Cybersecurity · Governance
EDR, immutable backups and MDR:
what to fund first
A small business should protect its backups first, then calculate the price at which EDR breaks even. The decision on MDR comes down to how much it costs on top of EDR. This guide gives the order, the method for redoing the calculation with your own figures, and what to demand technically from each measure.
Published 27 min readThe article in 30 seconds
- Three checks before buying anything. A second factor on remote access, patches on exposed equipment, and separate administrator accounts.
- Backups first. In our fictional example, they avoid €14,000 more than they cost each year, provided the lock holds against an administrator.
- EDR comes next. It is close to breaking even, but of the two measures it is the only one that addresses data theft. It must block, not just alert.
- Judge MDR on its additional cost. At €10,000 extra, it does not pay off on average. It comes close to breaking even if nobody reads alerts at night, and its right to act must be written into the contract.
- There is no official percentage. No reliable source sets the share of the budget to spend on it. Observed levels are at the end of this guide.

The method: two tests and six figures
This guide follows four steps, in this order.
Each step has its effect and its proof
Every security expense faces two tests. The first compares what it costs with what it avoids on average. The second asks whether the company survives its worst year.
The examples refer to the fictional case in this guide
A gap smaller than the uncertainty in the assumptions settles nothing: that is the case for EDR, at −€1,250.
The expected loss of a scenario is its annual probability multiplied by its cost. That cost is built as follows: days of downtime × cost of one day, plus the fixed costs of the incident.
The gap between the expected loss before and after gives the break-even point. It is the highest annual price at which the measure is justified by the average alone, and the figure to keep in mind when you read a quote.
| Figure | How to obtain it | Reliability |
|---|---|---|
| Cost of one day of downtime | Margin lost that day, wages paid without output, late-delivery penalties. Work it out with finance, sales and operations. | Can be calculated |
| Likely length of downtime | The time measured during a restore exercise, not the one stated in the contract. | Can be measured |
| Fixed costs of an incident | A provider’s intervention, rebuilding workstations, legal advice, informing the people affected. Ask your provider and your insurer. | From quotes |
| Annual probability | No source gives it for your company. Take a low value and a high value, and run the calculation with both. | Estimate |
| Cash that can be mobilised | What the company can release within a few weeks, credit lines included. | Known |
| Full cost of the measure | The grid below. | Quotes and staff time |
Reading a quote: the full cost over three years
A quote shows only the subscription. The cost to compare with the break-even point has five lines.
| Line | What to ask for or estimate |
|---|---|
| Subscription or licences | The price per device, per user or by volume, servers included. The price for the second and third years. |
| Set-up | Deployment, configuration, training. These costs are often billed once: spread them over three years. |
| In-house operation | The days spent each year tuning the tool and handling its alerts, multiplied by the cost of a day. |
| What the measure replaces | Licences or services that stop. Deduct them. |
| Exit | Recovery of data and logs, notice period, end-of-contract fees. |
The example that follows applies this method to an imaginary company. Replace its figures with your own.
The fictional example: doing nothing also has a cost
Take a fictional distributor with 80 employees, €12 million in revenue, 100 workstations and servers, and €250,000 in available cash. Its IT technician works during the day and receives alerts on their phone in the evening.
Management is concerned about two events.
| Scenario | Annual probability (estimated) | Cost if it happens | Expected annual loss |
|---|---|---|---|
| Ransomware, ten days of downtime | 8% | €400,000 | €32,000 |
| Data theft without encryption | 5% | €150,000 | €7,500 |
| Total | €39,500 | ||
With the formula, the €400,000 for ransomware breaks down into 10 days of downtime at €30,000 and €100,000 in fixed costs.
The columns have different levels of certainty. You can calculate the cost of an outage with sales, the warehouse and the finance director: lost margin, overtime and recovery costs. Probability remains an estimate.
The company will never receive a €39,500 bill. In most years, there is no attack and the loss is zero. In the year it happens, the bill reaches €400,000, more than the cash available. Apply two tests: does the spending reduce expected loss by more than it costs, and can the company survive the worst scenario?
