The labor cost of accreditation is regressive: it falls hardest on the institutions least equipped to absorb it.

A flagship research university in a reaffirmation year — with an AACSB business school running its own AoL cycle underneath it — has a dedicated accreditation office, a professional institutional research lab with five or six analysts, an assessment director with staff, an associate provost whose job description is built around this work, and budget flexibility to fund course releases.

A regional comprehensive with 8,000 students has perhaps a third of that infrastructure. A small private liberal arts college with 1,200 students has, in many cases, a single overworked associate dean who also chairs a department. A community college with 6,000 students has a chief academic officer absorbing the work between two other portfolios.

In other words: the flagship has an actual org chart for accreditation.

Everyone else runs it on a shadow org chart that the Part 2 of the series had mapped: the unofficial second structure of faculty and staff doing this work on top of the jobs they were hired for.

All four of these institutions face broadly the same documentation expectations from their regional accreditor. They face the same federal compliance regime. They face the same programmatic accreditors when they offer programs in nursing, education, business, or engineering.

The standards do not scale to institutional capacity. The labor required to meet them does.

What the Original Vanderbilt Study Already Knew

The same 2015 BCG/Vanderbilt study that produced the famous $27 billion figure included a finding that has been quoted far less often than the headline number. Vanderbilt's own CFO, Brett Sweet, called it one of the study's most significant findings: small and medium-sized colleges were disproportionately impacted by federal regulations, with compliance eating up a larger share of their expenditures than at wealthier peers.

The study also estimated about $6 billion in annual compliance costs for community colleges as a sector, though with only one community college in the 13-institution sample, that figure rests on thin ground. Its direction matches everything else we know.

The FAFSA Verification Case Study

If you want to see how the regressive math plays out at the institutional level, look at FAFSA verification — a federal requirement that sits outside accreditation but shows the same pattern more cleanly.

A 2020 working paper by Guzman-Alvarez and Page estimated that institutional compliance costs of FAFSA verification totaled nearly $500 million annually, with the burden falling disproportionately on public institutions and community colleges. The costs by sector are striking:

• Community colleges spend approximately 22 percent of their financial aid office operating budget on verification.

• Four-year publics spend 15 percent.

• Four-year privates spend 1 percent.

The authors put it bluntly: the mandate places a disproportionate burden on institutions serving the neediest students. Verification load follows low-income enrollment, not institutional size: the students most likely to be selected are those with the greatest financial need.

Verification is also exactly the kind of high-volume, rule-bound document work that the current AI conversation points to first, and it's a fair example. Matching fields, flagging discrepancies, and assembling files is volume labor, the layer where automation helps.

But notice the asymmetry the cost data reveals: the institutions carrying 22 percent of an office budget are also the ones with the least capacity to deploy and supervise that technology well. The tool that would help them most is the one they are least able to adopt safely.

Small Private Colleges: Same Standards, No Staff

Community colleges absorb the load through volume — high enrollment, lean staffing, and chronic underfunding. Small private institutions face a different version of the same problem.

A small private with 1,200 students faces broadly the same documentation expectations as a regional public with 18,000 students. But with no dedicated accreditation staff, an IR office that may be one part-time analyst, and faculty already on three or four committees each.

One study points the other way, and it deserves a hearing. A 2020 University of Minnesota dissertation, The Cost of Accreditation for Small, Private Institutions of Higher Education, found that accreditation cost two small private colleges about 0.13 percent of their operating budgets per year. In budget terms, that is modest.

But consider what it means in practice. At a college with, say, a $40 million operating budget, 0.13 percent is about $52,000 a year — roughly one modest salary. Except no one is hired with it. The work lands on the associate dean, the part-time analyst, and the faculty already on four committees, on top of the jobs they were hired to do. A budget share tells you whether an institution can afford accreditation. It doesn’t tell you who carries it. At a 1,200-student college, that gap is the whole story.

Hartwick College’s 2012 self-audit referenced in the Vanderbilt study, estimated annual compliance costs around $300,000 but estimated the real figure could reach 7 percent of its non-compensation operating budget once decentralized costs including faculty time were counted. A 7 percent budget line that is mostly invisible because it is distributed across more than 100 roles instead of concentrated in one office.

This is where the labor cost gets genuinely structural, and where the AI argument runs into the regressive math head-on.

Small institutions cannot solve the problem by hiring an accreditation director; the marginal cost of that hire is too large relative to the operating budget. They cannot solve it by automating documentation, because the per-institution cost of a compliance platform is roughly the same as at a flagship.

And the temptation that follows is the dangerous one: with no dedicated staff, the under-resourced institution is the one most likely to let automation creep past the volume work into the judgment work — to let the tool decide what the evidence means, not just collect it.

As the previous post argued, much of what looks like document labor is compressed judgment. The institutions with the least capacity to keep a human on those calls are precisely the ones the regressive math pushes hardest toward giving them up.

The only remaining coping strategy is to absorb the load into the existing workforce — the same pattern that produces the shadow org chart.

The Programmatic Layer Makes It Worse

For institutions running professional programs, programmatic accreditation sits on top of regional accreditation, not in place of it. A nursing program holds CCNE or ACEN. A business school adds AACSB. An education program adds CAEP. Engineering adds ABET. Counseling adds CACREP. Each has its own self-study cycle, evidence requirements, site visits, and annual reporting.

The Vanderbilt study estimated programmatic accreditation alone at roughly $3 billion annually. For institutions running multiple professional programs, the labor load is not additive but compounds further. Because the same faculty often hold leadership roles across several accreditation processes at once, and the same IR analyst is pulling overlapping but non-identical data cuts for each.

