{"resourceId":"jisc-knowledgeflow-pro-services-ai-20260914","versions":[{"version":"external-afe39c4523a1d5c3c316934b6e902680256ab0c95bbae022c7d3c3f60014d91f","resource":{"id":"jisc-knowledgeflow-pro-services-ai-20260914","title":"Jisc KnowledgeFlow pilot tests secure AI assistants for HE/FE professional services including finance, HR and administration","organization":"Jisc National Centre for AI","sector":"Higher and further education (UK)","geography":"United Kingdom","publishedAt":"2026-09-14","publicationDate":"2026-09-14","eventDate":"2026-09-14","sourceName":"Jisc National Centre for AI blog","sourceLabel":"Public-sector association pilot announcement","sourceUrl":"https://nationalcentreforai.jiscinvolve.org/wp/2026/09/14/exploring-ai-assistants-for-professional-services-the-knowledgeflow-pilot-gets-underway/","evidenceClass":"public-sector-association","outcomeClass":"emerging","topics":["knowledge-work","operating-model","governance-procurement","accessibility-workforce","data-security"],"finding":"On 14 September 2026 Jisc launched a FEA-I KnowledgeFlow pilot (14 Sep–18 Dec 2026) with ten FE/HE institutions including Keele University to test secure, institution-specific AI assistants for professional and operational services—explicitly including quality, policy, governance, data, finance, HR and administration. Priorities are knowledge access to authoritative sources, quality/reporting support, and natural-language questions on operational data. Evaluation is planned for January 2027; human oversight is stressed for consequential decisions. No outcome metrics yet.","sledRelevance":"Preferred-domain association evidence that campus-operations AI is moving from teaching tools into governed professional-services assistants—directly transferable as a model for U.S. multi-campus admin AI sandboxes, with UK policy/context limits.","evidence":"Opened Jisc blog lists ten participating colleges/universities, pilot window, three priority clusters (institutional knowledge; quality/planning/reporting; operational data/services including finance/HR/admin), and January 2027 full evaluation. States purpose is not only speed but quality, workload, governance and security in real workplaces.","architectureImplications":"Institution-specific assistants grounded in campus policy/procedure corpora with pointers back to authoritative sources; multi-assistant environment vs shadow general-purpose tools. Implies RAG over controlled document stores, identity-aware access, and logging—not open web chat for admin data.","governanceImplications":"Pilot frames human oversight where decisions affect staff/students; sector evaluation in 2027 will set evidence bar. Campuses should define which finance/HR/admin tasks are assistive vs decision-making before scaling.","securityPrivacyImplications":"Secure, customisable platform claim still requires local data-processing agreements, access control to policy/finance corpora, and monitoring for sensitive operational data in prompts/outputs.","caveats":"Pilot start announcement; findings not due until 2027. Mostly FE colleges plus one university (Keele); UK tertiary context. Vendor platform (FEA-I KnowledgeFlow) capabilities not independently benchmarked here.","streamIds":["campus-operations"],"roles":{"sales":"Jisc’s KnowledgeFlow pilot is a credible opener for professional-services AI in higher ed: ten institutions will test secure assistants for policy, quality, finance, HR and administration through December 2026, with evaluation in January 2027. Sell a similarly bounded admin-knowledge pilot—not a full ERP agent—focused on retrieval to authoritative sources and drafting support under human review. Emphasize that Jisc itself refuses to treat speed alone as success. Unsupported: any claim that KnowledgeFlow already proved finance/HR savings.","engineering":"Architecture pattern is institution-scoped assistants over policy/procedure and operational data stores, with routing back to authoritative documents and multiple specialist assistants in one environment. Prerequisites: curated corpus, identity tiers for finance/HR content, evaluation instrumentation for accuracy and workload, and clear non-decision boundaries. Proof of value: staff task samples comparing assistant-aided vs baseline time/error rates on policy lookup and report drafting—not production autopilot on HR actions.","delivery":"Delivery mirrors Jisc’s onboarding, training, and peer-learning model for professional-services AI. Select two or three admin units, load only approved policy and procedure corpora, train staff on source-checking and escalation, and run a fixed evaluation window with mid-pilot accuracy review before any wider release. Ownership sits with digital or professional-services leadership plus unit subject-matter experts. Acceptance criteria should require source-linked answers, documented human decision ownership for consequential cases, and a written limitations register. Risks include policy hallucination, scope creep into HR decisioning, and anecdotal-only evaluation."