Before buying: three checks that mostly cost time
None of the three products in this guide makes up for remote access protected by a password alone. In the Sophos 2026 report [1], 67% of intrusion causes were identity-related.
The CISA ransomware guide [2] calls for phishing-resistant multi-factor authentication on email, VPNs and accounts that access critical systems. It also calls for patching internet-facing servers as a priority.
| Check | The question to ask | What to fix |
|---|---|---|
| Second factor | Can email, the VPN or remote desktop be opened with a password alone? | If so, enabling a second factor on those access points comes before any purchase. |
| Patches | Which devices are visible from the internet, and when were they last updated? | If nobody can answer, the inventory is the first task. |
| Administrator accounts | Does an administrator read email with the account that administers the servers? | If so, separate the accounts. |
These checks are not part of the calculation that follows, because I have not put a figure on their effect. They mostly take staff time.
Immutable backups first: they reduce the worst-case loss
An immutable backup cannot be changed or deleted during a set period. It does not prevent the attack. It reduces the cost by letting the company recover without relying on the attacker.
Attackers target that recovery route. In a survey commissioned by Sophos [3] of 2,974 victims, 94% reported attempts to compromise their backups. Median recovery costs were $3 million when those attempts succeeded and $375,000 when they failed. Sophos sells these services, and the difference may also reflect weaker security overall.
Storage accounts for little of the bill. Ten terabytes protected, or 22 terabytes stored with history, cost about $1,800 a year at Backblaze B2’s public price [4]. The rest is software and staff time: the backup licence and two restore exercises a year. Our company budgets €6,000 a year in total.
The example assumes that tested recovery reduces the cost of ransomware from €400,000 to €150,000. Only a timed exercise can confirm it. The cost of data theft stays the same, since restoring cannot retrieve data that was copied.
The expected loss falls to €19,500, so the break-even point is €20,000 a year. After deducting the €6,000, the net benefit is €14,000 a year. The worst case costs less than the cash available. Both tests pass.
Cost of a ransomware attack in the fictional example
The €150,000 is an assumption. Only a timed exercise can confirm it.
What to demand
The word “immutable” covers uneven levels of protection.ANSSI [5] points out that the strength of the mechanism varies with the technology used. Five points separate a backup that holds from a backup that merely reassures.
| Criterion | Why | How to check it |
|---|---|---|
| A lock the administrator cannot lift | At AWS [6], “governance” mode can be bypassed with a special permission. Only “compliance” mode forbids deletion by anyone, the root account included. Ask your provider for the equivalent setting. | Attempt a deletion on a test set, with the most privileged account. |
| Backup accounts outside the production directory | An attacker who takes the directory takes everything that authenticates against it. ANSSI calls for independent authentication and dedicated administrator accounts. | The backup server is not a member of the production Windows domain. |
| An offline copy | ANSSI considers it essential, even if less frequent, and more robust than an online lock. | A medium is physically disconnected, and someone knows where it is. |
| A timed restore, in a written order | ANSSI calls for a procedure that is carried out regularly, and an order that follows dependencies: directory and DNS before applications. | Date of the last exercise, time measured, name of the person who ran it. |
| A backup of the backup system | Without the catalogue, the encryption keys and the installation media, the data exists but cannot be restored. | These items are kept offline, with the import procedure. |
Three possible set-ups
| Set-up | Its advantage | Its limit |
|---|---|---|
| Locked object storage with a hosting provider | Quick to set up, inexpensive by volume. | Restoring depends on internet bandwidth. ANSSI recommends encrypting before sending and checking that the data stays in the European Union. |
| Locked repository on site | Restoring does not depend on the internet link. | It stays connected to the network. The strength of the lock, software or hardware, needs to be examined. |
| Disconnected media: tapes, disks taken off site | The most robust according to ANSSI. | Regular handling, less frequent backups, time needed to bring the media back. |
ANSSI describes an acceptable compromise: regular backups to locked storage, and a less frequent offline copy. That is the set-up I would choose for our distributor.
What an attacker in control of the network can still reach
The proof: an exercise, not a certificate
Restore a business server and the directory on isolated hardware, and time it. The measured time replaces the assumption in your calculation. If it is twice as long as expected, the cost of the worst case rises and the survival test must be redone.