A regional comprehensive with nursing, education, and business programs is effectively running four parallel accreditation cycles on staggered timelines. There is rarely a year in which none is in active review.

What This Argues For

The regressive distribution of accreditation labor is not a bug in any single accreditor’s standards. It is a structural feature of a system designed around uniform documentation expectations applied across radically different institutional capacities.

There are practical responses worth taking seriously: shared-evidence frameworks across regional and programmatic accreditors, joint site visits where programs hold multiple accreditations, capacity-aware reporting cycles for institutions below certain enrollment thresholds, and consortium-based institutional research support.

AI belongs on that list, but only if it is scoped honestly.

The promise it can plausibly keep is at the volume layer — aggregating evidence across overlapping programmatic and regional requirements, generating the repeated data cuts a single IR analyst now produces by hand, flagging gaps before a site visit. For an institution running four parallel cycles with one analyst, that is not a marginal convenience; it is the difference between a sustainable process and an unsustainable one. The promise it cannot keep, is the judgment layer: deciding what the evidence means, which fights are worth having, how to frame a weakness for a specific reviewer.

The regressive math makes both halves of that sentence matter more here, not less. The institutions with the least slack stand to gain the most from genuine volume relief, and stand to lose the most if automation quietly absorbs the judgment work too.

Separating the paperwork from the judgment calls is hard anywhere, as Part 2 argued. At a small institution, where the same person pulls the numbers and decides what they mean, it is hardest of all.

Used for the paperwork, with people keeping the judgment calls, AI can help level the field. Used carelessly, it widens the very gap it claims to close.

None of these responses are new ideas. All of them require accreditors and institutions to acknowledge openly that the current distribution of labor is not equitable and is not sustainable for the institutions doing the most expensive version of the work.

The honest question for the next decade of higher education policy is not whether accreditation is worth its cost in aggregate. It is whether we are willing to redesign the system so that the institutions serving the most vulnerable students are not also the ones paying the highest labor cost to remain accredited.

That is the conversation the final post in this series turns to.

We’re excited to announce the second round of our roundtable conversation series.

After receiving insightful suggestions from our previous series, feedback and points to ponder, we want to continue building this platform for conversation. The last theme centered on artifact scoring, which is one of the most burdensome parts of evidence generation for direct evaluation in accreditation processes.

It is also the most crucial component when considering AI-related solutions. The stakes are high, and accuracy drives everything that follows. Curricular recommendations, rescoring, and closing the loop all require accurate calibration of the rubric to begin with.

From what we heard, even selecting an artifact for a given program learning outcome can be challenging when rubric definitions are too vague or subjective to reliably identify the right artifact. This can lead to an even bigger problem: scoring the wrong artifacts.

We want to help reduce the time and burden on faculty. The solution we’ve narrowed in on is a comprehensive rubric design that consistently supports:

• Consistent artifact scoring by faculty or AI

• Rubric definitions that lead to better artifact selection

Starting next month, we’ll be running the same structure: exploring the use case of rubric design in accreditation, hosting a practical workshop on designing an agent to build rubrics, and discussing governance concerns when applying agentic AI to rubric design.

Module 1: The applied use case of rubric design in accreditation, and how agentic AI can help
Module 2: Workshop, designing an agent to build rubrics
Module 3 : Governance and guardrails for agentic AI in rubric design

Registration will be opening soon. Please stay tuned for the details! As before, we’re keeping the group small to keep it a conversation rather than a webinar.

If you were in the room for round one, thank you. This next series exists because of what you raised. If this is your first time joining, welcome in.

Subscribe now

On August 26, we hosted the third module of our Roundtable Series, which closed the first cycle of our roundtable conversation series.

This session focused on governance and the reality of responsible agentic AI adoption in higher education: specifically, that responsible adoption requires a clear understanding of its true operational costs and a realistic blueprint for how institutions sequence implementation.

Why governance for Agentic AI is a design problem

Before exploring responsible adoption, we defined governance itself: deciding in advance what an AI system is allowed to do, who is accountable when it acts, and how you know if something goes wrong. Its core pillars are authority, accountability, and visibility.

Most AI governance in higher education today addresses teaching and learning, specifically policies on how faculty and students use AI in coursework. Far less attention goes to administrative operations, where the institution executes its own processes. This is precisely where agentic systems create exposure.

The distinction matters because the nature of risk changes:

• Traditional AI Governance evaluates output risk: Is the output correct?

• Agentic AI Governance evaluates action risk: What is the system authorized to do, and who owns the consequences when it acts?

This shift makes governance an architectural design decision rather than a post-hoc compliance audit. An end-of-process review only reveals what already happened; it cannot show how augmented decisions were reached.

In practice, checkpoints must sit within each step of the workflow. Using artifact scoring for Assurance of Learning (AoL) as an example, responsible design requires explicit approval points and clear ownership of AI-generated evaluation outputs:

• Human Confirmation: An AI-generated score never enters the official record until a faculty member confirms or adjusts it.

• Scoped Access: Access is restricted strictly to the artifact set selected for the active review cycle.

• Traceable Logs: Every action is logged with user attribution, justification, and timestamping.

• Dynamic Scope: System permissions are reassessed every cycle.

Verification is "New Work" in Agentic AI

An AoL committee participant summarized current practice concisely:

You have a committee, you have artifacts, you have a rubric, someone assesses them, and you’re done.

Standard processes today rarely confirm scoring consistency across assignments or courses.

