Choosing a security provider is a decision you make against evidence you can verify, not against a brand you recognize. The single most useful discipline is to force every vendor claim down to something checkable: an independent test result, a published standard, an insurance limit, a contract clause. Everything below is about how to do that, in the order the decisions actually come up.
Start with the independent test, then read the fine print behind it
For consumer antivirus on Windows 11, AV-TEST is the reference point most people should start from, because it publishes a fixed maximum and tells you who hit it. As of 2024-07-01, AV-TEST awards a maximum of 18 points per certification test, and a product needs at least 17.5 out of 18 overall to earn its "TOP PRODUCT" designation. In the March–April 2024 home-user comparison on Windows 11, AV-TEST tested 17 products and 11 of them reached TOP PRODUCT.
That last number is the one that matters for your decision, and it is the one vendors never quote. When 11 of 17 products clear the same bar, the bar stops discriminating. You are no longer choosing between a good product and a bad one; you are choosing among a group that has already demonstrated comparable detection. That reframes the whole purchase. Detection being roughly equal, your decision should turn on the things AV-TEST does not score: how much the product slows the machine, how aggressively it upsells, whether the renewal price triples, how the support line behaves when something goes wrong, and whether the company has a history of quietly bundling things you did not ask for.
The limits are worth stating plainly. A certification test is a snapshot of a test run on a particular platform in a particular window; it tells you about detection and performance under lab conditions, not about the vendor's billing practices or its response to a vulnerability disclosure. It also does not transfer across platforms — a strong Windows result says nothing about the macOS or Android product, which are often built by different teams or licensed from someone else entirely. Use the score to build a shortlist, then decide on the operational stuff.
Match the provider type to the threat you actually have
The word "provider" hides at least four different purchases, and people routinely buy one while needing another. Get the category right before you compare names inside it.
- Endpoint antivirus and EDR. For malware on devices you control. The consumer question is the AV-TEST shortlist above; the business question is whether you need detection-and-response telemetry and someone to read it, which is a staffing decision as much as a software one.
- VPN services. NIST, in its guide to IPsec VPNs, defines a VPN as a virtual network built on top of existing physical networks that provides a secure communications mechanism for data and information transmitted between networks or between different nodes on the same network. That is a description of a tunnel, not of anonymity. If your actual problem is "our staff need to reach internal systems from outside the office," you are buying an IPsec gateway, not a consumer privacy subscription.
- Cyber insurance. A financial backstop, and the one place where the provider's limits are written down in numbers you can read in advance. More on this below, because the numbers are sobering.
- Managed services and consultancies. You are buying people and process. The provider's own security posture and its contractual obligations to you matter more than any product it resells.
The cost of getting this wrong is usually paid twice: once for the thing you bought, and again for the thing you actually needed after an incident. A small firm that buys a consumer VPN subscription to "secure remote work" has bought encryption for traffic in transit and nothing else — no device management, no identity checks, no logging anyone will read.
VPN choices: the split-tunneling decision is where the risk lives
If you are selecting a VPN provider or configuring one, the configuration question outranks the vendor question, and the default is usually the less safe option. NIST describes split tunneling as a remote device establishing a non-remote connection with a system while also communicating via some other connection to a resource in an external network — in practice, work traffic goes through the tunnel and everything else goes out the ordinary internet connection.
The popular intuition is that split tunneling is safer because less traffic passes through the VPN, reducing load. NIST says the opposite: split tunneling introduces security complications, can expose traffic outside the VPN, and is strongly discouraged for IPsec VPN clients. Its telework guidance is explicit that split tunneling protects only internal-resource traffic through the VPN while excluding other communications from the tunnel, which limits an organization's ability to inspect and protect that traffic. NIST also acknowledges the trade-off honestly — preventing split tunneling is more secure, while the default setting that allows it is a balance between security and functionality, and split tunneling can reduce load and improve efficiency.
So the decision is a real one, not a formality. Full tunneling costs you bandwidth and latency at the gateway and can be painful on poor home connections; split tunneling buys that performance back by moving traffic outside the inspection boundary. For a provider selection, ask directly whether split tunneling is on by default, whether it can be disabled centrally, and whether the client can be forced into full tunnel. A vendor that cannot answer those three questions in writing has told you something useful.
One correction worth carrying into any conversation with a sales engineer: NIST states that the most common use of IPsec implementations is providing VPN services. IPsec is not a niche optional feature bolted onto a VPN product — it is the mainstream mechanism. If a provider describes its transport vaguely and cannot say whether it is IPsec-based, that is a gap, not a simplification.
Insurance limits: read the number, then read it against the loss
When the provider is an insurer, the only figure that matters is the limit, and the useful exercise is to hold it next to a realistic loss. Public-sector incidents give unusually clean numbers because the costs get reported.
The City of New Orleans carried a cyber insurance policy limit of $3,000,000 at the time of its 2019 ransomware breach, according to reporting dated 2019-12-16. The total estimated cost of that December 2019 attack was $7,200,000, and the city expected cyber insurance to cover $3,000,000 of it. The gap between the limit and the loss is the part the policyholder absorbs.
