What problem does it solve?
Access reviews exist because entitlements drift: people change roles, contractors leave, temporary access becomes permanent because revoking it is nobody's job. SOX drives periodic certification as part of its IT general control audit expectations, and SOC 2, an attestation many SaaS and cloud vendors pursue rather than a legal requirement, expects the same discipline (editorial framing, not drawn from the evidence on this page). So a manager or application owner ends up having to look at every entitlement and decide.
This usually becomes a spreadsheet or ticketing exercise across dozens of applications, sent to reviewers who do not know what most of the access they are certifying actually does. Reviewers rubber stamp long lists to get through the campaign, which satisfies the audit requirement without reducing the access risk, and the process can stretch over weeks for a large organization.
How does it work?
- Collect entitlements continuously. The agent syncs identities and their access grants from every connected application, including ones without a native governance integration, instead of a manual export per system.
- Learn what normal access looks like. Using department, title, group and prior grants, the agent builds a profile of the access someone in a given role typically holds.
- Score every grant. Each entitlement gets a risk or trust score based on how well it fits the peer profile, prior review decisions and how sensitive the resource is.
- Pre approve or group the routine. Low risk, expected access is grouped or pre approved for a one click decision; only access that deviates from the profile needs individual judgement.
- Route exceptions with context. Each exception goes to the right business owner with the score, the reason it was flagged and the peer comparison, not a bare entitlement name.
- Record and revoke. Every decision, approve or revoke, is logged with who decided and why, and revocations are pushed back to the source application automatically where integrated.
Public evidence covers the steps unevenly. AmeriGas uses SailPoint's AI recommendations inside its certifications: the system shows what share of a department holds an entitlement, and a person decides whether the user is a real outlier (steps 2 and 3). SailPoint reports the same kind of use in its own campaigns. Netskope uses Lumos, whose LLM powered analytics flag anomalies and toxic combinations of access and route prioritized items to owners (steps 3 and 5).
Step 1 and the audit reporting part of step 6 are documented by the data automation deployments at Sprout Social and Hippo Insurance Services, whose sources describe no AI. No named organization's public source describes pre approval of low risk access (step 4) or automated revocation, so treat those as the part of the design not yet shown in production. All of this evidence is published by vendors.
The cycle time benchmark on this page pools two vendor claims: Netskope's up to 70% (AI) and Sprout Social's 75% (data automation, no AI described), so read it as the range reported for access review programs, not as the effect of AI scoring alone.
- Audience
- Employee facing
- Autonomy
- Supervised agent
- Adoption
- Emerging
- Channels
- Internal tools, API and system to system
What is it worth?
Benchmarks are computed from the public deployments below: one data point per organization per KPI, with who made each claim.
| KPI | Median | Reported range | Data points | Claimed by |
|---|---|---|---|---|
| Cycle time reduction | Too few to pool | 75% Not pooled: up to 70% | 1plus 1 up to | 1 vendor |
Value drivers: Risk and loss reduction, Employee productivity, Compliance quality, Speed and cycle time.
Indicative value
A company running quarterly access reviews for 3,000 employees across 50 applications
USD 48,000 to USD 480,000
Reviewer time released across annual access review cycles per year
How this is calculated
Formula: employees * grantsPerEmployee * cyclesPerYear * minutesPerGrant / 60 * automationShare * hourlyCost. The low scenario uses every low input, the high scenario every high input.
| Input | Low | High | Basis |
|---|---|---|---|
| Employees in scope employees, employees | 3,000 | 3,000 | The reference organization. |
| Access grants reviewed per employee per cycle grantsPerEmployee, access grants per employee | 8 | 15 | Editorial assumption across core systems and SaaS applications. Replace with your own entitlement count. |
| Access review cycles per year cyclesPerYear, cycles per year | 4 | 4 | Editorial assumption. Quarterly is a common review cadence; replace with your own. |
| Minutes a reviewer spends per grant without AI support minutesPerGrant, minutes per grant | 2 | 4 | Editorial assumption. Replace with your own campaign timing data. |
| Share of review effort the agent removes automationShare, fraction of review effort | 0.3 | 0.5 | Editorial assumption. Lumos reports that access reviews at Netskope are accelerated by up to 70%, a vendor claim and an upper bound (the same page also gives 77%), so a range of 0.3 to 0.5 stays well below it. No source measures the share of review effort an AI agent removes. The Sprout Social and Hippo Insurance Services figures describe data automation, not AI, so they are not used. Replace with your own reviewer time study. |
| Fully loaded cost of a reviewer or GRC analyst hourlyCost, USD per hour | 50 | 80 | Editorial assumption. Replace with your own rate. |
What it leaves out: Reviewer time only. It leaves out the platform cost, the integration work per connected application, and the audit and breach risk reduced by faster, more consistent reviews, which is harder to price but is often the larger benefit.