This creates a critical realization for anyone evaluating AI in assessment: an AI review checkpoint does not replace an existing task—it introduces entirely new work.

In the words of one participant, adding that checkpoint brings in “100% of new involvement.” Another participant highlighted the mirror risk: institutions expect AI to save time, only to discover faculty end up with more checks to perform than before.

That has a direct consequence for AI-assisted scoring, where the AI evaluates each criterion, provides its rationale, and routes only contestable points to a human. Efficiency claims have to be benchmarked against a baseline of zero pre-existing verification, because early implementations will rely on faculty and rubric owners verifying scores regardless of how the guardrails are designed.

The true value of our upcoming pilots will be measuring how calibrated AI scores are against human graders, and whether this model ultimately yields actual time savings.

Sequencing a pilot to build evidence and secure leadership endorsement

Participants offered strategic guidance for securing endorsement from accreditation committees and senior leadership, which is worth borrowing if you are considering a similar solution in your own work:

• Run in Parallel: Begin with full human scoring, generate AI scores alongside for direct comparison, and start with a single faculty member on the AoL committee before expanding to a second class and then broadening access.

• Convert Trust to Data: Parallel execution shifts the conversation with skeptical faculty and leadership from an emotional debate over trust to an empirical evaluation of measurement validity.

• Align with Power Structures: The committee accumulates its own record of where AI and human judgment agree and diverge, then presents that record with the Dean behind it. At one participant’s school, the Dean sits on the AoL committee, allowing them to establish institutional trust from day one.

• Budget the Verification Burden: Weak sponsorship is a leading cause of digital transformation failure. The initial verification workload in Cycle 1 is an intentional design input, not a platform failure—it must be budgeted upfront rather than argued away.

Where AI earns trust first

Two low-friction applications emerged that require far less initial trust than automated scoring:

• Pattern Flagging & Anomaly Detection: The AoL cycle is used as much to revise rubrics as to generate scores. A participant shared that during one cycle, ratings repeatedly came back marked “out of scope,” which revealed the committee had been assessing the wrong course entirely. AI is well suited to flagging anomalies like that and capturing rubric modifications alongside their justification, which is exactly the material an accreditation report asks for.

• Supporting Non-Subject-Matter Evaluators: At one institution, committee members rather than course faculty perform the assessment, meaning the person evaluating a communication outcome is often not a communications expert. AI can provide contextual rubric guidance to support non-experts working outside their primary field.

• A Related Validity Caution: One participant noted their institution’s writing measurements predated generative AI. On subsequent cycles, nearly all students exceeded benchmarks because they used AI to write. When a direct measure no longer discriminates performance, it creates a validity issue that must be addressed before introducing new AI assessment tools.

Critical structural constraints:

• Peer Rating Data: Peer rating surveys from team assignments can be primary evidence for a certain competency, and should not be treated as edge cases.

• Rubric Separation: Program-level AoL rubrics are distinct from grading rubrics used by course faculty; conflating the two will lead to incorrect conclusions.

What to Prepare Before Adoption Review

Legal and privacy reviews dictate the implementation timeline. Participants described privacy hurdles that can halt projects outright, ranging from restrictions on using public faculty LinkedIn data for CV collection to mandatory multi-tiered approvals for any tool touching student data.

When evaluating an AI tool that handles student work, requesting these key deliverables upfront will save time:

• Vendor Security Assessment: Completed within your IT department’s required framework.

• Data Flow Diagram: Clear mapping of artifact data routing, including external model providers.

• Data Retention & Disposal Terms: Explicit policies detailing what happens to data and access when the pilot ends.

• Model Training Exclusions: Written commitments that institutional data will never be used to train foundational models.

• Scoped Access Control: Granular permissions restricting tool access strictly to selected cycle artifacts.

• Comprehensive Audit Trails: Detailed logs tracking action attribution by user, step, and model version.

Responsibility ultimately rests with the deploying institution, which holds process oversight and accreditation obligations. Vendor solutions must carry these requirements through their platforms and model providers. General AI frameworks are helpful, but because none are written specifically for higher education, we are working out these requirements directly with schools.

The unresolved question: peer review visibility

Can external peer reviewers detect excessive AI involvement lacking human judgment, and what does that look like in practice?

The room agreed on the general direction: explicit rigor with human checkpoints mapped out clearly.

However, a key operational question remains: How would a peer reviewer actually detect human involvement, and is naming the report author sufficient? While AACSB is currently developing guidance on this topic, the standards remain unpublished.

That open question is why we treat the evidence trail as a core feature rather than a byproduct of scoring. If the record of who confirmed a score — and why an adjustment was made — cannot be exported in a form a peer review team accepts, the governance was never really there.

Where we're headed

This session concluded our first roundtable cycle, which we will re-run in upcoming months for new cohorts. Based on these insights, we are making three key adjustments in our own work:

• Instrumenting Agreement/Divergence: Tracking AI vs. human score alignment directly to generate empirical pilot data.

• Decoupling Pattern Flagging: Scoping anomaly detection and change capture as standalone capabilities exportable into accreditation reports.

• Pre-empting Security Requirements: Completing full vendor security documentation prior to initial pilot discussions.

We are building this framework in the open alongside faculty, assessment directors, and accreditation leaders. If you are interested in testing these governance design principles against your institution’s Assurance of Learning workflows, we’d be interested to hear how you are approaching them.

Most institutions do not have an accreditation department. They have an accreditation officer, which is usually one role tucked under the provost or dean’s office, sometimes a fraction of an associate provost’s job, occasionally a dedicated director at larger institutions. On the official organization chart, that’s where accreditation actually lives.

