Blog/Enterprise security

Identity and access management for a team that keeps growing

What to fix at ten people, thirty, and a hundred: one source of truth, group-based access, service accounts, session controls and reviews that actually happen.

CyberWatch AI21 November 2026 · 7 min read
An open-plan office with several desks, laptops and an empty chair

Here is a test you can run this afternoon. Ask HR for a list of everyone who has left in the past twelve months. Then ask whoever administers your identity provider for a list of accounts that have not signed in for ninety days. Compare them.

In most growing organizations the two lists do not match, and the gap is where the real identity problem lives. Not in the sophistication of the authentication, but in the plain fact that access accumulates and nothing removes it, because growing fast means every process is optimised for granting.

Identity management is the one area where the work genuinely changes shape as headcount rises. What is adequate at ten people is negligent at a hundred, and the transitions are predictable enough to plan for.

First, one source of truth for who works here

Every identity control depends on being able to answer a simple question definitively: who currently works for this organization, in what role, since when. Until one system is authoritative for that, usually the HR or payroll system, nothing downstream can be verified. You will be comparing one incomplete list against another.

Practical version: HR owns the list of people. The identity provider owns the list of accounts. Somebody compares them on a schedule, monthly at first. When those two lists match without manual reconciliation, you have the foundation. When they do not, every other control inherits the error.

This is also the change that most often requires a conversation rather than a configuration, because it makes HR part of the access process rather than an adjacent department. That conversation is worth having early, while it is small.

Stop granting access to people, start granting it to roles

The default in a small team is per-person permissions. Someone needs the finance folder, so you give them the finance folder. It works until roughly the point where you cannot remember who has what, which arrives sooner than expected.

Replace it with groups that correspond to jobs. A new salesperson gets the sales group and inherits exactly what salespeople get. When they move to account management, they lose one group and gain another. The gain here is not tidiness. It is that an access review becomes answerable: instead of asking whether this particular person should have this particular folder, you ask whether salespeople should have it, which is a question a business owner can actually answer.

Two rules keep this from decaying:

  • No permanent individual exceptions. If someone genuinely needs something outside their role, grant it with an end date and a reason recorded. Exceptions without end dates become the new baseline.
  • Base the templates on what people do, not on what they ask for. Build the first set of role templates by asking each team lead what their people need, then grant slightly less and let them tell you what broke. Starting from too much and trimming never finishes.

Our articles on least privilege without the jargon and access reviews cover the detail.

Separate administration from everyday work

Anyone with administrative rights should have two accounts: an ordinary one for email, documents and browsing, and a separate administrative one used only for administration, with no mailbox attached. This costs nothing but a licence decision and removes the configuration attackers most want to find, which is an account that both reads email and can reconfigure the estate.

Alongside that, keep the administrative population small enough to name out loud. Count it quarterly and expect the number to fall over time rather than rise. Our piece on admin account security goes into the specifics.

Then set up at least one break-glass account: a highly privileged emergency account, excluded from the conditional access rules that could lock it out, with its credentials split between two people and stored offline, and alerting on any use. Without one, a misconfiguration or an outage at your identity provider can shut you out of your own systems at precisely the moment you need in. Our article on break-glass access covers how to set one up safely.

The population everyone forgets: service accounts and tokens

Human accounts get attention because humans complain. The non-human ones do not complain, which is exactly why they accumulate.

Make a list of every credential that is not a person: integration accounts between systems, API keys, personal access tokens in code repositories, accounts created by a contractor during a project three years ago, scanners and monitoring tools with broad read access, and automation accounts in your finance system.

For each one, record four things: what it is for, which system uses it, who owns it, and when the credential was last rotated. Most organizations doing this for the first time find several they cannot explain, and the ones nobody can explain are the ones to investigate first.

Then apply three rules. Service accounts get the minimum privilege the integration actually needs, not administrator because that was quicker. Each one has a named human owner who is asked to confirm it is still needed at review time. And anything that cannot support multi-factor authentication is compensated for with network restrictions, source address limits or a conditional access policy, because a password-only credential with broad rights is the cheapest way into an estate.

Connected applications and the consent problem

As a team grows, people connect third-party applications to the company's identity provider themselves: a scheduling tool that reads calendars, an email plugin that reads mailboxes, a design tool that syncs files. Each consent grant is a standing permission that persists after the person stops using the app, and it survives a password change.

Go to your identity provider's list of connected or third-party applications and read it. It will contain things nobody remembers approving. Then decide whether users can grant consent themselves, and if so, to what scopes. Requiring administrator approval for anything that reads mail or files is a reasonable line, and it is a configuration change rather than a project. Our piece on third-party app access walks through it for Workspace, and the equivalent exists in Microsoft 365.

Authentication: necessary, and no longer the whole story

Multi-factor authentication everywhere is the baseline, with no exemptions for executives or legacy systems. Beyond that, two things matter more than they used to.

Method strength. Push notifications and one-time codes can both be defeated by an attacker who proxies a real sign-in page or simply fatigues the user into approving. Phishing-resistant methods based on hardware keys or platform passkeys remove that whole category. Roll them out to administrators and finance first, where the consequence is highest. Our article on MFA fatigue attacks explains the technique being defended against.

