School Liaison That Does Not Become a Second Job

The operational problem

School meetings matter. Chasing the same facts across five inboxes should not.

This brief applies the leadership approach in A Practical AI Roadmap for NGO Leadership Teams to one workflow. This is an operational collaboration brief, not clinical, educational or safeguarding advice. Follow local information-sharing agreements, safeguarding arrangements and lawful-basis requirements.

School liaison preparation is a good candidate for the administrative-burden pattern the wider sector already reports, precisely because it is rebuilt from scratch each time rather than maintained continuously. The Federation of European Social Employers put the Dutch care-worker administrative share at roughly 36% of working time in 2023, a figure driven in part by exactly this kind of cross-referencing work, and worker representatives in Austria and Belgium separately described documentation getting harder to keep straight as more systems were added rather than fewer. A liaison lead’s Wednesday evening is a plausible local instance of that pattern, worth timing directly rather than assumed.

Eurofound’s 2023 social-services report ties the same pattern to staff shortages and uneven digitalisation across EU care services, which matters for liaison work specifically because the lead is usually reconciling information that lives in systems belonging to other professionals, not just their own. And because close to a fifth of social-care workers in the EU never use a digital device at work, any fix built around a shared digital pack has to have a working non-digital equivalent, or it simply moves the burden from the liaison lead to whichever colleague cannot access the system.

What a real week looks like

A meeting is booked for Thursday. On Wednesday the liaison lead locates the current plan, checks whether a strategy changed, asks a therapist for an update, finds a parent email and reconstructs the last action list. In the room, people repeat information because nobody is certain which version is current. After the meeting, actions live in personal notes until the next chase begins.

The useful product is a proportionate meeting pack, not a larger dossier. It should show what has changed, what the school needs to know, what remains uncertain and who owns the next action. That requires an agreed source of truth and an information-sharing decision before any summarisation begins. The Department for Education’s 2026 draft guidance and the ICO data-sharing code support timely, documented and proportionate sharing; the European Commission’s disability-rights strategy keeps participation and inclusion in view.

The liaison lead who prepares these meetings is the only person who can show where the reconstruction time actually goes, because they are the one chasing the therapist for an update and searching the inbox for the last confirmed version. A map built by that person, across a handful of comparable meetings, will separate genuine relationship-building time from pure retrieval time in a way a policy review from outside the workflow cannot.

What the evidence says in 2026

The Charity Digital Skills Report 2026 (launched 9 July 2026, n=807) found 79% of charities using AI, up from 76% in 2025 and 61% in 2024, with 92% among large charities, while only 28% had a documented digital strategy, down from 44% the previous year, and 33% of boards were rated poor on AI skills. For a workflow that already involves moving information between organisations, that governance gap is the number to sit with: it means most charities have not yet written down what an AI tool may see when it is helping pull together a pack that draws on records from a school, a therapist and a family, several of which are not the organisation’s own data to begin with.

The barriers reported alongside that adoption point towards keeping any pilot narrow: 63% named squeezed finances as the biggest digital barrier, 56% named a lack of skills or technical expertise, and 35% distrusted AI tools outright. Staff training was the top funding need for 44%, yet only 17% had received dedicated digital funding. A liaison lead juggling several children’s meetings a week is not well placed to also become the organisation’s data-sharing policy expert, so the first pilot has to come with the governance decisions already made, not left for them to work out mid-preparation.

There is a specific reason liaison packs are a reasonable place to test assistance: the same survey found 63% of charities already using AI for administration and project management, evidence that pulling together and formatting confirmed material is a task the sector has found genuine use for. What it has not tested at scale is whether that assistance can be trusted with information that originates outside the organisation’s own systems, which is the harder problem this workflow actually presents.

Where technology helped - and where it stalled

A controlled pack can pull current goals, agreed strategies, questions and actions from approved records. It stalls when it creates a shadow record, includes more detail than the school needs, or confidently summarises a historic plan. More software cannot restore liaison time while the authoritative record is unclear.

For a liaison pack, the honest boundary is drawn at the proportionality decision, not the formatting. A tool can pull together confirmed goals and actions from approved records and flag where a required field is missing. It cannot decide that a piece of information is necessary for the school to know, or that sharing it is lawful and proportionate under the relevant information-sharing agreement. Handing that judgement to a drafting tool because it produces a comprehensive-looking pack is how over-sharing happens quietly, one meeting at a time, without anyone deciding to do it.