The official organization chart is a fiction. What institutions actually have is a shadow organization chart; a network of roles that quietly activates when a reaffirmation cycle ramps up, performs the work of accreditation for two to three years, then dissolves back into “regular duties” until the next cycle. It never appears in a budget document, is rarely acknowledged in workload policies, and is almost never compensated proportionally to what it produces. Yet without it, no institution would maintain accreditation.

The previous post in this series argued that the decade-long debate about the $27 billion cost of accreditation has been pitched at the wrong altitude, that the more important question is not how much the work costs in aggregate, but who specifically, is absorbing it. This post is about who.

And like any org chart, the shadow one has tiers, so the cleanest way to answer is to explicitly map the chart no budget document ever does, top to bottom.

The named owners

The top tier of the shadow chart is the only one that overlaps with the official one. A small number of people on every campus have accreditation in their job description. Their names appear in the self-study; they get the email from the accreditor; they sit on the steering committee.

The accreditation liaison officer or associate provost owns the timeline, the institutional narrative, and the political risk. In peak years this is functionally a full-time role layered on top of an existing full-time role.

Deans and department chairs assemble program-level evidence such as assessment plans, curriculum maps, alignment to accreditor standards. A chair with 3 classes per semester load running a department of fifteen does this on top of, not instead of, normal work. A 2024 study in the Journal of Higher Education Management found that chairs already report administrative work as the dominant source of role strain; reaffirmation compounds an existing overload rather than creating a new one.

Faculty assessment coordinators often on stipends in the low four figures run rubric calibration, closing-the-loop documentation, and artifact collection, frequently with no course release. Annualized against actual hours, the stipend often falls below the institution’s own adjunct pay rate.

These are the people whose accreditation labor at least exists somewhere in the institutional record. Their hours are still undercompensated for the amount of work, but at least they are nominally counted.

The bottleneck infrastructure

One tier down sit the offices everyone else’s deadline runs through. On the official chart, they report somewhere else entirely; on the shadow chart, every arrow passes through them. They aren’t “in charge of” accreditation, but nothing about it works without them.

The institutional research office generates the data: completion and retention rates, time-to-degree, faculty credentials, financial ratios, learning outcomes.

The same dataset gets cut differently for the federal report, the regional accreditor, the state licensing authority, the internal dashboard, and any programmatic accreditors on their own timelines.

IR offices are commonly staffed at one or two analysts for the entire institution. During a self-study year, they become the single point of contention for half a dozen simultaneous deadlines.

The CFO’s office demonstrates financial sustainability; facilities demonstrates classroom adequacy; the library demonstrates information literacy outcomes. Registrar, financial aid, student affairs, IT, and HR staff field document requests on accelerated timelines, usually with no acknowledgment in their evaluations.

Accreditation work moves to the top of their queue without ever being formally added to it. The hours come from whatever else those offices would otherwise be doing.

In normal times, these offices have their own work and their own service portfolios. During reaffirmation, accreditation work moves to the top of their queue without ever being formally added to it. The hours come from somewhere. They come from whatever else those offices would otherwise be doing.

The invisible labor

The bottom tier of the chart is the one the field talks about least. It is also, like the bottom of most org charts — the widest. And on the shadow chart, it is drawn in invisible ink.

As per the American Association of University Professors’ Fall 2023 data, roughly 68 percent of faculty appointments in U.S. higher education are now contingent non-tenure-track, often part-time, typically paid by the credit hour with little or no compensation for their service work.

Accreditation requires those faculty to do service work anyway.

CAEP, ABET, AACSB, ACEN, CCNE, and CACREP all expect documentation that adjuncts are evaluated, oriented, and participating in continuous improvement. Someone has to collect it. In practice, either full-time faculty absorb the load on behalf of uncompensated adjuncts, or adjuncts are pulled into uncompensated service to make the documentation defensible.

This is not marginal.

At community colleges, where contingent faculty teach the majority of credit hours, accreditation labor has a built-in equity problem most institutions have not honestly grappled with: the faculty with the least job security and the least time are the ones whose participation is most documentation-dependent.

The system has been resting on their goodwill, and that goodwill has nowhere left to come from.

Why naming this matters

When institutions estimate the cost of accreditation, they almost always undercount this shadow chart, for two reasons.

The first is methodological. Most cost studies use accounting categories that map to budget lines such as fees, dedicated salaries, consultant invoices. The shadow chart shows up in none of them; its labor is absorbed into salaries counted as “regular duties.”

The second is cultural. Higher education has long treated service work as something that simply happens, performed by professionals whose job description is flexible enough to absorb it. That tradition was built when workloads and the regulatory environment were lighter. The State Authorization Network’s 2026 report found that 85 percent of institutions reported a significant or moderate increase in compliance workload, and that is one regulatory domain among many.

When the shadow chart is invisible, three things follow. The labor stays uncompensated. The people doing it burn out and leave, taking institutional memory with them, often right after a reaffirmation. And the next cycle gets harder, because the people who knew how the last one worked are gone.

What the AI conversation usually misses

Everything above describes labor that stays invisible because it’s disguised as paperwork.

The associate provost’s negotiation looks like a scheduling email. The IR analyst’s framing decision looks like a spreadsheet query. The coordinator’s norming call looks like data entry. That disguise is exactly why the current conversation about AI in accreditation is getting something subtly, consequentially wrong.

Most of it is about AI, not with AI.