Who already uses it?
5 public deployments, strongest evidence first. Grades: A regulator or audit, B the organization itself, C vendor case study, D anonymous or estimate.
AmeriGas
United States · Energy and utilities · 2026
AmeriGas, the largest propane company in the United States, governs more than 7,000 identities and 600 plus roles with SailPoint. Certifications used to run on manual provisioning, spreadsheets and screenshots. Its Identity and Access Security Manager says the AI features inside SailPoint certifications now give reviewers recommendations: the system shows what share of a department holds an entitlement, so the reviewer can tell whether a user is an outlier who really needs the access. This matches the peer comparison step of an AI assisted access review. The case study is SailPoint's own and reports no figure for time saved on certifications, so no metric is recorded.
No outcome disclosed.
Netskope
United States · Technology and software · 2026
Netskope, a cloud security vendor, replaced a patchwork of identity tools with the Lumos Autonomous Identity Platform. Lumos describes LLM powered analytics that monitor risk, flag anomalies and toxic combinations of access, and route prioritized items to owners, with an AI agent that mines roles from assignment, usage and business context and keeps a human in the loop for high risk decisions. Lumos reports that access reviews at Netskope are accelerated by up to 70% and standing access is reduced by 80%. The figures come from the vendor's case study, not from Netskope.
- Cycle time reduction: up to 70%, per access review
"Automated role mining and policy management are driving security and efficiency where access reviews are accelerated by up to 70%, standing access reduced by 80%, and fewer managed roles reduces IT tickets and lowers costs."
Claimed by: vendor
SailPoint
United States · Technology and software · 2026
SailPoint runs its own identity governance product, SailPoint Identity Security Cloud, to manage its employees' access, a program it calls Customer Zero. Its case study says AI recommendations help managers see what they are approving and judge whether access is warranted during certification campaigns, and that automated revocation of risky entitlements doubled the effectiveness of those campaigns. The deployment is the vendor using its own product and the claims are the vendor's own, with no figure that maps to a taxonomy KPI, so no metric is recorded.
No outcome disclosed.
Sprout Social
United States · Technology and software · 2024
Sprout Social's GRC team ran access reviews through a ticketing based, manual process to meet SOX and SOC 2 requirements, which grew harder to scale as the range of applications in scope expanded. It adopted Zilla Security (now CyberArk Comply, part of CyberArk) to sync access data directly from cloud applications, including ones without an API, and automate the review campaign instead of manual data pulls. Zilla Security, now CyberArk Comply, part of CyberArk, reports a large reduction for Sprout Social in how long each review campaign takes, alongside better communication with reviewers and easier scaling to new systems. The gain is data sync and workflow automation; neither source describes AI or machine learning anywhere in the deployment.
- Cycle time reduction: 75%, per review campaign
"After quickly deploying Zilla Security, Sprout Social saw a 75% reduction in project duration of access reviews, improved communication with reviewers, and enhanced scalability to support a growing range of applications and systems."
Claimed by: vendor
Hippo Insurance Services
United States · Insurance · 2023
Hippo Insurance Services, a licensed property casualty insurance agent, ran user access reviews across dozens of SaaS and cloud applications holding sensitive financial information, ahead of quarterly compliance certifications by external auditors. The access reviews, taking up to two weeks per application, were a team effort across InfoSec, HR and legal, and preparing the audit materials was 100% manual: checking access rights, cross referencing job roles and validating changes by hand.
Its InfoSec team adopted Zilla Security (now part of CyberArk) to automate the data collection and mapping across those applications. Hippo's CISO reports the process per application went from about two weeks to around two days, and most certifications could now be completed by two InfoSec team members alone, down from that cross team effort. Zilla's case study narration says over 90% of the previously manual tasks became automated.
The gain is data collection and mapping automation; the source does not describe AI or machine learning.
- Cycle time: about 2 days
"The process that used to take two weeks per application was reduced to just around two days, and most of the certifications could now be completed by two members of the InfoSec team."
Claimed by: organization
How do you implement it?
A model agnostic playbook: what to prepare, the order to build in, and what goes wrong.
Data you need
- Identity and role or department data from HR, kept current
- Entitlement data from every application in scope, including ones without a native connector
- History of prior review decisions, to train the peer comparison
- A defined owner for every application and resource in scope
Systems to integrate
- Identity provider or directory (for identity and group membership)
- HR system, for role, department and employment status
- Target applications and their entitlement or role model, including ones synced without a native API
- Ticketing or GRC platform, for exception routing and audit evidence
Complexity: Medium
Reading entitlements from well integrated systems (identity provider, core HR, major SaaS) is fast. Coverage of applications without an API, and getting business owners to actually review rather than rubber stamp, take longer and depend on executive sponsorship.
