Fostering Agency Software Checklist: What to Check and Where AI Fits

A small fostering agency should check five things before buying case-management software: where records and panel papers live, how supervision logs are stored and found, who can see which child's file, what leaves the system, and how you export everything if you leave. AI fits into admin drafting only, never assessments or recommendations.

That order matters because the usual buying process runs the other way. A vendor opens with a demo of the clever features, an AI note summary that looks magical, and the hard questions come last, if at all. Records about children and carers are among the most sensitive a small organisation holds. A tool that is brilliant at summarising and weak at access control is the wrong tool.

Follow me on Instagram@sagnikteaches

Start with what the agency must be able to prove

Software for a fostering agency exists to answer a short list of questions, usually under time pressure, usually from someone who is not in the room. Can you show every contact with this household since approval? Who approved this placement, and on what papers? When was this carer's last supervision, and what actions came out of it? When did this person last see the file? Write those questions on a page before you contact a vendor. Each one becomes a test in the demo.

Connect on LinkedInSagnik Bhattacharya

A worked example, as an illustration. A small agency with 14 approved households and about 20 children placed has three social workers and one administrator. Three hours of the administrator's week go on assembling panel papers from emails and shared folders. The supervising social workers lose roughly 20 minutes after each visit finding the previous visit's actions. The manager spends two days before any inspection building an evidence folder. Those are the three costs to compare against a software fee, and they tell you what the software must do: hold one record per household, produce a panel pack from it, and show the last visit's actions beside the new note.

Subscribe on YouTube@sagnikteaches

The record-keeping checks, group by group

Score each item below as yes, partly or no during demos, and ask to see it done live on dummy data, not described on a slide.

Records and structure

• One record per household, linked to each child and each adult. Ask what happens when a carer moves household or a child moves placement. The history should stay and the links should change, not be overwritten.
• Chronology generated from entries, not typed twice. A chronology that depends on someone remembering to add a line will be wrong by the next inspection.
• Dated, attributed entries. Every entry shows who wrote it and when, and late edits show as edits, not silent changes.
• Required fields and checklists that you can edit yourself. If adding a field means a paid change request, the system will drift from how you work.
• Document storage with versioning. A report that gets revised three times needs three versions, not one overwritten file.
• Search that respects permissions. A search for a surname should not reveal records the searcher is not allowed to open.

Panel papers and approvals

• Can the system assemble a panel pack from the household record, with a contents list, in a fixed order, as one file?
• Can panel members be given time-limited access, or does every pack have to be emailed? Email is the weak point in many agencies.
• Does it record the panel's date, attendance, recommendation and the decision-maker's decision as separate items with dates?
• Can you see all carers whose reviews, checks or training are due in the next 60 days, in one list?
• Are reports and minutes locked once approved, with later changes stored as new versions?

Supervision logs

• A template for each type of visit, such as supervisory visit, unannounced visit and carer supervision, that you can edit.
• Actions that carry an owner and a due date, and roll forward on the next visit's template until closed.
• Visit frequency rules that flag overdue visits automatically, using intervals you set.
• A way to see the carer's, the child's and the worker's views together, with the child's own voice recorded in a dedicated field.
• Offline or poor-signal use, because many visits happen where the signal is weak.

Access controls and audit

• Roles that match your agency: manager, supervising social worker, administrator, panel member, finance, agency decision-maker. Ask to see the role editor.
• Access by caseload, not only by role, so a worker sees their own households and a manager sees all.
• An audit log of views as well as edits. Ask who opened a given file in the last 30 days and watch the vendor produce the answer.
• Two-step sign-in for every user, and a way to remove a leaver in minutes.
• Restricted-record flags for sensitive files, such as a family with a safeguarding concern or a carer who is related to a staff member.

Sharing and exit

• How are reports shared with placing authorities, panel members and carers? Look for a secure portal or encrypted sharing, not attachments.
• Where is the data hosted, and can the vendor tell you in writing? Ask for a list of sub-processors.
• What is the exit route? A complete export of records, documents and audit logs in an open format, with a price and timescale. This matters for fostering agencies in particular, because records may need to be kept for many years after a child leaves care. Check your own retention duties with your regulator or adviser.
• What are the price rises at renewal, the notice period, and the minimum term?

Where AI fits in a fostering agency, and where it must not

AI in this setting needs a firm boundary, set before any vendor demo. The rule: AI may help with the agency's own admin, and it may never replace professional judgement about a child, a carer or a placement.