Accrediting bodies and policy groups such as CHEA, UNESCO, the Middle States Commission’s 2025 policy, the ACCME’s January 2026 guidance for continuing medical education have been writing about how institutions should govern, audit, and ethically deploy AI in their own teaching and operations. That work matters. It is also a different conversation from the one this post is in.

The narrower conversation about using AI to do the labor of accreditation, not to be regulated by it, is smaller and nuanced.

Tawnya Means’ April 2026 essay, argues that AI could make continuous improvement actually executable: faculty design the rubrics and interpret the results, AI handles the volume work of evidence aggregation and pattern detection.

The National Accreditation Commission’s AIHub project, funded by the GitLab Foundation, OpenAI, and Ballmer Group, is the most concrete attempt we know of to actually build this.

We want to add one thing, and it follows directly from the shadow chart as metaphorically elucidated above.

The data layer and the judgment layer are not cleanly separable, because the shadow chart already fused them. The associate provost negotiating with a chair over what counts as adequate evidence is doing data work and judgment work in the same conversation. The IR analyst deciding how to cut a dataset is doing both in the same query. The coordinator running a norming session is doing both in the same room.

Much of what looks like “document labor” from the outside is compressed judgment labor — decisions about what the evidence means, who needs to see it framed which way, and which institutional fights are worth having.

This is why “AI handles the documents, faculty handles the judgment” is a useful first frame but an incomplete one. The harder design question is which specific parts of the document layer are safe to automate, which parts carry compressed judgment that automation will quietly erase, and how to tell the difference at the design stage rather than after the fact.

That distinction is what this series returns to later.

For now the point is structural: AI is most likely to help with the labor in the bottleneck infrastructure layer such as the IR cuts, evidence aggregation, rubric application at scale, and most likely to cause real damage in the parts that look like documentation but are actually institutional judgment.

Telling those apart is the work the field has not really started yet.

Session Details:

What We Are Exploring

In Module 2, we explored what agentic AI workflows could look like in practice within Assurance of Learning.

Module 3 takes the next step: what needs to be in place for a workflow like this to be adopted responsibly inside an institution?

What We’ll Cover

The conversation is built around three questions:

• Where does governance belong? We treat review checkpoints and approval gates as design decisions rather than a compliance review at the end. We want to test that with you.

• What does your institution actually require? Data security, IRB, FERPA, procurement. We want to hear what the real gates are and who holds them.

• What would responsible adoption look like on the ground? Including how you document AI’s role in a way a peer review team would accept.

You do not need to have any of this figured out. The schools we have spoken to are at very different stages, and that range is part of what makes the conversation useful.

Facilitators

• Daisuke (Dai) Nishimura, Managing Director, Macnica

• Dolma Rawat, Research Lead, Macnica

• Jerelyn Co, AI Engineer, Macnica

About the Roundtable Discussion

Our roundtables are practitioner-led discussions built to explore emerging questions around AI and accreditation. They’re intended as a space for shared learning, thoughtful discussion, and exchanging perspectives across institutions.

Interested in Participating?

If this topic resonates with you and you’d like to take part in this roundtable or future ones, we’d love to hear from you.

Register using the button below. If you have any questions, reply to this post or get in touch—we’re always looking to bring new perspectives into the discussion.

Register for the Session

On July 28, 2026, we ran the second module of our Roundtable Series: a small, practitioner-led conversation about how to build agentic AI workflows that hold up inside accreditation work, and specifically inside Assurance of Learning (AoL).

The cohort was deliberately small, but brought together faculty and assessment leaders from several institutions, giving us the range of perspectives we were hoping for.

An hour of conversation produced more usable insight than a week of desk research. We are writing it down here, not only for us, but for you.

Objective of the roundtable

Our core objective hasn’t changed since Module 1: we are learning a lot through our internal research, but nothing compares to hearing directly from the people living inside the process day to day.

What did change was what we brought to the table.

Since Module 1, we have been actively building alongside listening. This session was less about asking for pain points and more about putting our working assumptions in front of leaders who could stress test them against real-world experience.

We set out to get two things:

1. An honest reaction to the solution we are shaping.

2. A clear sense of whether the underlying framework is something institutions could apply to their own work—with or without us.

We built dedicated check-in points into every section of the agenda for exactly this reason. As we noted at the top of the session: the parts where participants push back are the parts where we learn the most.

Keeping the human in the loop through intentional design

Most current excitement around agentic AI focuses on autonomous agents—where the model directs its own steps, decides what to do next, and loops continuously until it considers the task complete.

That capability is genuinely impressive.

It is also flexible, open-ended, and unpredictable because the autonomy sits with the model.

What we are building instead is an agentic workflow, where human practitioners design the steps and the model executes them. It is fixed, consistent, and predictable—putting the human in full control.

Our reason for this distinction is simple: Assurance of learning is already a designed process.

It has deliberate steps, deliberate checkpoints, and human judgment placed at specific points—because at the end of the day, a peer review team will ask you to defend each decision.

We start with a fixed workflow not because AI models lack capability, but because human control is built into AoL as a feature. Our job is to amplify that control, not route around it.

This sharpens our primary question: It is not whether AI can do accreditation, but which specific parts of the loop it can genuinely assist with.

Large language models excel at digesting large volumes of text. Reading a hundred student artifacts against a rubric is a textual understanding problem; surfacing patterns across those scores is a textual understanding problem.

When we surveyed participants before the session on where they saw the best fit, their answers landed on those exact areas:

• Curriculum mapping

• Artifact evaluation

• Pattern and gap analysis

The Caveat: The rest of the process still needs somewhere to run, and it requires robust governance to remain responsible and compliant. AI provides neither out of the box. That reality shapes the three pillars we build around: a unified platform, task automation, and governance.