- 1
Start with the systems that hurt the most
Pick the applications with the largest, messiest entitlement lists and the most reviewer complaints first; that is where automation saves the most time and builds the most trust.
- 2
Feed it clean identity data
The peer comparison is only as good as role and department data from HR. Fix known gaps before turning on scoring, or the agent will learn from a wrong baseline.
- 3
Calibrate on one cycle before trusting the scores
Run the first campaign with the agent's recommendations visible but not acted on automatically, and compare its flags against what reviewers actually decided.
- 4
Pre approve narrowly, expand carefully
Start pre approval with the lowest risk, best understood access only, and widen the scope as false positive and false negative rates on real campaigns prove out.
- 5
Close the loop on revocations
A review that flags excess access but does not remove it has not reduced risk. Automate the revocation step back to the source system wherever an integration allows it.
Guardrails
- Every revocation and every exception decision is made or confirmed by a named human owner
- Pre approval applies only to access below a defined risk threshold, reviewed and set by security
- Every score shows the peer comparison and the reason behind it, so a reviewer can check it in seconds
- Full audit log of every decision, the reviewer, and the score the agent presented
- Privileged and administrative access is always routed for individual review, never pre approved
KPIs to instrument
- Project duration per review cycle, before and after
- Share of grants pre approved versus individually reviewed
- Reviewer decisions that overturn the agent's recommendation, to catch drift or miscalibration
- Time from a flagged exception to a recorded decision
- Access actually revoked as a share of access flagged as excessive
Human in the loop
A named business or application owner decides every exception and every revocation of access above the pre approval threshold. Security or GRC teams set and periodically review the threshold, and internal audit samples decisions to check reviewers are not rubber stamping the agent's recommendations.
Common failure modes
- Automated rubber stamping
- Pre approval becomes the new rubber stamp if the threshold is set too wide. Keep the threshold narrow and sample decisions for quality.
- Peer groups built on bad role data
- A wrong or stale department or title field makes the agent's normal look abnormal, or the reverse. Fix identity data quality before trusting the scores.
- Coverage gaps outside the connected applications
- Access in systems without an integration never gets reviewed at all, but the campaign looks complete. Track and report coverage explicitly, including what is out of scope.
- Flags without revocation
- Exceptions get flagged and discussed but access is never actually removed. Instrument revocation completion as well as review completion.
What are the risks and rules?
EU AI Act
Depends on design
A tool that only certifies IT system access and revokes over provisioned entitlements is not itself a use listed in Annex III, and makes no decision about a natural person's employment, benefits or creditworthiness. But scoring how an employee's access compares to their peers sits close to Annex III's category on monitoring and evaluating the performance and behaviour of workers: if the score is used, or could be read, to judge conduct or performance rather than only system entitlements, treat it as high risk and run the full assessment. Keep the scoring scoped to entitlement risk, documented and reviewed, to stay on the lower risk side of that line.
Controls to put in place
- Inventory entry for the agent with an owner, its scoring logic and the applications in scope
- Documented pre approval threshold, reviewed and signed off by security or GRC
- Full audit log of every review decision, retained for the compliance period required
- Periodic sampling of pre approved decisions by internal audit to check for rubber stamping
Frequently asked questions
- Does an AI agent decide who keeps access?
No. The agent scores and groups entitlements so a named business or application owner spends time on the exceptions, and that owner confirms every revocation or exception decision. At AmeriGas, the AI recommendations in SailPoint show what share of a department holds an entitlement, and a person decides whether the user is a real outlier. The Lumos case study for Netskope also describes keeping humans in the loop for high risk scenarios.
- How much time does this actually save on access reviews?
Every figure found is a vendor claim, not an independent measurement. Lumos reports that access reviews at Netskope are accelerated by up to 70%, and gives 77% for its LLM powered analytics elsewhere on the same page. Zilla Security, now part of CyberArk, reports a 75% cut in the project duration of access reviews at Sprout Social, and Hippo Insurance Services' CISO says review time per application fell from about two weeks to around two days, but both of those cases are data collection and workflow automation with no AI described. SailPoint says AI recommendations and automated revocation doubled the effectiveness of its own certification campaigns, with no baseline.
- Is this only relevant to regulated industries?
Regulation is often what sets the review cadence: SOX and SOC 2 are common drivers, though SOC 2 is an attestation many SaaS vendors pursue rather than a legal requirement. Any organization with SaaS sprawl and entitlement drift has the same underlying problem of reviewers certifying access they do not understand.
How to cite this page
Blits.ai AI Use Case Library, "AI agent for identity and access review and certification", last verified 13 October 2026, https://www.blits.ai/ai-use-cases/identity-and-access-review-agent. Licensed under CC BY 4.0. Method: how we verify use cases.
Changelog
- 13 October 2026: First published