Which Parts of Care Need a Person?

A leader’s guide to keeping the relationship at the centre as AI enters social service work

It is Friday afternoon at a family service centre. A social worker has six home visits’ worth of case notes to write before the weekend. An AI tool could draft them in minutes from her voice memos.

Her supervisor is unsure. Will the notes still sound like someone who was in the room? Will a family ever read one and feel processed? And if the drafting goes to a machine, what else might follow?

Nobody in the room is against technology. They are asking a harder question.

The question underneath

Social service work creates value through relationships. In casework, counselling and outreach, the presence and judgement of a person is part of what is being delivered.

Generative AI can draft, summarise, translate and respond. Those functions sit close to the relationship, and sometimes inside it. So every agency adopting AI faces the same question, whether it names it or not: which parts of our work depend on a person, and which do not?

There is no settled answer. One leader may see AI as a threat to the quality of care. Another may see it as protection, absorbing the paperwork so staff can be more present. A third may find the answer differs from one service to the next. All three can be reasonable. What matters is that the question gets answered deliberately, by the people who understand the work.

Why Singapore leaves the decision with you

As at September 2026, Singapore has no standalone AI law. IMDA’s Model AI Governance Frameworks offer voluntary guidance, and existing laws such as the PDPA apply when AI handles personal data.

Where guidance is advisory, the organisation becomes the place where judgement happens. Leaders are not ticking a compliance box. They are deciding, and those decisions shape what care looks like in their agency. That is a responsibility worth taking on openly.

Draw the line together

Here is a practical way to make the decision.

1. Map the work around the relationship and the work inside it. List the tasks in one service. Some sit around the relationship: case notes, funder reports, scheduling, translating information for families. Some sit inside it: conversations, judgements about care and risk, moments when a client would expect a person.

2. Ask the people who do the work. Staff know where the relationship lives in their day. Ask what they would gain from help with the work around it, and what they would never hand over. Where appropriate, ask the people you serve too.

3. Write the line down. Decide which tasks AI may support, which need a person to review, and which stay with a person. A written line protects staff and clients, and gives your board something concrete to approve.

4. Review it. Set a date, perhaps 90 days out, to check whether the line is holding and whether the time saved is going back to people.

What the time is for

Saving hours on admin is only the start. The aim is to transform how the organisation delivers its mission, guided by its strategic goals.

The time AI releases has to go somewhere. It can become more home visits, shorter waiting lists, a new programme, or reaching families the agency cannot reach today. Decide where it goes before the tool arrives, or it can quietly disappear into the next round of paperwork.

One story, many voices

A change like this is heard by many people at once: staff, volunteers, the people you serve, the board and funders. Each will build their own version of what is happening if leaders leave a gap.

I use a Story World approach to close that gap: one shared story of the change, set by leadership, with room for each group to add its experience. Staff hear what happens to their role. The board hears the case and the safeguards. Families hear what stays human. When everyone can tell the same story in their own words, the change has somewhere to land.

Where to start

Begin with one service and one conversation with the people who deliver it. My free guide, The Question Before the Tool, sets out five questions to answer before you adopt any AI tool.

In your organisation, which part of the work would you never hand to a machine, and have you ever written it down?

rYo