A first look at the platform, with artifact scoring as the use case

We walked the room through a static mockup of the platform, built around the learning goal as the unit of monitoring, with three views: university, program, and learning goal. Each learning goal moves through setup, baseline, gap analysis and intervention, and remeasure, and then closes the loop.

We then demonstrated the scoring prototype live:

• The Scorer: Reads the rubric and student artifact, scores each criterion, and provides its reasoning alongside citations of the specific passages relied upon.

• The Critic: A second model audits those scores, evaluating if any judgment is contestable between adjacent rubric levels. If so, it flags the criterion for human review.

• Human-in-the-Loop: Criteria that AI shouldn’t judge are routed to a human before scoring begins. Reviewers can override any score, and both the override and its rationale are saved for the final report.

The Feedback That Stood Out

The questions raised during the demonstration underscored why we host these roundtables in the first place:

1. Reporting Integration: Participants asked whether indirect measures could sit alongside direct ones, and if results could auto-populate the updated Table 5-1 spreadsheet under the 2026 AACSB standards. Both are on our roadmap.

2. Timing & Process Interference: The hardest question focused on timing: If a program is mid-cycle with artifacts already under evaluation, how do you introduce AI without clouding the results? As one participant noted, introducing AI is itself an intervention that must be documented. Our current approach is to adapt to your existing timeline — taking your current curriculum map as a fixed input.

3. Rubric Clarity: Someone asked whether the AI defines what exceeds, meets, does not meet, or absent mean, or if humans do. The answer is that the rubric must define them. One participant shared that while their rubric lists performance levels, it has never explicitly defined them. If exceeds isn’t defined, neither five faculty members nor an AI model can apply it consistently. Getting “AI-ready” requires foundational rubric clarity first.

Governance as a condition of adoption, not an afterthought

Governance came up on its own well before we got to it on the agenda, which tells you something about where it sits for practitioners.

The question was about data security, and it came in operational terms: institutions have internal processes that any external agent has to clear before it goes anywhere near student work.

We treat governance as a foundational pillar rather than a final compliance audit because it belongs in the design phase. Agentic solution design cannot treat governance in isolation. Review checkpoints, approval gates, and audit trails are architectural decisions that must be considered alongside adoption.

Because standards and institutional requirements vary across campuses, governance frameworks must be co-designed directly with the schools using them.

Where we are headed

Module 3, planned soon, goes deep on exactly this: the governance and guardrails that have to be in place for a workflow like this to actually be adopted inside an institution.

If you want to test these concepts against a real workflow at your institution, we welcome exploratory conversations. We are building the solutions in the open, with faculty, assessment directors, and accreditation leads, and these sessions are where the building actually happens.

If you’ve sat through more than a handful of conversations about accreditation reform, you’ve heard the number.

$27 billion a year in federal regulatory compliance costs across U.S. higher education, with roughly $3 billion of that attributed to regional accreditation alone.

The figure comes from a 2015 Boston Consulting Group study commissioned by Vanderbilt University, and in the decade since it has done remarkable work. It’s been quoted in congressional testimony, pulled into op-eds, leaned on by every think tank with a position on higher ed regulation, and treated, more or less, as settled.

It is not settled.

And the fact that it has been treated as settled for ten years tells you something important about how higher education talks about its own labor.

The figure, and the pushback

The original BCG/Vanderbilt study sampled 13 institutions and extrapolated to the sector. It found that regulatory compliance represented 3 to 11 percent of core university operating expenses at the studied institutions, and that faculty and staff spent between 4 and 15 percent of their time on compliance activities.

From there, the study estimated roughly $17 billion in non-research higher-education compliance, $10 billion in research-related compliance, and inside that total, $3 billion for regional accreditation and another $3 billion for programmatic.

The pushback came almost immediately.

Belle S. Wheelan, then-president of the Southern Association of Colleges and Schools Commission on Colleges, and Mark A. Elgart of AdvancED published a sharp essay in Inside Higher Ed arguing that the headline figures were “significantly inflated as well as irresponsibly misleading.”

Their case had two parts.

First, much of what the study counted as accreditation cost was actually federal research-grant compliance — a separate regulatory regime that happens to overlap with accreditation timelines.

Second, they argued that genuine accreditation activity is “part of regular faculty service and committee work” that contributes to institutional improvement regardless of whether an accreditor exists.

Wheelan and Elgart pointed to a different number: across three accreditation regions, the average combined direct and indirect cost of accreditation at doctorate-granting research universities came out to roughly $415,000 per institution over the full review cycle.

That is not nothing.

It is also nowhere near the per-institution figure the $3 billion sector estimate implies.

So which number is right?

They’re both right, and that is the problem.

Here's the uncomfortable truth: both estimates are defensible, because they're answering different questions.

The Vanderbilt study counted everything an institution spends on activities that touch federal compliance, including faculty and staff time valued at full loaded cost. That methodology produces a big number because faculty and staff time is a big number when you actually count it.

Wheelan and Elgart’s framing counted only the activities that exist solely because an accreditor requires them — the marginal cost of accreditation, separate from work that institutions would do anyway.

You can argue all day about which framing is more honest. What you cannot argue is that either figure settles the question for the people doing the work. Because here’s what neither one tells you: how the cost is distributed inside the institution.

A $415,000 cycle cost at a research university sounds manageable in budget terms. Spread across a five-to-ten-year reaffirmation cycle, it is.

But that figure says nothing about whether the assistant provost coordinating the self-study is doing it on top of a full administrative load.