The first three rows are non-negotiable for any agency. Notes about a child's behaviour, a carer's marriage or an allegation are the kind of material that must never be pasted into a consumer chatbot, even with names removed, because circumstances alone can identify a family in a small area. The related tutorial on supervision templates with AI notes and action logs shows how a note can be drafted from a worker's own bullet points without the AI ever deciding anything.

How to test the AI features in a demo

Ask for three things, each on dummy data you supply:

1. Switch it off. Can the agency turn the AI features off for the whole organisation, or by role? If a feature cannot be disabled, you cannot control it.
2. Show me where the data goes. Which model provider receives the text, is it kept, and is it used to train anything? Ask for the answer in writing, and for the retention period.
3. Show me a wrong answer. Feed it an inconsistent set of notes and see whether the summary invents a detail. A summary that adds "the carer was cooperative" when no note said so is making an assessment.

A sample bullet list, and the kind of draft you want and do not want, helps here. The dummy bullets: "visit 14 Oct, 45 min; child at school, saw carer only; carer reports sleep improving; contact with birth family changed to fortnightly; training course booked; action: confirm respite dates." A good draft restates those lines in order, flags that the child was not seen, and lists the action. A bad draft says "the placement is stable" or "no concerns were raised". Nobody wrote either, and the second is the kind of sentence that surfaces later in an inspection or an enquiry.

A scoring sheet for comparing three vendors

Create a simple table, one row per check above, with three columns for the vendors. Score 2 for done live, 1 for described or on the roadmap, 0 for no. Weight the first five rows three times:

Weight 3: audit log shows views; roles by caseload; full export; panel pack in one file; AI can be switched off
Weight 2: chronology from entries; versioned documents; overdue-visit flags; offline use; two-step sign-in
Weight 1: reports to templates; carer portal; branded letters; calendar link

Add the totals, then look at the cheapest vendor's zeros. A zero on a weight-3 line is a stop, not a deduction. In the example agency, vendor A scored well on features but could not show who had viewed a file; vendor B scored lower overall but passed every weight-3 test. The agency chose B, because a missing audit log is expensive to live with and hard to retrofit.

Moving from email and folders without losing the trail

The switch itself is the riskiest month. Plan it in four steps. First, list every folder, spreadsheet and inbox where carer or child information sits today. Second, pick a cut-off: all new records go into the new system from a set date, and old records move household by household, starting with active placements. Third, check each migrated file against the original, and keep the original read-only for an agreed period. Fourth, delete the stragglers: personal drives, desktop copies and forwarded emails. A migration that leaves old copies scattered around has doubled the exposure instead of reducing it.

Budget the staff time as well as the fee. For the 14-household example, migrating records might take a worker an hour or two per household, so 20 to 30 hours in total. Treat that as an illustration and ask the vendor how they handle imports, and whether it is included in the price. For sample questions on the AI side of the contract, the security questions to put to an AI scribe vendor lists the retention and training questions that apply to any AI note feature, even outside healthcare.

An access test to run live on dummy data

Access control is where a polished demo most often hides a weakness, so run this test yourself, on dummy records, in a trial login. Do not rely on a screen recording. Set up three dummy users and two dummy households, then work through the steps.

1. Create a supervising social worker with only Household A on their caseload. Log in as them and search for Household B's surname. The record should not appear, and ideally the search should say nothing about whether it exists.
2. Type the direct web address of a Household B document into the same browser. It should refuse. Some systems hide a link but still serve the file to anyone who knows the address.
3. Log in as the administrator and open a restricted record. Does the system ask for a reason, or at least log the view?
4. Remove the social worker from the system. Do any open sessions, mobile apps or saved links keep working?
5. As manager, ask for the log of everyone who opened Household A in the last seven days. How long does the answer take, and can you export it?
6. Download a panel pack as a panel member with time-limited access, then try again after the access expires.

Write down each result in the scoring sheet. A vendor that passes five of the six and explains the sixth honestly is better than one that passes all six and cannot say how. A realistic failure in the third step is a system that logs edits but not views; in a small agency that means nobody can later tell whether a curious colleague read a sensitive file, and you cannot answer a carer who asks.

What the software costs once you count the whole year

Fees differ by vendor, and none are quoted here because pricing for this niche is usually by quotation. Ask each vendor for the same set of figures, in writing, so you can compare like with like. A sample comparison, with invented figures for illustration only, for the 14-household agency over three years:

The point of the table is the lines most buyers forget: the AI add-on and the exit fee. On those figures vendor A has the cheaper licence, yet it costs $4,400 more over three years once set-up, the AI add-on and the exit fee are counted. Whatever the real numbers are, ask for the exit fee and the renewal cap before you ask about features.