EDR next: it addresses risks that backups cannot
EDR monitors workstations and servers, detects suspicious behaviour and can isolate a device. It reduces the probability that an attack succeeds, including when data theft is the aim.
Public prices vary widely and cover different things: $36 per user per year from Microsoft [7], with servers as an add-on, and $184.99 per device from CrowdStrike [8] for Falcon Enterprise. The licence is only part of the cost. Someone has to configure the tool and read its alerts.
Our company estimates €6,000 for licences and €5,000 to operate it, or about 15 days of staff time. Total: €11,000 a year.
The example assumes EDR halves the probabilities. TheMarsh McLennan study [9] suggests a similar order of magnitude: a factor of 2.23 between organisations with and without advanced endpoint protection. This is a correlation based on US data from 2020–2021.
What to demand
An installed EDR is not necessarily an EDR that protects. Five settings make the difference, and none of them appears on the quote.
| Criterion | Why | How to check it |
|---|---|---|
| The whole estate covered | CISA [2] recommends EDR on all assets. One forgotten server is enough as a starting point, and some offers charge for servers separately. | Compare the machine inventory with the list in the console. |
| Blocking, not just alerting | In detection-only mode, the tool reports the attack without stopping it. At night, nobody reads the report. | The policy applied to workstations and servers is in blocking mode. |
| Protection against being disabled | Attackers often try to switch off security tools, notes Microsoft [10]. With this vendor, the protection can block on Windows and macOS, but only reports on Linux, in preview. | Try to stop the agent with a local administrator account, on a test machine. |
| Remote isolation | Cutting a machine off from the network limits the spread. Someone still needs the right to do it. | The names of the people authorised, and a trial on one workstation. |
| Enough history | After an incident, you need to be able to trace what happened. How long events are kept must be checked in each offer. | The retention period is written into the contract. |
The EDR you already pay for, or another one?
If an existing subscription already includes EDR, its cost comes down to operating it. Check three points before buying anything else:
- it covers your servers and your non-Windows systems;
- your team or your provider knows how to tune it;
- the MDR service you are considering can monitor it.
Another vendor is justified if one of these three points fails.
The proof: a simulation
Ask your provider for a supervised attack simulation on a test machine. Measure three things: whether the tool blocked, who received the alert, and how long it took someone to react.
The two basic safeguards make sense together
After backups are in place, EDR reduces a loss that has already been halved. Its benefit falls accordingly.
| Measure | Full annual cost | Expected loss afterwards | Net annual benefit |
|---|---|---|---|
| Immutable backups | €6,000 | €19,500 | +€14,000 |
| EDR added after backups | €11,000 | €9,750 | −€1,250 |
| Both together | €17,000 | €9,750 | +€12,750 |
EDR breaks even at a full cost of €9,750. The €1,250 gap is smaller than the uncertainty in the assumptions. If the licence is already included in a subscription, the cost drops to €5,000 and net benefit rises to €4,750.
I would keep it for another reason: of the two measures, it is the only one that addresses data theft.
The safeguards hold up under several assumptions. If both measures are half as effective, the benefit is still €375. If the probabilities are half as high, it falls to −€2,125 and only the survival test justifies the spending. If the probabilities double, it reaches €42,500.
SOC or MDR: the additional cost decides
EDR generates alerts and someone needs to read them. The Sophos 2026 report [1] found that 88% of ransomware deployments occurred outside business hours across 661 cases. Sophos sells these services, but the question remains: who responds at 3 am?
Compare MDR with what it replaces. The service outsources monitoring and first response, often including EDR. Our company receives an €18,000 annual quote. It replaces €6,000 in licences and leaves €3,000 in internal operating costs instead of €5,000. The actual additional cost is €10,000, rather than €18,000.
The example assumes MDR further reduces probabilities by 25% and incident costs by 20%. Expected loss falls from €9,750 to €5,850, saving €3,900.
