Shortlists for external attack surface management look almost identical on paper. Every proposal promises continuous discovery, accurate attribution, risk-based prioritisation and a tidy integration with your service desk. The decks differ mainly in colour. So choosing an EASM vendor is not really an exercise in comparing feature tables, because every feature is claimed by everyone; it is an exercise in asking a handful of questions whose answers cannot be faked in a slide.
What follows is seven of them, in the order they tend to matter, along with what a weak answer sounds like. One of the seven is a test you run yourself, and it is worth more than the other six combined.
1. Can you find our estate from one seed domain alone?
Give the vendor your primary domain. Nothing else. No asset list, no IP ranges, no export from your CMDB, no helpful nudge towards the acquired subsidiary you are quietly worried about. Then let them run discovery and show you what came back.
Good discovery expands outward the way an attacker does: from the outside. Certificate transparency logs, because every publicly trusted certificate issued for your names is logged and searchable. DNS records, including historic entries that no longer resolve but still reveal naming conventions. Domain registration and WHOIS data, including the registrant details that tie a forgotten domain back to your legal entity. Hosting relationships, where one confirmed host sits inside a range that turns out to contain four more of yours. If you want the mechanics in detail, we have written about how discovery expands from a seed into an inventory.
The number that matters is not the total asset count. A big number is easy to produce and easier to pad. The number that matters is the delta in two directions: what did they find that your register does not contain, and what did they hand you that is not yours at all. The first is the value you are buying. The second is the cost. That comparison is the whole evaluation in miniature.
Ask for the evidence trail behind each discovered asset, too. A row that says “yours”with no reasoning attached is an assertion. A row that says “found in a certificate issued 14 March 2025 covering this name and your primary domain”is something an engineer can check in ten seconds.
Two answers should worry you. One is a request for your asset inventory before discovery starts. The other is a demo that shows you, with some pride, the assets you already told them about.
2. How do you decide an asset belongs to us?
Discovery produces candidates, not certainties. Any vendor who displays attribution as settled fact is not solving the hard part, they are hiding it, and the work lands on your team later when somebody has to explain why a competitor’s web server appeared in your exposure report.
There are two failure modes and they pull in opposite directions. In the first, a marketing agency registers a subdomain under your brand for a campaign, a certificate is issued, the name shows up in certificate transparency logs, and the tool files it as yours: now you are tracking an exposure you cannot patch and do not control. In the second, a genuine forgotten asset is filtered out as third-party noise because it sits on shared hosting, and the one thing you most needed to see never reaches the inventory. The second failure is the dangerous one, and it is invisible in a demo. We have covered the ownership question at length in third-party subdomain discovery.
So ask three mechanical things. Who adjudicates a disputed asset, and is it your engineer or their analyst. How is the decision recorded, with what reasoning attached. And does the decision stick, so that an asset you rejected last month does not reappear next week for somebody else to reject again. Attribution that is not remembered is just a recurring argument.
Then ask about shared infrastructure. A single public IP behind a hosting provider or a CDN edge may serve dozens of unrelated tenants. A tool that attributes by IP alone will cheerfully give you all of them. For UK organisations that will have asset inventory expectations written into regulation, this stops being a tidiness problem and becomes a reporting one, which is worth thinking through alongside the Cyber Security and Resilience Bill, currently progressing through Parliament at the time of writing. Reporting an incident on an asset you never confirmed was yours is an uncomfortable conversation.
3. How fast do you see change, source by source?
Continuous is a marketing word until somebody breaks it down per data source. Press for the refresh interval of each feed separately, because they do not move at the same speed and no honest answer is a single number. Certificate transparency entries can surface within minutes of issuance. Full service enumeration across a large IP footprint runs on a far longer cycle, for the ordinary reason that scanning takes time and bandwidth. Both are genuinely continuous. Only one will tell you about a new host on Friday afternoon.
The metric to request is time from change appearing on the internet to an alert reaching a human. Not the scan schedule. The scan schedule is the vendor’s internal business; the lag is yours.
