{"resourceId":"marshall-ibl-sovereign-ai-sandbox-educause-20261008","versions":[{"version":"external-ea0fbf5d86b956cd4182dc2df69d8806199cc8b4b979c562ecec6da8827fbda6","resource":{"id":"marshall-ibl-sovereign-ai-sandbox-educause-20261008","title":"ibl.ai case study: Marshall University takes full source code of ibl.ai agent platform as a 'sovereign' sandbox to measure usage and ROI, announced at EDUCAUSE 2026","organization":"Marshall University / ibl.ai (vendor case study)","sector":"Public university / agentic AI platform for administrative and student services","geography":"United States (West Virginia)","publishedAt":"October 8, 2026","publicationDate":"2026-10-08","eventDate":"2026-10-08","sourceName":"ibl.ai","sourceLabel":"Vendor case study restating its Oct. 8, 2026 press release and CIO remarks","sourceUrl":"https://ibl.ai/case-study/marshall-university","evidenceClass":"vendor-claim","outcomeClass":"emerging","topics":["infrastructure","developers-agents","governance-procurement","operating-model"],"finding":"Per the vendor, Marshall receives the full ibl.ai codebase (agent framework, APIs, LTI integrations, documentation) and runs it on infrastructure it controls; Marshall IT governs access, security and model choice, and no outside party trains on institutional data. The deployment begins with a portfolio of agents across admissions, transcript evaluation, financial aid guidance, advising, tutoring and IT support, which Marshall will evaluate and prioritize. CIO Jodie Penrod, speaking at EDUCAUSE 2026, framed it as a sandbox to measure usage and return on investment before deciding how to proceed. Pricing is usage-based, not per seat, and the platform is model-agnostic.","sledRelevance":"Campus-operations signal on architecture and procurement: an institution choosing self-hosted, source-owned agent infrastructure with usage-based pricing and ROI measurement, contrasting with per-seat enterprise licenses under renewal pressure elsewhere.","evidence":"Opened the ibl.ai case-study page dated Oct. 8, 2026, which quotes Penrod's EDUCAUSE remarks and states it draws on the Oct. 8 press release (EIN Presswire version returned HTTP 403 to direct retrieval).","architectureImplications":"Self-hosted agent platform with LTI integrations and model-agnostic routing; Marshall carries hosting, patching and integration work.","governanceImplications":"Institutional logs of every interaction enable usage and cost review; agent prioritization and evaluation criteria are not yet published.","securityPrivacyImplications":"On-premises/controlled hosting keeps data under institutional control, but shifts security responsibility to Marshall IT.","caveats":"Vendor marketing; no deployment, usage or ROI results. Code-ownership and no-training claims are vendor statements. Which agents go live is undecided.","streamIds":["campus-operations"],"roles":{"sales":"Sales framing: ownership and usage pricing are buying criteria. Marshall chose a source-owned agent platform as a sandbox to measure usage and ROI before committing. Use this with CIOs comparing per-seat licenses to self-hosted options. A bounded total-cost and capability comparison is credible; ROI, savings or agent effectiveness claims are not supported, as the vendor reports no results.","engineering":"Engineering framing: sovereignty transfers operational burden. Fit is institutions with capacity to host and maintain an agent platform. Assess hosting, patching, model-routing and LTI integration effort, and logging for cost per agent. Proof of value should deploy one or two agents (for example IT support) against campus systems and measure accuracy, cost per interaction and maintenance hours.","delivery":"Delivery framing: prioritize agents with owners and metrics. Marshall will select agents from a portfolio; delivery should require a business owner, baseline and success measure for each, plus a security review of self-hosted components. Risks include under-resourced platform operations, scope sprawl across many agents and vendor-dependence persisting despite code ownership."},"roleDetails":{"sales":{"customerProblem":"A campus wants to test AI agents across services while controlling data, models and cost before committing to per-seat licenses.","stakeholders":["CIO","Information security officer","Financial aid and admissions directors","Procurement / general counsel"],"discoveryQuestions":["Which agents will be prioritized and by what criteria?","What hosting and staffing does source ownership require?","How will usage and cost per agent be reported?","What happens to support if the vendor relationship ends?"],"valueHypothesis":"Interpretation: a usage-priced, self-hosted sandbox could inform buy-or-build decisions; no ROI is evidenced.","potentialEngagement":"A bounded platform-evaluation: TCO comparison with per-seat options and a two-agent pilot with cost and accuracy measures.","unsupportedClaims":["No usage, accuracy or ROI results.","Code ownership and no-training claims are vendor statements.","Agent effectiveness in campus systems is untested."]},"engineering":{"fit":"Institutions with platform engineering capacity seeking model-agnostic, self-hosted agents.","architectureIntegration":"Self-hosted agent framework with APIs, LTI to the LMS, model routing across providers and institutional interaction logs.","prerequisites":["Hosting infrastructure and platform team","Integration access to SIS/ITSM","Model provider contracts","Security review"],"constraints":["Institution bears operations and patching","Usage costs vary with model pricing","FERPA obligations for student-data agents"],"security":"Apply institutional identity, logging and vulnerability management to the self-hosted stack; verify model-provider data terms.","proofOfValue":["Pilot IT-support agent and measure resolution accuracy","Report cost per interaction by model","Track platform maintenance hours"]},"delivery":{"work":"Select initial agents, integrate with campus systems, and stand up usage/ROI reporting.","dependencies":["ibl.ai engineers","Campus systems owners","Model providers"],"ownership":"Marshall IT under the CIO, with service owners per agent.","skillsAdoption":"IT staff need platform-operations skills; service owners need training to evaluate agents.","governanceCheckpoints":["Security review before each agent goes live","Owner and baseline defined per agent","ROI review before expansion"],"acceptanceCriteria":["Proposed: each live agent has a documented baseline and target","Proposed: cost per interaction reported monthly","Proposed: no unresolved high-severity findings in platform security review"],"risks":["Platform operations under-resourced","Agent sprawl","Unproven vendor claims"]}},"retrievedAt":"2026-10-09T21:54:43-05:00","enrichedAt":"2026-10-09T21:54:43-05:00","enrichmentBasis":"retrieved source","accessibilityWorkforceImplications":"Penrod framed access for faculty, staff and students; the IT team takes on platform operations.","procurementImplications":"Usage-based pricing and code transfer are alternatives to per-seat enterprise licenses; contracts must define support after code transfer.","operatingModelImplications":"Positions IT as an internal AI platform provider rather than a license manager.","sourceVerification":{"openedUrl":"https://ibl.ai/case-study/marshall-university","referenceExcerpt":"Penrod framed ibl.ai as a sandbox: put AI in the hands of faculty, staff and students, then measure usage and return on investment before deciding","promptVersion":"sled-research-v4.0","model":null,"basis":"agent-reported inspection"}}}]}