What MDR replaces is deducted before comparing (fictional example)
The additional cost exceeds the loss avoided: at this price, MDR does not pass the average test.
| Situation | Loss avoided by MDR | Additional annual cost | Net result |
|---|---|---|---|
| IT technician responds at night, €18,000 quote | €3,900 | €10,000 | −€6,100 |
| Nobody reads alerts at night, €18,000 quote | €8,775 | €10,000 | −€1,225 |
| IT technician responds at night, €10,800 quote | €3,900 | €2,800 | +€1,100 |
The second row assumes that EDR without overnight monitoring reduces probabilities by only 25%. MDR adds more here because it restores some of EDR’s effectiveness. The remaining gap is about €100 a month.
The third row uses Huntress’s public price [12]of $8.99 per endpoint per month, treated as euros for a rough comparison. It covers endpoints. In the Sophos cases [1], 67% of intrusion causes were identity-related. A quote that also covers email and accounts will cost more.
In this example, MDR breaks even at an additional €3,900 a year, or €325 a month. At €10,000, an attack would still have to cost €385,000 after recovery.
Per year, in the fictional example, each measure added to the previous one
Each step avoids less than the previous one, because it acts on a loss that has already been reduced.
What to demand from the contract
Two offers at the same price can have almost nothing in common. Six clauses decide what the service really reduces.
| Clause | What it changes | What to read in the contract |
|---|---|---|
| The scope monitored | A service limited to endpoints does not see a hijacked email account. | The list of sources: workstations, servers, accounts, email, cloud. |
| The right to act | A provider that must reach you before isolating a machine loses the time you are paying to save. | The actions authorised without your approval, at any hour: isolating a machine, disabling an account. |
| The committed response times | “24/7” says nothing about reaction time. | A stated time between alert and action, the same at night and at weekends. |
| Your share of the work | Some decisions remain yours: stopping a production server, warning a customer. | Who is called on your side, and at what time. Without an on-call rota, plan a written delegation. |
| The confirmed incident | Monitoring and full response are sometimes two separate services. | What is included after the first response: investigation, restoration, and at what rate. |
| The exit | The logs and the history will be useful to you after the contract. | The format in which data is handed back, the notice period, and whether the EDR tool remains yours. |
Three ways to outsource
| Option | What it covers | The question to ask |
|---|---|---|
| EDR managed by your IT provider | Tuning the tool and following up alerts. | Does it monitor at night and at weekends, or only during business hours? |
| MDR | Continuous monitoring and first response, often with the tool included. | On what scope, and with what right to act? |
| Outsourced SOC | Collecting and analysing logs beyond endpoints: network, servers, applications. | What do log collection and storage cost, on top of the subscription? |
ANSSI qualifies detection providers, known as PDIS, and publishes their list [13]. Nothing obliges a small business to choose one. The list serves as a point of comparison.
Depending on your situation: where to start
| If… | Then | Why |
|---|---|---|
| Remote access or email opens with a password alone | Fix it before buying anything. | None of the three products makes up for that door. |
| No restore has been timed for a year | Run the exercise before comparing quotes. | Without a measured time, the cost of the worst case is a guess. |
| A domain account can erase the backups | Lock and separate accounts first. | It is the measure that passes both tests in the example. |
| A subscription includes an EDR that is not deployed, or set to alert only | Deploy it and switch it to blocking. | The cost is limited to operating time. |
| Nobody reads alerts at night and at weekends | Cost an MDR on its additional cost. | This is the case where it comes close to breaking even. |
| A customer or an insurer requires continuous monitoring | MDR is required. Negotiate the scope and the right to act. | The calculation no longer decides; the contract does. |
| The company has no IT technician | Everything rests on the provider’s contract. | Check who restores, how quickly, and who responds at night. |
| You hold data whose theft would be serious | Identity and EDR come first. | A backup does not take back what has been copied. |
The three checks at the start can be done straight away: they need almost no budget. For spending, with a limited budget, I would keep this order: backups, identity and patching, EDR, then MDR.
The decision and what would change it
Management commits €17,000 a year to backups and EDR. The worst case falls to €150,000, below available cash. Management holds off on the €18,000 MDR offer and asks for a quote with a precise scope.
