPO?

Taking action

Start with a two-hour workshop with finance, operations and IT to fill in the “assumptions” tab of the model. Then test the result against your current contract: does the existing SLA cover what your business requires? If the gap is large, it is a quantified argument for renegotiating at renewal.

To go further, read our guide to business continuity and disaster recovery for your ERP, our article on the 12 clauses to negotiate before signing and the one on e-invoicing platform outages in France. If you would like your ERP or SaaS contract reviewed against your downtime cost, get in touch for a contract audit.