It says nothing about whether the institutional research office gets backfill for the analyst pulled into evidence prep for nine months.

It says nothing about whether the department chairs absorbing program-level documentation are getting course releases or just gratitude.

This is the gap that the entire $27 billion debate has been talking past.

The argument has been pitched, for a decade, as a referendum on whether accreditation is “too expensive” in aggregate.

The more interesting question — and the one that actually predicts whether a reaffirmation cycle will damage an institution — is whether the distribution of the cost is sustainable for the specific humans absorbing it.

That question has been almost completely absent from the policy conversation, and the people who could answer it most precisely are also the people too busy carrying the load to write about it.

What the more recent data adds

The 2015 Vanderbilt figures are now a decade old, and the regulatory environment has only thickened since.

The State Authorization Network’s 2026 report Between the Lines, Behind the Work offers a more current window into one slice of the load.

Eighty-five percent of institutions reported a moderate or significant increase in licensure compliance workload under newer federal certification requirements.

Nearly three-quarters added between one and ten staff hours per week.

Twelve percent added 21 or more hours weekly — the equivalent of a half-time hire’s worth of new work, often with no half-time hire to do it.

That is a single compliance domain at a single staffing level.

It does not include faculty time, IT coordination, cross-departmental meetings, or the cascading effect on other priorities those hours displace. Multiply across the full accreditation and compliance landscape, and the labor question gets sharper, not vaguer.

What the number was always for — and what the next decade will ask of it

The $27 billion figure has been useful precisely because it is big and quotable.

It moved the conversation.

It got accreditation onto agendas it would not otherwise have reached. That has value, even if the number itself is contested.

But a decade in, the figure has become a stand-in for a conversation we are not actually having.

The real question was never “is accreditation too expensive in aggregate.” It was always “who, specifically, is paying for it, and is the load distributed in a way the institution can sustain.”

That question matters more now than it did in 2015, for a reason that hasn’t fully landed in the policy literature yet.

The same decade that produced the $27 billion debate also produced a new generation of tools — large language models, agentic systems, document-aware AI — that are genuinely good at exactly the kind of work the shadow labor of accreditation consists of: reading, summarizing, mapping, evidence aggregation.

Whether and how those tools belong inside accreditation work is a real question, and one we’ll come back to later in this series.

But it cannot be answered honestly until the labor itself is honestly counted.

You cannot redesign work you cannot see.

Session Details:

What We Are Exploring

Module 1 surfaced where the AoL process feels most burdensome and where AI could genuinely help.

Module 2 moves from that discussion into practice: we’ll walk through what an agentic AoL workflow actually looks like, step by step.

What We’ll Cover

We'll walk through a practical example of an agentic workflow, using it as a starting point for discussion. Along the way, we'll examine how different components work together, where faculty expertise remains essential, and the questions institutions should consider when evaluating similar approaches.

Some of the themes include:

• Where AI adds value—and where it doesn’t

• Human oversight, trust, and workflow design

• Institutional considerations for adopting AI responsibly

• Open practitioner discussion

Facilitators

• Daisuke (Dai) Nishimura, Managing Director, Macnica

• Dolma Rawat, Research Lead, Macnica

• Jerelyn Co, AI Engineer, Macnica

About the Roundtable Discussion

Our roundtables are practitioner-led discussions built to explore emerging questions around AI and Accreditation. They’re intended as a space for shared learning, thoughtful discussion, and exchanging perspectives across institutions.

Interested in Participating?

If this topic resonates with you and you’d like to take part in this roundtable or future ones, we’d love to hear from you.

Register using the button below. If you have any questions, reply to this post or get in touch—we're always looking to bring new perspectives into the discussion.

Register for the Session

We recently concluded our webinar on June 10, 2026, where we held the first session the Roundtable Series: a small, practitioner-led conversation about how agentic AI can be leveraged in accreditation, particularly Assurance of Learning (AoL). Our cohort was extremely small, but spread across two institutions in the United States and Canada, which gave us a diverse perspective. An hour of conversation led us to various insights, which we noted down not only for us, but for you as well.

Purpose of the Roundtable

While we are discovering a lot through our own research, nothing compares to hearing it from the people living inside the process on a day to day basis.

We wanted to hear their voices, and make the session as conversational as possible.

Faculty Engagement as the #1 Hurdle

One of the questions we really wanted to understand was which part of the whole AoL process takes the most amount of time. As we have been exploring the role of AI in accreditation, this one was crucial for us to see whether there was a fit.

To our surprise, two practitioners without a flinch named faculty engagement as the biggest hurdle.

What's interesting to us is decoding why faculty engagement is the hurdle. Is it because the process is too tedious? Is it because of the additional work? Or because of the hidden responsibilities?

From our research, we discovered that the additional responsibilities are what weigh on faculty and keep them from engaging with the accreditation process in a meaningful way. The indirect consequence of this is that AI can be leveraged to ease some parts of the process, making AoL more efficient and smoother.

Understanding Agentic AI

What followed was the team clearly defining the capabilities of Agentic AI, and how it differs from the standard GenAI LLMs that are more commonly used.

Agentic AI works differently from GenAI. It goes a step beyond answering to a prompt, it can take a goal, plan out the steps, use tools like files, databases, or APIs, and keep working through the task in a loop until the goal is met.

One analogy from the session that stuck with us while thinking about Agentic AI: if generative AI is the engine, then agentic AI is the whole car, and it can run several of those engines at once.

Where it Fits in Accreditation, particularly in the AoL process

After establishing the definition and capabilities of Agentic AI, the team then explained all the parts of the process in AoL to explore of parts where Agentic AI can be a good fit and parts where it can’t.