Three changes are worth naming in the demo. A developer publishes a new subdomain late on a Friday for a Monday launch. A firewall edit makes an admin interface reachable from the internet that was previously internal only. A certificate lapses overnight and a customer-facing service starts throwing warnings. None of these is exotic. Each has a different detection path, and a tool that only catches the first one is not monitoring your surface, it is watching your DNS.
Finally, ask what happens to assets that disappear. A host that was decommissioned in March and still sits in the inventory in October inflates your numbers, your risk picture and, if the licensing works a certain way, your bill. Retirement logic is the least glamorous thing in the product and one of the most revealing, which is part of why a point-in-time scan falls short against a surface that moves.
4. What do you report beyond CVEs?
CVE counts are the easiest finding type to demo and frequently the least useful part of an external view. They populate a dashboard nicely. They are also the thing your existing vulnerability scanner probably already covers for the assets it knows about.
The external issues that actually bite tend to be duller. Misconfigurations. Management and admin interfaces reachable from the open internet. Default credentials on a service nobody remembered was listening. Certificates that have expired, that cover the wrong names, or that were issued by a provider your organisation stopped using two procurement cycles ago.
Dangling DNS deserves its own question. Picture a CNAME still pointing at a cloud storage bucket that was deleted when a campaign microsite was retired. Months later the record resolves to nothing, or worse, to a namespace somebody else can claim. Ask the vendor to show you that finding type live: does the tool flag it as takeover-prone, or does it file it as an ordinary DNS record among thousands, and what does it tell you to do about the record itself. The mechanics are set out in our piece on subdomain takeover and dangling DNS.
Ask for the prioritisation logic in writing, too. Which inputs raise a finding’s severity, which lower it, and can you reweight them for your own context. A customer payment path and a dormant brochure site should not be ranked by the same formula, and if the weighting is a black box you will spend your first quarter arguing with it.
5. Is this finding likely, or definitely?
Passive external observation infers. It reads banners, response headers and behavioural fingerprints, and from those it concludes that a system is probably running a given version and is so probably affected by a given flaw. It has not tested the flaw. That distinction is the single most important thing to establish in a demo, and we have unpacked it in likely versus definitely.
So go through the demo findings and ask, for each one, whether it was inferred or validated. Then raise the awkward case: a host behind a CDN or reverse proxy, where the banner reflects the edge rather than the origin. Version-based findings in that situation are frequently wrong in both directions, and a vendor who has thought about it will say so plainly rather than quoting you a confidence percentage they cannot justify.
If active checking is offered, get the operational detail. What is in scope and what is explicitly out. What rate limits apply. What is on the safe-check list, and who maintains it. Who signs the authorisation, and whether that signature covers assets discovered after the authorisation was given. And what the tool does when it meets a fragile legacy host that falls over if you look at it firmly. Ageing kit exists in every estate and the honest answer is usually to exclude it and handle it manually.
The reassurance a buyer is entitled to, in plain terms: external attack surface management observes what is already published to the internet. It does not exploit systems, and it does not make your organisation more interesting to attackers, who are enumerating the same public sources anyway.
6. What are we actually paying for, and what happens at renewal?
Unpick the licensing unit before you discuss discount. Per domain, per discovered asset, per IP address, per seat, or a managed service with a block of analyst hours attached. These are not variations on a theme; they produce completely different bills for the same estate.
Per-asset pricing contains a perverse incentive worth naming out loud. The supplier whose discovery works best, the one that finds the acquired company’s forgotten staging environment and the eleven microsites, also generates the largest invoice. That is not a reason to reject the model. It is a reason to get the overage mechanism in writing: what happens when you cross the threshold mid-term, whether the excess is billed immediately or trued up at renewal, and whether assets you decommission drop out of the billable count or sit there permanently.
Separate the tool from the work. A platform that produces findings you must triage assumes internal capacity you may not have. A service that validates findings and writes remediation guidance costs more and assumes less. Both are defensible. Pricing one while planning to operate the other is how an EASM programme quietly dies in month four, and it is worth costing honestly against what attack surface management actually costs you in internal time.
Then exit terms, which nobody asks about during a demo and everybody wants during a renewal negotiation. Can you export the full inventory and the complete finding history, in what format, by API or by a support ticket. What happens to your data after the contract ends, and how long is it retained. A vendor who shrugs at data portability has told you something about the renewal conversation to come.