Elsewhere the same month, the City of Pensacola reportedly faced a ransom demand of $1,000,000 in its December 2019 ransomware incident, per reporting dated 2019-12-26. By 2024-09-24, in a request for proposals for risk broker services, Pensacola specified a minimum cyber liability insurance limit of $3,000,000, alongside a professional liability minimum of $1,000,000 per occurrence.
| Entity and date | Figure | What it tells you |
|---|---|---|
| New Orleans, 2019 breach | $3,000,000 policy limit | The cover available at the time |
| New Orleans, 2019 attack | $7,200,000 total estimated cost | The loss the limit was measured against |
| New Orleans, 2019 claim | $3,000,000 expected from insurer | The limit was reached, not the loss |
| Pensacola, Dec 2019 | $1,000,000 ransom reportedly demanded | Scale of a single demand |
| Pensacola RFP, 2024 | $3,000,000 minimum cyber limit | What a buyer later required |
| Pensacola RFP, 2024 | $1,000,000 per-occurrence professional liability | Coverage bought alongside cyber |
What this means for choosing an insurer is less about the carrier's name and more about the structure: the limit, the retention, the sublimits for things like ransomware or notification costs, and the conditions — insurers increasingly require specific controls before they will pay. The honest caveat is that these are public-sector examples with published figures, and a small business's loss profile differs; the method transfers, the numbers do not. Read your own policy's exclusions before you compare premiums, because the cheap quote is often cheap in the sublimits.
Verify the claims that vendors make about themselves
Every provider you shortlist will present third-party validation, and most of it can be checked in minutes. The check is not cynicism; it is the difference between a marketing department's paraphrase and the underlying result.
- Test scores. Go to the testing body's own site and find the specific test, date, and platform. A TOP PRODUCT badge from an older test window, or from a different operating system, is not the same claim. AV-TEST's 18-point maximum and 17.5-point threshold are precise; a vendor quoting "top scores" without the number and the date is telling you less than it could.
- Standards and definitions. When a vendor says "military-grade encryption" or "enterprise-grade tunnel," replace the phrase with the standard. NIST's definition of a VPN is narrow and useful: a virtual network built on existing physical networks providing a secure communications mechanism. If the product does not do that, it is not a VPN regardless of the label.
- Certifications and audits. Ask for the report, not the logo, and check the scope — an audit of one data center says nothing about the others.
- Insurance and financial backing. For insurers and managed providers alike, ask for the limit or the balance sheet in writing. The Pensacola RFP is a good model of how to do this: it specified minimum limits before anyone was selected.
- References and incident history. Ask what happens after a breach, who is on call, and what the notification timeline is. A provider that has never had an incident either is new or is not telling you.
The cost of this diligence is time, and it is real — a thorough comparison of five providers can eat a week. The trade-off is worth naming: for a personal antivirus purchase, the AV-TEST shortlist plus a look at renewal pricing is enough, and deeper diligence is wasted effort. For anything with a contract, a data processing agreement, or a coverage limit, the diligence is the purchase.
Price the whole relationship, including the exit
The quoted price is the smallest part of what a provider costs you, and the exit terms are the part people discover too late. Before you sign, get the renewal price in writing — not the introductory one — along with the notice period for cancellation and what happens to your data when you leave.
Three things go wrong predictably. First, introductory pricing that renews at a much higher rate, common in consumer antivirus and VPN subscriptions, which is why the renewal figure belongs in your comparison table next to the first-year figure. Second, term lock-in with auto-renewal buried in the agreement, so cancelling requires a phone call inside a narrow window. Third, data egress and migration costs that make leaving expensive enough that you stay — the lock-in is technical rather than contractual, and it is just as effective.
There is also a coverage question that belongs in the price comparison rather than after it. If the provider is an insurer, the premium is meaningless without the limit and the retention beside it; a policy with a $3,000,000 limit and a large retention is a different product from a cheaper one with a smaller limit, even at the same headline premium. If the provider is a managed service, the price should be read against what is contractually promised — response times, staffing, and who carries the liability if something is missed.
The honest limit of all this: no amount of provider selection substitutes for the basics you control. Patches, backups you have actually restored from, multi-factor authentication, and someone who reads the alerts will outperform a better vendor chosen badly. Choose the provider to remove a specific gap, write down what you expect it to do, and check that expectation against the contract before the invoice is paid.
When the provider is the tunnel: configuration clauses worth demanding in writing
Ask a VPN vendor whether its client can be forced into full tunnel and whether split tunneling can be disabled centrally, and treat a vague answer as a "no." NIST's IPsec VPN guidance strongly discourages split tunneling for clients because of its security complications, and its telework guidance notes that split tunneling protects only internal-resource traffic while excluding other communications from the tunnel — which limits an organization's ability to inspect and protect that traffic. The trade-off is genuine: NIST acknowledges split tunneling can reduce load and improve efficiency, and that the default setting allowing it is a balance between security and functionality. What you are buying with full tunnel is the inspection boundary; what you are paying is gateway bandwidth and latency. Put the enforcement mechanism — central policy, forced full tunnel, no user override — in the contract, not the sales deck.
Two questions that eliminate most weak providers
Ask what standard the tunnel is built on and what the insurance limit is, and make both answers specific. NIST states that the most common use of IPsec implementations is providing VPN services, so a provider that cannot say whether its transport is IPsec-based has a gap, not a simplification. On coverage, the Pensacola RFP dated 2024-09-24 is a usable template: it specified a minimum cyber liability limit of $3,000,000 and a professional liability minimum of $1,000,000 per occurrence before selecting anyone. Compare that discipline to New Orleans, which carried a $3,000,000 cyber policy limit at the time of its 2019 breach against a total estimated attack cost of $7,200,000 — the limit was reached, not the loss. A provider that answers both questions precisely is telling you how it operates; one that deflects has answered too.