Session lifetime and conditions. An attacker who steals a session token does not need your password or your second factor. Shorter session lifetimes for sensitive applications, re-authentication for administrative actions, and policies that treat an unmanaged device differently from a company laptop all limit what a stolen session is worth. Set these to match the sensitivity of the application rather than applying one global value, which will either be too loose or will generate a stream of complaints.

What to do at each stage of growth

Up to about fifteen people. Multi-factor authentication on everything. A password manager with shared vaults so credentials are not in a spreadsheet. Separate admin accounts. A written offboarding checklist, even if it lives in a shared document. One break-glass account.

Fifteen to fifty. Groups instead of individual grants. HR as the trigger for joining and leaving rather than a manager's email. Single sign-on across your main applications, which our piece on single sign-on for smaller organizations covers. A quarterly access review on the systems holding money and customer data. The service account inventory.

Fifty to a couple of hundred. Role templates maintained by team leads rather than IT. Automated provisioning and deprovisioning from the HR system. Contractor accounts with mandatory expiry dates. Privileged access that is requested and time-limited rather than permanently held. Conditional access tied to device compliance. Logging of administrative actions somewhere the administrators cannot quietly edit.

The sequence matters more than the pace. Automating provisioning before you have role groups just automates the mess.

Leavers, properly

The single highest-value identity process is the one that runs when someone leaves, and it is the one most often done badly because it happens on a day when everyone is busy with something else.

What good looks like: accounts disabled rather than deleted, at the agreed time, triggered by HR. Sessions and refresh tokens revoked, not just the password reset, because a live session survives a password change. Mobile device access removed. Mailbox access delegated to a manager and auto-reply set. Any shared credential the person knew rotated the same week. Physical access and building passes handled in the same pass. And a record of when each step was done.

Disabling rather than deleting matters: it preserves the audit trail, the mailbox and the file ownership, which you may need later. Our article on former staff access covers what goes wrong when this is rushed, and joiners, movers and leavers has the full checklist.

Someone forwarded a suspicious sign-in alert? Paste the message into CyberWatch AI Scan for a free check before anyone clicks the link in it.

Where this fits

Identity is the control plane for a modern organization. It decides what every other control is protecting, and it is where most serious incidents begin, which is why it sits near the front of the sequence in our guide to building a security program from nothing. Mergers and rapid hiring put the most strain on it, as covered in our piece on security during mergers and rapid growth, and the people-facing half of the problem is in our security awareness training guide. For the broader baseline, see our small business cybersecurity guide.

Frequently asked questions

What is the single most important identity control?

A single source of truth for who works here. Until one system is authoritative, usually HR, every other control is guessing, because nobody can produce a definitive list of current staff to compare accounts against.

When should we introduce single sign-on?

As soon as you run more than a handful of business applications, or earlier if any of them hold customer data. The security benefit is less about convenience and more about having one place to disable a person and one place to see what they can reach.

Why are service accounts such a problem?

They outlive the projects that created them, rarely have owners, often hold high privileges, and usually cannot take multi-factor authentication. They are a standing population of credentials that nobody reviews, which makes them the most attractive thing in an estate.

What should happen on someone's last day?

Their accounts are disabled within minutes of the agreed time, their sessions and tokens revoked rather than just the password changed, their mailbox delegated to a manager, and any shared credential they knew rotated. Disabling rather than deleting preserves evidence.

Is multi-factor authentication enough?

It is necessary and no longer sufficient on its own. Attacks increasingly target the session that exists after successful authentication, through token theft or a proxy page that captures a live sign-in, so phishing-resistant methods and shorter session lifetimes matter alongside it.

How do we handle contractors and guests?

Same lifecycle as staff, plus an expiry date set when the account is created. Time-limited by default is the only approach that survives contact with a busy organization, because nobody remembers to close an account for a project that quietly ended.

How often should access be reviewed?

Quarterly for systems holding money or customer data, annually for everything else, and immediately whenever someone changes role. Reviews that cover everything at once get skipped, so scope them deliberately.

What is a break-glass account and do we need one?

An emergency administrator account, held outside normal access, used only when the usual route fails. Yes, you need at least one, with its credentials split and stored offline, and its use alerted on, otherwise a problem with your identity provider locks you out of your own estate.

Should everyone's admin account be separate from their normal account?

Yes, for anyone with administrative rights. Browsing the web and reading email from an account that can reconfigure your estate is the configuration attackers most want to find, and separating the two is free.

Sources

  1. SP 800-63-4, Digital Identity Guidelines, US National Institute of Standards and Technology
  2. Introduction to identity and access management, UK National Cyber Security Centre
  3. 10 Steps to Cyber Security: Identity and access management, UK National Cyber Security Centre
  4. Multi-factor authentication for online services, UK National Cyber Security Centre
  5. SP 800-207, Zero Trust Architecture, US National Institute of Standards and Technology
  6. SP 1800-15, Securing Small-Business and Home Internet of Things Devices, US National Institute of Standards and Technology
  7. CIS Critical Security Controls List, Center for Internet Security
  8. Cross-Sector Cybersecurity Performance Goals, US Cybersecurity and Infrastructure Security Agency
  9. Zero Trust Maturity Model, US Cybersecurity and Infrastructure Security Agency
Share this article
LinkedIn WhatsApp X
For organizations

Your people get these messages at work too.

CyberWatch AI sends your staff realistic practice attacks, trains the gaps it finds, and shows management exactly where the organization stands.