The first 90 days after you sign

Buying is half the job. In the first month, set roles and caseloads, switch the AI features to off, and write the one-page rule on what staff may and may not do with them. In the second month, migrate active households, run the system alongside the old folders for one supervision cycle and check every new record against the old one. In the third month, run your first overdue-visit report and your first mock panel pack, and compare it with the old way. Note the minutes saved, but also note surprises, such as a field the vendor's template lacks. Raise those with the vendor while you still have goodwill and a quarter of the contract ahead.

Only after that, and only for admin tasks, consider switching on one AI feature at a time: a supervision-note tidy-up first, because it is the easiest to check. Keep a short log of each AI draft that needed a correction and what the correction was. After a month, that log tells you whether the feature is saving time or adding risk. If the log shows invented details, switch it off again and tell the vendor.

Four red flags that should end a demo

• "AI summaries are included" with no way to switch them off. Ask for the setting, and if they cannot show it, walk away.
• A vendor that cannot show a view log in the demo. "It is on our roadmap" is not a control.
• Export as PDF only. PDFs of each page are not a portable record. You want structured data plus documents.
• Pricing per child record with a fee to archive. Your costs can rise just because children left. Ask whether archived records count.

A related caution on features that are already in tools you own: the guide to AI features already in your software explains how to find and switch them off, which is relevant if a vendor turns something on after an update. For how care-sector vendors price AI as an add-on, see home care agency software pricing and what the AI features add; the pattern is similar.

Four small-agency situations

These are illustrations of how the checklist bends:

• A start-up agency with four households. A full system may be heavy. Choose software with a low entry tier but still test the view log and export, because the records you create now are the ones you will have to keep.
• An agency that places sibling groups. Check that records can link siblings in different households and that access rules follow the child, not the carer.
• An agency using a part-time social worker. Ask how access is granted and removed for contractors, and whether their sessions expire.
• An agency about to be inspected. Do not migrate in the three months before. Tidy your existing records, add the new system afterwards, and use the checklist to prepare evidence.

Finally, give someone the job of reading the vendor's terms. If you would like a second reader on the shortlist and the AI settings, that is the sort of problem my AI implementation consultation covers. Nothing in this checklist is legal advice, and your regulator's expectations on records and retention come first.

Questions fostering managers ask about software and AI

Can AI help write a foster carer assessment?

It should not. An assessment is a professional judgement about a family, made by a qualified social worker who has met them. AI can tidy the layout of your own notes or check a draft for missing sections, but the evidence, analysis and recommendation must be the assessor's own, and nobody should paste applicant details into a general chatbot.

Is a spreadsheet enough for a very small agency?

For a handful of households, a locked-down spreadsheet can work for a time, but it struggles with audit trails, access by role, deadlines and safe sharing with panel members. Once you have more than a few carers, or any inspection scrutiny, dedicated software usually earns its fee.

What should we ask a vendor about data when we leave?

Ask for a full export of every record, attachment and audit log in an open format, for a stated price and timescale, and ask what the vendor deletes afterwards and when. Get the answer in writing before you sign, because the moment you want to leave is the moment your bargaining power is lowest.

Do we need to tell carers and children we use AI?

If AI touches anything about them, say so in plain words in your privacy notice and in carer information packs, and check your regulator's expectations. Better still, keep AI away from children's records entirely until your policy, a data protection impact assessment and your vendor contracts are in place.

Further reads

• Children's Home Software: Which Systems Add AI and What to Check First — A sister checklist for children's homes with similar records.
• How to Write an AI Usage Policy for Your Small Business — Draft the staff rule on what AI may touch.
• Is It Safe to Use AI With Children's Data in a Nursery? — Why children's data needs a firm line with AI.
• Nursery Staff Training: A One-Hour AI Briefing on Children's Data — A short staff briefing you can adapt for carers and staff.
• Use the AI Already in Your Practice Software Before Buying More — Check AI already inside your software before buying more.
• Tutor Safeguarding Checklist: Policy, Checks and Records With AI — Another safeguarding-heavy agency with similar records.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: no vendor is named or priced here; the checklist is built from common record-keeping, access-control and contract questions, so verify each answer with the vendors you shortlist.