},"roleDetails":{"sales":{"customerProblem":"Admin staff waste time hunting policies and assembling routine quality/finance/HR documentation across fragmented systems.","stakeholders":["COO / VP Administration","Finance and HR leads","Quality/planning offices","CIO","Data protection officer"],"discoveryQuestions":["Which policy corpora are authoritative and kept current?","Which finance/HR tasks are assistive drafting vs decisioning?","What accuracy and workload metrics will the pilot capture?","How will outputs be traced to source documents?"],"valueHypothesis":"Secure institution-specific assistants can reduce search and drafting time for professional services if grounded in authoritative sources and human-reviewed—Jisc is testing this; results are not yet published.","potentialEngagement":"12-week professional-services AI pilot on policy lookup and quality-report drafting with January-style evaluation rubric.","unsupportedClaims":["Proven finance or HR FTE savings from KnowledgeFlow","Safe autonomous decisions in consequential admin cases","Immediate transfer of UK FE results to U.S. R1 campuses"]},"engineering":{"fit":"Multi-assistant RAG platform for professional services knowledge, quality reporting, and operational data questions.","architectureIntegration":"Institution-specific assistants over controlled document/data stores; identity-aware access; citation back to authoritative sources; optional analytics on operational indicators where permitted.","prerequisites":["Curated policy corpus with owners","SSO and role-based corpus access","Logging/evaluation hooks","DPA/processing terms for FEA-I or equivalent"],"constraints":["UK tertiary pilot mix (mostly FE)","Outcomes deferred to Jan 2027","Operational-data Q&A may touch regulated datasets"],"security":"Keep sensitive finance/HR corpora in approved stores; block exfiltration to consumer AI; require human confirmation for consequential outputs.","proofOfValue":["Source-linked answers on a gold-question set","Measured drafting-time delta on agreed report types","Documented escalation rules for consequential cases"]},"delivery":{"work":"Onboard units, load corpora, train staff, run fixed pilot window, collect usefulness/accuracy/governance feedback, decide scale/stop.","dependencies":["Unit content owners","Identity and data protection teams","Vendor onboarding support","Evaluation design"],"ownership":"Digital transformation / professional services lead with optional sector peer learning.","skillsAdoption":"Staff training on prompting, source-checking, and when not to rely on assistants—paired with literacy programs.","governanceCheckpoints":["Corpus approval before pilot","Mid-pilot accuracy review","End-pilot go/no-go with risk register"],"acceptanceCriteria":["Agreed share of answers cite authoritative sources","No unapproved consequential decisions by assistants","Workload and limitation findings documented"],"risks":["Policy hallucination accepted as authoritative","Pilot scope creeps into HR decisioning","Evaluation delayed or anecdotal only"]}},"retrievedAt":"2026-09-30T21:56:45-05:00","enrichedAt":"2026-09-30T22:05:00-05:00","enrichmentBasis":"retrieved source","accessibilityWorkforceImplications":"Explicitly about professional-services staff workload and confidence; accessibility of assistant UX and alternatives for staff who cannot use AI tools should be planned. Bargaining/consultation norms may apply when tooling changes job design.","procurementImplications":"Sector-facilitated pilot of a named platform (FEA-I KnowledgeFlow)—campuses still need local contracts, data terms, and exit plans if moving from pilot to service.","operatingModelImplications":"Shifts professional services toward AI-assisted knowledge work with human judgment retained; implies new content-ownership and quality roles for policy corpora.","sourceVerification":{"openedUrl":"https://nationalcentreforai.jiscinvolve.org/wp/2026/09/14/exploring-ai-assistants-for-professional-services-the-knowledgeflow-pilot-gets-underway/","referenceExcerpt":"explore how secure, institution-specific AI assistants might support areas such as quality, policy, governance, data, finance, HR and administration","promptVersion":"sled-research-v4.0","model":null,"basis":"agent-reported inspection"}}}]}