Skip to content

Systems

IT assessment checklist: what to inspect, in order, before switching providers

Five passes, identity to contracts, each read from the systems rather than from the people who run them.

Computing America IT. Help desk support, backups and network infrastructure.

Send the inventory
6 min readComputing America
A flashlight beam on a dusty network rack of tangled cables in a server closet.

In short

  • The gap between the estate as remembered and the estate as it is, is the finding. Anything gathered by asking has to be confirmed against the system that would know.
  • Run the passes in order: later questions are about the accounts the identity pass finds, and the contracts pass reads best once the other four show what the contracts cover.
  • "We have backups" is four claims: a job that runs, data that can be retrieved, a system that boots from it, and a copy an attacker holding your credentials can't delete.
  • The dependency map is the output people underrate. It shows which of the things found the business depends on and which are merely present.
  • A review is a snapshot with a shelf life. It can't see anything seasonal, anything that needs a vendor's cooperation to observe, or the estate under a load it isn't carrying now.

Inspect five things, in this order: identity, assets, network, data survival and contracts. Every pass starts from an export or a test, not an interview.

To price running an estate you must know what's in it, and the record you have is the one the firm you're considering leaving kept. Our review rebuilds the record from the estate itself. It's a fixed fee, one to two weeks, and it ends with three documents you keep whether or not you go further: an inventory, a dependency map and a risk register. This piece gives each pass's method.

The one rule: read the estate, not the memory of it

Where does a finding come from? Every useful one is a difference between what the systems say and what people believe. Ask an operations manager how many laptops the company has, then count what the directory and the network see. The difference is the ones in drawers, the ones the sales team took home, and the replaced ones still on the license count.

Still do the interview. It tells you what breaks and what people work around, and it gives you questions to check. But a review that ends in a document assembled from conversations has measured the organization's beliefs about itself.

Five passes, and why identity is first and contracts last

PassThe questionWhat it typically finds
1. IdentityWho exists, what can they reach, and who administers the thing that decidesAccounts of people who left, service accounts with no owner, administrative rights granted for one afternoon and never taken back.
2. AssetsWhat devices, licenses and subscriptions exist, and who they are assigned toA license count that doesn't match headcount in either direction, endpoints silent for months, a system a department pays for directly.
3. NetworkHow the sites connect, what the rules are, and whyFirewall exceptions with no comment and no owner, a VPN configuration untouched since it was stood up, and one path everything depends on.
4. Data survivalWhat happens if a system is gone tomorrow morningBackups that run but cannot be restored, or can be deleted with the estate's own credentials.
5. ContractsWho holds the agreements, under whose name, with what noticeAuto-renewals inside their notice window, circuits contracted to the provider and not to you, a domain registered to someone who has left.
The five passes, what each one asks, and what it typically finds

Identity goes first because it defines the population. Every later question (which endpoints matter, which mailboxes hold data, which accounts a firewall rule was written for) is about principals. Contracts go last because they only mean something once the four passes before them show what the contracts are for.

The identity pass

This pass rests on three enumerations any directory can produce and few estates run, because they produce work.

  • Every privileged account, cross-referenced against last sign-in. A dormant account with standing administrative rights is a high-value target, and nothing about it generates a ticket.
  • Every account that isn't a person: service accounts, shared mailboxes, integration identities, the account the backup software runs as. For each, who owns it, what it is for, and whether its credential is ever rotated. Expect one whose purpose no one can state.
  • Every guest, federated or external identity with access to anything. Former contractors and wound-down vendors persist here, because offboarding is written for employees.

None of it needs specialist tooling. Somebody outside the arrangement should run it, because an incumbent enumerating the standing access it granted itself is filing a finding against its own housekeeping.