The governance gap the 2026 survey found, a third of boards rated poor on AI skills, is a particular liability here because liaison packs routinely draw on records the organisation does not own outright, from a school, a therapist or a family. Without a clearly named data boundary, a liaison lead under Wednesday-evening pressure may let a tool pull in more than the meeting actually requires, and there is no natural point in the workflow where anyone would notice before the pack is already in front of the school.

Wednesday evening is an ownership gap before it is an AI gap

Not every part of Wednesday evening needs an AI system in the loop. The liaison lead’s reconstruction work, chasing a therapist for an update, checking whether a strategy changed, happens largely because no single record has been agreed as authoritative across the school, the therapist and the family. That is a source-of-truth and information-sharing decision this brief already calls for, and naming it honestly matters more than adding a summarising tool on top of a record nobody has fixed.

The pattern worth watching sits across a caseload of these meetings, not inside any single one. A liaison lead preparing several children’s meetings a week faces the same reconstruction repeatedly, and a bounded agentic layer that reads confirmed updates across that caseload can flag, ahead of Thursday’s meeting rather than during Wednesday’s chase, that a strategy has not been confirmed in the expected window or a required field is still missing, then prompt the named professional responsible rather than waiting for the liaison lead to discover the gap alone. That is the proactive, boundary-respecting use of agentic systems this series argues for: reasoning about recurring risk across many meetings rather than answering a single query. McKinsey’s 2025 State of AI survey found that adoption without redesigning the underlying workflow, in this case the record and information-sharing agreements this brief describes, rarely produces a measurable gain however fluent the resulting pack looks, a fair test for any vendor proposing a liaison-pack assistant without first asking whether those records already work. SAS’s 2026 research on keeping automation in service of judgement, rather than replacing it, describes the alternative worth building instead.

A safer AI-assisted path

1. Map preparation and follow-up for a small set of comparable meetings. Include the time spent locating facts, checking consent, asking for updates and recording actions. The liaison lead builds this map by hand; no AI role here, because the organisation must see where reconstruction time actually goes before trusting an assistant with any part of the pack.

2. Agree a minimum pack with a named owner, version date, audience and expiry. Separate confirmed information from questions for the meeting, and include only material that is necessary and proportionate. Information-sharing discipline is human-only: a tool may later pull from this pack structure, but it cannot decide lawfulness or proportionality.

3. Use assistance for bounded tasks such as comparing versions, checking required fields and formatting confirmed content. The liaison lead must decide accuracy and proportionality, and the relevant professionals must confirm discipline-specific updates. Never let a generated summary become a new source of truth.

4. For four weeks, audit whether packs were current, actions had owners and unnecessary chase messages fell. Compare that audit with proactive flags from a bounded agentic layer across the caseload, such as strategies not confirmed in the expected window or required fields still missing before Thursday’s meeting. The liaison lead decides whether to expand the pilot only when those signals and the audit both show the record and action route are holding.

Where humans must intervene

The liaison lead decides what is accurate and proportionate. Families and the child or young person should be involved according to local practice, while safeguarding concerns use dedicated escalation routes. Information-sharing agreements and lawful basis should be recorded before material is sent outside the organisation.

NIST’s Generative AI Profile maps onto liaison preparation reasonably directly: know what the drafting tool cannot judge (proportionality of disclosure, chiefly), keep provenance so every item in the pack can be traced to a confirmed source, and treat the finished pack as a consequential output that needs a named human sign-off before it leaves the organisation, every time. GDPR and, where it applies, the EU AI Act, govern what may be shared outside the organisation regardless of how close the meeting is.

Risks and failure conditions

A polished, complete-looking pack is not the same as an accurate or proportionate one, and that gap is the central risk of this workflow: a shadow folder growing alongside the official record, a teacher planning around a strategy that has since changed, meeting actions that nobody owns once everyone has left the room, and a fluent summary being treated by a busy sender as if it were the same thing as a considered decision to disclose. The design that avoids this makes the current version and its owner obvious at a glance, so nobody has to guess which pack is the real one.