The steps pointed out in the session were the following:

One observation quickly became clear: participants saw the greatest potential for agentic AI in the second stage of the process—rater norming. Which primarily indicates the rating unevenness across faculty while scoring artifacts.

While AI can never be a replacement for a faculty member grading the rubric, there is context to the grading process that an AI as of today can never understand, whether it's how the curriculum was delivered, or how the curriculum changed before a course was phased into the semester, which led to a difference in learning outcomes. All of this is still out of scope.

However, what AI is specifically good at is creating a baseline evaluation, which ensures there is consistency in grading.

If five faculty members apply a rubric to student work and each applies it slightly differently, then there is a calibration problem. An AI system doesn't drift between graders, and human spot-checks can confirm reliability without re-doing every score by hand.

Faculty Oversight is Non-Negotiable

One thing the room was very clear on is that faculty oversight is not optional. AACSB allows AI to be used across many steps of the process, as long as faculty review and approve the outcomes, and that requirement doesn't go away.

One participant also pointed out that a process which looks too automated can fail the "smell test" with a review team, even if the outputs are good.

What's also interesting is that oversight tends to erode over time. As people build trust in a system, the review gets lighter pass over pass. It's not a flaw in any one person, it's just a very human pattern, and it's something any workflow has to be designed against from the start.

Data Privacy as a Primary Hurdle for AI Adoption

Prior to our roundtable, we also asked the participants to fill out a survey, and one of the questions was about the biggest hurdle they foresee in adopting AI.

And rightly so. Handling student data is extremely sensitive. Sensitive data such as student names, IDs, and grades are all things that faculty may be hesitant to upload to a cloud-based service such as ChatGPT or Claude.

That's why AI shouldn't be adopted without being mindful of the risks. It also means adopting technologies that are specifically FERPA-compliant and aligned with institutional data privacy frameworks, while building workflows that keep all the right guardrails in mind for how the infrastructure handles that data.

But compliance is really just the floor.

The harder questions sit above it: Who holds the vendor contract, and what does it say about data use? Where is student data actually stored, and in which jurisdiction? Is there any possibility the model is being retrained on institutional inputs? Who inside the institution has access to query logs, and who audits that access?

These are data governance questions, and most institutions haven't had to ask them this rigorously until now. AI is forcing that conversation whether schools are ready for it or not.

Where We’re Headed

Through our conversation, it was clear that there's a real fit for AI in the workflow of AoL.

The key question now is where to focus: identifying which parts of the process AI can genuinely strengthen, and which are better handled without it. That's the calibration: understanding each institution's specific needs, how they handle data internally, and at what stage AI can add the most value.

In our next session on July 28, we'll move from theory to practice by demonstrating of how to design an agentic workflow within the AoL process. If you work in AoL and and this resonated with you, we'd love to have you join the conversation.

Registration details to follow.

There’s a particular kind of work in higher education that almost nobody outside it has heard of, and the people inside it tend to talk about in tired, slightly resigned tones (oftentimes vocal about the burdens of extensive paperwork).

It goes by different names depending on which accreditor your particular department/school answers to. Assurance of Learning if you're in a business school. Outcomes assessment if you're in engineering. Competency-based programmatic review in the health professions. The vocabulary changes; but the work doesn't. You define what students are supposed to learn, you collect evidence that they actually learned it, and every few years you prove to an accrediting body that you've documented the whole process meticulously, often rigorously and as accurately as possible.

On paper it’s straightforward. In practice, the task is much more complicated than it looks —and the work falls disproportionately on a small number of people in each program outside their formal work duties.

Why we started writing

The starting point was something narrower than a blog. We were running a proof-of-concept study with Claremont Graduate University’s AI for Humanity Lab, exploring whether AI could meaningfully support rubric-based competency evaluation—the kind of scoring work that sits at the center of most accreditation cycles.

At first, it was a research question: can an LLM model help here, and where does it fail?

We started Aolia because we kept finding ourselves in conversations with assessment directors, program coordinators and associate deans across different institutions, and noticing how similar their pain points were — and how rarely they had a space to compare notes.

So this is an attempt to change that: a community of practice for people inside accreditation, built by people who’ve spent time listening carefully to what the work actually looks like from the inside.

What we'll be writing about

Some of them will be about AI, because that question is unavoidable now and most of the writing on it is either too optimistic or too dismissive to be useful.

Most of the posts will be not be about AI directly.

It will be about the administrative burden problem. What reviewers actually look for. Why so much assessment data never makes it back into curriculum decisions.

It’s about the gap between what accreditation asks programs to demonstrate and the infrastructure they’re actually given to demonstrate it with.

If you’ve ever felt that gap personally, you’ll recognize it.

Beyond the writing

That listening turned into something more structured this year, starting with conversations across the AACSB community. We started the AoL Roundtable, a small, invite-only series where Assurance of Learning (AoL) directors, faculty, and deans get hands-on with Agentic AI, exploring how it can be applied within their own institutions workflows.

As the series runs, we’ll be publishing what comes out of it here, alongside what we’re learning from the research side.

Before you go

If you have a few seconds, we'd genuinely like to hear this: where in your assessment or accreditation cycle does the process feel least supported?

If something is slow, manual, duplicated, or quietly held together by workarounds—we’re especially interested in that.

Hit reply and tell us what you work on and what you wish someone would take seriously. Much of what we publish here will come from those conversations, and we read every response.

Glad you’re here. We’re looking forward to building this with you.