7. What does the proof of value look like before we sign?
Design the trial yourself. One seed domain, two to four weeks, and acceptance criteria written down and agreed before access is granted. Criteria written afterwards are just a description of whatever happened.
Measure four things honestly. Unknown assets surfaced and then confirmed by your own team, not merely claimed. Attribution errors in both directions, including the forgotten asset the tool filtered out as somebody else’s. Findings actually closed, because an unremediated finding is a liability with a timestamp. And how many alerts a human had to read to reach those outcomes.
That last one is a procurement question, not an operational detail. Ask how a typical week-one backlog for an estate your size is triaged, who does the triage during the trial, and whether suppression survives a re-scan. A tool that reopens the same fifty accepted risks every Monday will be muted within a month, and a muted tool is worse than useless, because it gives false confidence.
Test the plumbing while you have the chance. Ticket creation in your service desk with the right fields populated. Ownership routing, so the finding reaches the team that runs the host rather than a shared mailbox. SIEM or SOAR integration if you have one. And whether the report your auditor asks for comes out of the product without someone spending a day in a spreadsheet.
Decide, before the trial begins, where this sits alongside your existing vulnerability scanning and internal tooling, so the pilot tests the gap it is meant to fill rather than duplicating cover you already pay for. The NCSC’s guidance on asset management makes the underlying point without ceremony: you cannot protect what you have not identified. If you want a structured starting point for the evaluation itself, our practical checklist for choosing an EASM product covers the requirement-setting stage that comes before the demos.
The Short Version
- Hand over one seed domain and nothing else, then compare the result against your register in two directions. What they found that you did not is the value. What they wrongly attributed to you is the cost.
- Attribution presented as certainty creates work. Ask for the evidence behind each asset, who adjudicates disputes, and whether a rejection is remembered next run.
- Break “continuous”down per data source and ask for time from internet change to human alert, not the scan schedule.
- CVE lists are the easy part. Make the vendor demonstrate dangling CNAMEs, exposed admin interfaces and certificate hygiene.
- Establish which findings are inferred and which were validated, especially behind a CDN.
- Get the licensing unit, the overage mechanism and the data export terms in writing before price is discussed.
- Write the acceptance criteria before the trial starts, and count the alerts a human had to read.
Choosing an EASM vendor comes down to whether the product tells you the truth about its own limits. The ones that do will answer these seven questions without flinching, and the ones that do not will offer you a feature table instead.
Frequently Asked Questions
What should I ask when choosing an EASM vendor?
Ask whether they can discover your estate from your primary domain alone, how they decide an asset belongs to you, how quickly each data source refreshes, what they report beyond CVEs, which findings are inferred rather than validated, exactly what the licence counts, and what the proof of value will look like. The pattern behind all seven is the same: push every claim down to a mechanism you can verify during a trial.
How is EASM usually priced, and what counts as an asset?
Common models include per domain, per discovered asset, per IP address, per seat, or a managed service with analyst time included. The definition of an asset varies between suppliers, which is why it belongs in the contract rather than the proposal: a hostname, an IP, a running service and a certificate can each be counted separately. Ask specifically whether decommissioned assets leave the billable count, and how mid-term overage is handled.
Can a vendor run discovery using only our main domain?
A capable one can, because the sources are public: certificate transparency logs, DNS records, domain registration data and hosting relationships. Expect the output to contain candidates rather than confirmed assets, and expect some of them to be wrong. A vendor who asks for your asset inventory before discovery begins is testing their matching, not their discovery.
Does EASM testing risk disrupting our live systems?
Passive external observation looks at information already published to the internet and does not interact with your systems in any way that should affect them. Active checking is different and needs scope, rate limits, a documented safe-check list and written authorisation. Where you have fragile legacy hosts, exclude them from active checks and handle them manually.
How long should an EASM trial or proof of value run?
Two to four weeks is usually enough to see an initial discovery pass, a week-one triage backlog and at least one genuine change event. Shorter than that and you only see the initial inventory, which is the part every vendor does well. Agree the acceptance criteria in writing before access is granted, otherwise the trial has no pass mark.