A second risk is reading general sector movement as proof this specific workflow has improved. The 2026 report found 81% of charities reporting some digital progress, a figure broad enough to cover almost any change anywhere in the organisation. It says nothing about whether a liaison lead actually got Wednesday evening back, whether a school received a more accurate pack, or whether over-sharing quietly increased instead. Only a direct audit of this workflow’s own packs answers that.

Measures that matter

Measure median preparation time for routine meetings, the share with a current dated pack and named actions, and the number of duplicate update emails per child per month. Read these alongside an audit of proportionality and action completion, because shorter preparation is not enough if it produces an unhelpful meeting.

Ask the liaison lead directly, after each meeting cycle, whether Wednesday evening actually got shorter or just tidier. If the pack looks better but the reconstruction time has not moved, the pilot has fixed the presentation of the problem rather than the problem itself.

Decision questions

1. What information does the school genuinely need for this meeting? Anything beyond that answer is a proportionality risk dressed up as thoroughness.
2. Which record is authoritative and who keeps it current? Without a clear answer, the pack will keep being rebuilt from scratch because nobody trusts any single source.
3. What is the lawful and proportionate basis for each item shared? If that cannot be stated for a specific item, it should not be in the pack.
4. How will actions be assigned and checked after the meeting, not just recorded during it? An action with no owner by Friday is an action that will resurface unresolved at the next meeting.
5. Could a new liaison colleague prepare this pack without inbox archaeology? If the answer is no, the workflow still depends on one person’s memory rather than the organisation’s record.

FAQ

Is a single pack always appropriate? No. Its content and audience should be proportionate to the meeting and information-sharing agreement. A shared format is useful; indiscriminate sharing is not.

What can technology safely do? It can locate confirmed material and flag gaps. People must decide accuracy, relevance and what may be shared.

References

1. David Saliba, A Practical AI Roadmap for NGO Leadership Teams, AIMonger (2026). https://aimonger.com/whitepapers/ngo-practical-ai-roadmap-leadership/
2. Federation of European Social Employers / research partners, Evolving jobs, skills and training needs in the social services sector and the role of social partners in managing changes (2023). https://www.socialemployers.eu/wp-content/uploads/2023/09/report_Evolving_jobs_skills_and_training_needs_in_the_social_services_sector.pdf
3. Eurofound, Social services in Europe: Adapting to a new reality (2023). https://www.eurofound.europa.eu/en/publications/all/social-services-europe-adapting-new-reality
4. Charity Digital Skills Report 2026 (Zoe Amar Digital / Nissa Ramsay; launched 9 July 2026, n=807). https://charitydigitalskills.co.uk/report/
5. NIST, Artificial Intelligence Risk Management Framework: Generative Artificial Intelligence Profile, AI 600-1 (2024). https://nvlpubs.nist.gov/nistpubs/ai/nist.ai.600-1.pdf
6. Regulation (EU) 2016/679 (GDPR). https://eur-lex.europa.eu/eli/reg/2016/679/oj
7. Regulation (EU) 2024/1689 (EU AI Act). https://eur-lex.europa.eu/eli/reg/2024/1689/oj
8. Department for Education, Information sharing duty: draft statutory guidance (2026). https://consult.education.gov.uk/mais-strategy-policy-and-programme-unit/statutory-guidance-for-information-sharing-duty/supporting_documents/information-sharing-duty-draft-statutory-guidance-for-safeguarding-organisations-and-their-practitionerspdf
9. Information Commissioner’s Office, Data sharing: a code of practice. https://ico.org.uk/for-organisations/uk-gdpr-guidance-and-resources/data-sharing/data-sharing-a-code-of-practice/
10. European Commission, Strategy for the Rights of Persons with Disabilities 2021-2030. https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=CELEX%3A52021DC0101
11. McKinsey & Company, “The State of AI: Global Survey” (2025 edition). https://www.mckinsey.com/capabilities/quantumblack/our-insights/the-state-of-ai
12. SAS, “For Agentic AI ROI, Invest in Human Judgment” (22 July 2026). https://www.prnewswire.com/news-releases/sas-for-agentic-ai-roi-invest-in-human-judgment-302831330.html