Four facts could reverse that decision regardless of expected loss:
- a customer or insurer requires continuous monitoring
- nobody is available to act at night or on weekends
- the company holds data whose theft would cause serious harm
- a recovery exercise shows restoration takes much longer than expected.
Before signing, I would also check who can disconnect what, and when. MDR with no authority to act will have little impact.
Before signing an insurance policy, check exclusions, deductibles and the security conditions declared to the insurer.
Each step comes back to the same question: what does this expense add to the previous one, and what does it cost?
What the calculation alone cannot decide
Some spending is required. Article 32 of the GDPR [14] requires security appropriate to the risk, taking account of implementation costs. It also calls for regularly testing the effectiveness of safeguards where appropriate.
Contracts can also decide the matter: a customer requirement, an insurer’s condition or a clause with a provider.
This guide does not provide legal or insurance advice.
Questions to ask before signing
This page does not receive your answers. Checking a box does not constitute an audit.
For reference: what companies spend according to their size
No robust source says what percentage a company should aim for.ENISA [16] measures spending without setting a target. CIS [17] only estimates that the tools for its IG1 baseline cost less than 20% of IT spending, regardless of company size.
We have observed spending levels.
| Organisations surveyed | Share of IT spending | Source |
|---|---|---|
| Less than $50 million in revenue | 26.1% | IANS and Artico, 2024 data [18] |
| Less than $100 million | 21.6% | IANS and Artico, 2026 [19] |
| $600 million to $1 billion | 11.6% | IANS and Artico, 2024 data [18] |
| Over $10 billion | 7.7% | IANS and Artico, 2026 [19] |
| EU critical sectors, 83% large organisations | 9% | ENISA, 2025 [16] |
The ratio falls as company size increases.ENISA [20] also finds a higher ratio at small and medium-sized firms, which it attributes to fixed baseline costs. Relative to revenue, IANS reports over 2% for the smallest firms in its sample and about 0.6% for the largest group.
These figures do not apply to every small or medium-sized business. IANS surveys CISOs, so the companies already have a dedicated security lead. ENISA mainly covers large organisations. Neither measures a French small business without a CISO.
In France, the OpinionWay survey for CESIN [21] found that 42% of respondents spend at least 5% of their IT budget on cybersecurity. It surveyed 397 CESIN members, 83% from large and mid-sized companies.
Suppose our distributor spends €360,000 a year on IT. ENISA’s 9% would mean €32,400, and IANS’s 21.6% about €78,000. The €17,000 in basic safeguards is 4.7%. Benchmarks provide context, while the risk calculation guides the decision.
Where cybersecurity budgets go
The most useful breakdown I found comes from Deloitte [22] : 53 financial institutions detailed their cybersecurity spending in June 2023.
This chart does not break down the entire IT budget. It shows only the portion spent on security.
That portion is 9% of IT spending in European critical sectors (ENISA) [16]. By company size, IANS and Artico [18] [19] report 26.1% below $50 million in revenue, 21.6% below $100 million, 11.6% between $600 million and $1 billion, and 7.7% above $10 billion.
With security at 9% of IT spending, EDR’s 20% of the security budget would therefore represent about 1.8% of the IT budget.
9%: ENISA, 2025. Breakdown: Deloitte, 2023 (finance), regrouped by the author
Total IT budget
Security 9%Rest of the IT budget: 91%
Breakdown of that 9%, scaled to 100%
Share of the security budget
- EDR, endpoints and network20%
- MDR, detection and response33%
- Immutable backups and data7%
- Identity and MFA11%
- Other30%
The pie chart covers only security spending. The 9% benchmark is from ENISA in 2025, for a different population from Deloitte’s. The breakdown is the author’s interpretation of Figure 2 of Deloitte’s 2023 Cybersecurity for financial services survey. Deloitte does not report separate shares for EDR, MDR or immutable backups. Each share covers the entire area that includes them. The shares add up to 101% due to rounding.
