How to use the AI client helper

Each client portal includes an AI helper, branded as "Ask the AI helper" or whatever name you configure. It's a chat interface scoped to that specific client's portal: their files, their approvals, their deadlines, their invoices.

It's not a general-purpose chatbot. It's a contextual assistant that helps clients self-serve answers to "what's pending from me?" or "when is my next deadline?" without emailing your team.

What it does well

The helper is good at:

• "What's pending from me?" Returns the open file requests, pending approvals, and upcoming deadlines for this client.
• "When is my [filing/deliverable/review] due?" Reads the deadline tracker, returns the date and what's needed.
• "What did we agree on last week?" Searches recent approvals and messages, returns the decisions with citations.
• "How do I upload a file?" Explains the portal UI, points to the right section.
• "What does [your firm name] do for me on [recurring task]?" Reads the FAQ block you configured during setup, summarises in plain English.

The helper cites the specific items it pulled from. A typical answer reads:

You still owe us your P60 from the BBC contract (due 15 January, see file request "P60 2025-26") and your dividend voucher from Pinpoint Ltd (due 15 January). Once those arrive we can run the draft return.

Citations are clickable. The client jumps from the helper answer straight into the relevant item.

What it doesn't do

Equally important to know:

• It doesn't give legal, tax, financial, or medical advice. The model is instructed to defer to "your firm" when asked for substantive advice. "Should I take the higher-rate pension contribution?" → "That's a decision your accountant should walk through with you. Open a message with the firm to discuss."
• It doesn't read attached files in detail. It knows files exist (name, category, upload date) but doesn't summarise their contents. For documents inside the portal, the client opens them directly.
• It doesn't act on the client's behalf. It can't approve invoices, sign engagement letters, send messages, or trigger any state change. Read-only by design.
• It doesn't remember conversations across sessions. Each helper session starts fresh. This is a privacy and predictability decision.

Cost per query

The helper uses Claude Haiku via Kie.ai. Each query costs approximately:

• €0.02 for a short answer (most "what's pending" queries)
• €0.03-0.05 for longer answers with multiple citations
• €0.10+ for unusual queries (very long file lists, complex recap requests)

Average cost across a typical workspace: €0.025 per query.

The monthly cap

Each workspace has a per-month query cap, enforced across all clients. Defaults:

• Workspaces on a slot pack: 500 queries/month
• Workspaces on the optional Agency Plan subscription: 2,000 queries/month

You can see the current month's usage at /app/settings/ai-usage. The page shows:

• Total queries this month
• Cost incurred this month (computed at €0.025 per query average)
• Per-client breakdown
• The cap and the percentage consumed

When the cap is reached:

1. Clients see a "AI helper paused for the month, resumes [date]" message
2. The chat input is disabled
3. Clients can still send a normal message to your team via the portal Messages section
4. You get a notification: "AI helper cap reached for [workspace] this month"

You can raise the cap from Settings → AI usage → Raise cap. Raising the cap is a per-month addition (€10 per +200 queries). The raise resets at the start of the next month.

Disabling the helper

You can disable the helper at three levels:

Workspace-wide: Settings → AI usage → toggle off. The helper disappears from every client portal.

Per-client: /app/clients/<client> → AI helper → toggle off. The helper disappears for that one client's portal.

Per-category: Settings → AI usage → Categories. Disable the helper for clients tagged with specific categories (e.g., "litigation" clients see no helper).

Disabling is useful when:

• Your clients are in sensitive workflows (litigation, employment disputes, personal injury) where you don't want any AI in the loop.
• A specific client is uncomfortable with AI tools.
• Your firm's professional regulations or insurance policies require human-only client communication.

Configuration

In Settings → AI helper, you can configure:

• Helper name: defaults to "AI helper." Some firms prefer "Portal helper" or "[Firm name] helper."
• Welcome prompt: the first message the client sees when they open the chat. Default: "Hi, I'm your portal helper. Ask me what's pending, when your next deadline is, or where to find a file."
• Off-limits topics: free text describing what you don't want the helper to discuss. The model is instructed to defer to your team on these.
• Custom FAQ: pairs of (question, answer) the helper uses as canned responses. Useful for firm-specific things ("What does ClientNest365 cost?" → your standard pricing answer).

What clients are asking

After 90 days of production data, the most common queries we see (anonymised, aggregated):

1. "What's pending from me?" (~28% of queries)
2. "When is my next deadline?" (~14%)
3. "How do I upload a file?" (~11%)
4. "What's the status of [matter/engagement]?" (~9%)
5. "Where can I see my invoices?" (~6%)
6. "How do I message my [advisor/lawyer/designer]?" (~5%)
7. Other / open-ended (~27%)

The "other" bucket is mostly clients exploring what the helper can do. Most of those queries get a brief answer pointing to the portal section that handles the request.

What this looks like in practice

The accounting walkthrough shows the helper in the context of tax-season document collection. The agency walkthrough shows it for design-engagement clients. Both have real screenshots of the helper interface.