The asset, network and contract passes

  • Assets. Export the device list from the RMM or endpoint manager, the computer objects from the directory, and the lease table from DHCP. Reconcile the three. A device in the leases and nowhere else is unmanaged. A device in the RMM with no check-in for a period you choose (30 days, for example) is retired or lost. Check each license portal's assigned users against current staff.
  • Network. Export the running configuration of each switch and router, and the full rule base of each firewall, with hit counts if the firewall keeps them. Read the rules for any/any entries, rules with no comment or owner, and rules with zero hits over a period you choose. Check remote-access accounts against the identity pass, and compare each switch's MAC address table with the asset list.
  • Contracts. Pull every recurring vendor payment from accounts payable, then the carrier, registrar and software portals behind them. For each, record the account holder's name, the renewal date and the notice period. Read the notice dates first, because they run on a clock.

Data survival is four claims wearing one word

"We have backups" bundles four claims, and each can be true while the next is false.

Fig. 1
  1. Backup administered from the domain

    Estate credentials can delete it or shorten retention

    Within: Reachable over the network.
  2. Backup in a separate account

    Survives only if no estate credential works there

    Within: Reachable over the network.
  3. Offline encrypted copy

    No network path for any credential to use

    Within: Out of reach.

Claim 4: a stolen credential can delete a backup run from the domain, a separate account holds only if no credential is shared, and an offline copy survives a network-borne one.

SourceThe offline-copy distinction follows CISA and MS-ISAC's #StopRansomware Guide.

  1. 1.A job runs and reports success. Read the job history, don't summarize it. The dashboard answers only this claim.
  2. 2.The data can be retrieved. Pull a file out and open it. Failures to look for: retention shorter than believed, or a share dropped when a server was rebuilt.
  3. 3.A working system rebuilds from it. Restore to isolated infrastructure and boot it. A failure to look for: configuration and state missing from the backup set.
  4. 4.It survives a credentialed attacker. Check whether the credentials that run the estate can delete the copies or shorten retention. A failure to look for: backups administered from the domain they protect.

CISA and MS-ISAC's #StopRansomware Guide says to maintain offline, encrypted backups and to regularly test their availability and integrity in a disaster-recovery scenario. It adds that golden images should be kept for rebuilding critical systems, and that backups must be offline because many ransomware variants seek out accessible backups to delete or encrypt them.

The output: inventory, dependency map, risk register

The inventory lists what exists. The risk register lists what is wrong, worst first, with what each item would take to close. The dependency map between them is the artifact people underrate.

An inventory says a virtual machine exists. The map says the file share holding the production scheduling spreadsheet lives on it, that a nightly export another department relies on runs from it, and that it authenticates against the only domain controller at that site. That's how you learn which few of the objects the business can't run without. A scanner produces only part of it, because those relationships live in configuration, scheduled tasks and habits.

Order the register by how likely each item is, times what it would cost you, not by a generic severity score. The two orderings can disagree, because a generic score knows nothing about which systems your business can't run without, and the dependency map does.

What a review cannot see

  1. 1.Anything seasonal. A two-week window in August misses the month-end close, the annual audit export and the November week the warehouse runs double shifts. These go in the register as open questions, apart from findings.
  2. 2.Anything that needs a third party's cooperation. A carrier won't discuss a circuit with anyone but the account holder. If that isn't you, that's the finding, and the question gets asked before it's needed.
  3. 3.Anything under a load the estate isn't carrying now. A review can tell you a link saturates at four in the afternoon. It can't tell you what happens the first day the whole staff is in the building after a change in how they work.

The review is also a photograph, and estates drift: a new site, a departing administrator, a vendor's platform change. Keep the inventory and register current as the work happens.

Next step

Send us your own inventory.

Write down what you believe you have: the sites, the servers, the circuits, and who holds the domain. We will tell you which items in this piece that list already answers, and which gap we would open first.

Reply
A person replies, not a sequence: within one business day, from someone who would be on the engagement.

Packages

Start with a package

Four packages, each with its price on the page. Pick one and buy it online.

Billing