The grouping reflects my interpretation. The percentages come from Deloitte, but I chose how to group them. EDR’s share, for instance, is smaller than the 20% shown because that area also covers networks and industrial systems. Likewise, MDR’s 33% includes all internal detection as well as cloud and applications.
| Chart category | Deloitte areas grouped together | Why they were grouped |
|---|---|---|
| EDR, endpoints and network | Infrastructure and network | A 2020 Deloitte survey grouped endpoint and network security |
| MDR, detection and response | Threat detection and response, emerging technologies and cloud, and applications | The author groups these environments with detection and response |
| Immutable backups and data | Data protection and privacy | Closest match. Some spending may belong to infrastructure. |
| Identity and MFA | Identity and access operations | Direct match |
| Other | Strategy, talent and governance, third-party security, and other | Items outside this article’s scope |
The figures describe heavily regulated financial institutions. They do not set a target for a small business in another sector.
Other recognised organisations add context:
- Wavestone [23] : : across more than 200 organisations, cybersecurity accounts for 6.7% of IT spending. Incident recovery is the least mature function, at 44% compared with 54% to 57% for the others.
- PwC [24] : : among 1,740 security leaders, 21% put managed services in their top three investment priorities, 18% identity and 15% endpoint protection.
- McKinsey [25] : : small and mid-sized companies pay three to five times more per employee than large companies, depending on the tools. They expect to rely on MDR providers for advanced tasks.
- Forrester [26] : : in Europe, 63% of decision-makers plan to increase their use of managed security services.
- Gartner [27] : : in 2020, its analyst said a range of 1.7% to 11.5% of IT spending was neither good nor bad in itself. The company’s circumstances matter more.
For insurance,AMRAE [28] gives indicative average premiums for 2025: €645 below €2 million in revenue, €1,600 for €2 million to €10 million, €5,000 for €10 million to €50 million and €34,000 for mid-sized companies. The figures for the smallest organisations are the least reliable.
Sources and dates
Checked on 9 November 2026, except sources 2, 5, 6, 10 and 13, checked on 12 November 2026. The businesses, internal amounts and probabilities in the example are fictional. Public prices are in US dollars before tax and are not quotes.
- Sophos, Active Adversary Report 2026, cases from 1 November 2024 to 31 October 2025. ↩
- CISA, #StopRansomware Guide. ↩
- Sophos, The impact of compromised backups on ransomware outcomes, Vanson Bourne survey, early 2024. 2,974 victims. ↩
- Backblaze B2 pricing. Public price. ↩
- ANSSI, Sauvegarde des systèmes d’information : les fondamentaux, ANSSI-BP-100, version 1.1 of 27 November 2025. ↩
- AWS, Amazon S3, Locking objects with Object Lock. Vendor documentation. ↩
- Microsoft Defender for Business pricing. Public price. ↩
- CrowdStrike Falcon pricing. Public price. ↩
- Marsh McLennan, Using data to prioritize cybersecurity investments, 2023. Correlations based on US data from 2020–2021. ↩
- Microsoft, Tamper protection overview, Defender for Endpoint, page updated on 9 September 2026. Vendor documentation. ↩
- Insee, statutory working time. ↩
- Huntress pricing. Public price. ↩
- ANSSI, Prestataires de détection d’incidents de sécurité (PDIS). ↩
- CNIL, GDPR, Chapter IV. ↩
- ANSSI, MesServicesCyber, NIS 2. ↩
- ENISA, NIS Investments 2025, 1,080 organisations, interviews from May to August 2025. ↩
- CIS, The Cost of Cyber Defense: IG1. ↩
- IANS and Artico, budget data for small and mid-sized companies, 2024 survey published in 2025. ↩
- IANS and Artico, 2026 Security Budget Benchmark, summary article. ↩
- ENISA, NIS Investments 2024. ↩
- OpinionWay survey for CESIN, fieldwork from 17 November to 12 December 2025. ↩
- Deloitte, 2023 Cybersecurity for financial services survey, June 2023, Figure 2. ↩
- Wavestone, Cyber Benchmark 2026. ↩
- PwC, 2026 Global Digital Trust Insights, May to July 2025. ↩
- McKinsey, Cyber Market Map, October 2022. ↩
- Forrester, Security Planning 2026, July 2025. ↩
- Cybersecurity Dive, Gartner analyst remarks, October 2020. ↩
- AMRAE, LUCY 2026, 2025 data. ↩