Admin & setup
Call summary tips and tricks
How to get more from Cradle's AI call summaries by tuning your prompt, plus what FYI AI on the Elite tier can do with the summaries afterwards.
Call summary tips and tricks
The point of a Cradle call summary is to give an accountant back the part of the day that used to go on post-call admin. Notes, time entries, follow-up tasks, a draft email to the client. Done well, you can hang up the phone, glance at the summary, and the next steps are already in front of you.
This article picks up where Writing a good call summary prompt leaves off. Read that one first for the fundamentals. What follows is the deeper layer: prompt patterns specific to accounting firms, and how the summary connects to FYI AI on the Elite tier so the admin runs itself.
Make calls that summarise well
A summary is only as good as the conversation it's reading. A few small habits make a noticeable difference:
- Say the client's name early in the call. Even if you both know who you're talking to, naming them in the first 30 seconds gives the AI an anchor.
- State what the call is about up front. "I'm calling about your 2025 income tax return" is enough. The AI uses that line to frame the rest of the summary.
- Recap decisions at the end. "So we've agreed I'll send the engagement letter today and you'll get back to me by Friday." The recap becomes the action-items section, almost verbatim.
- Mind the audio quality. The summary is generated from the call transcript, so anything that hurts transcription accuracy (poor microphone, background noise, weak signal) hurts the summary too. See What do I need to know about call recording with Cradle? for the wider picture.
Prompt patterns that make the summary do more
Four patterns we've seen accounting firms use to good effect. Each one answers a single question: what is the accountant no longer doing manually?
You can paste any of these snippets into your AI call summary prompt in the admin portal and adapt them. They stack, so a single prompt can do all four.
1. Lead every summary with the client name
The summary becomes scannable at a glance. Looking back through your call history, you can see who each call was about before you read a single word of the body.
Prompt snippet:
Begin every summary with the client's name (or company name) on the first line, prefixed with "Client:". If the call was to or from a public agency, vendor, or unknown number, write "Client: external" instead.
2. Tag calls to or from tax authorities
If you ever need to find every call you've had with IRD, HMRC, or the ATO in a given month (compliance audit, time recording, internal reporting), a flag at the top of the summary makes it a one-glance lookup.
Prompt snippet:
If this call was to or from Inland Revenue Department (NZ IRD), His Majesty's Revenue and Customs (UK HMRC), or the Australian Taxation Office (ATO), add a flag on the second line in the format "[IRD]", "[HMRC]", or "[ATO]" followed by a one-line topic.
3. Classify by tax type
Hand the AI a list of the tax types your firm actually handles, and every summary will tell you which ones came up. Useful for time-recording, year-end reconciliation, and management reporting.
Prompt snippet (edit the list to match what your firm covers):
If any of the following tax types are discussed, list them in a "Tax types:" line after the client name. Tax types to recognise: GST, VAT, income tax, PAYE, PAYG, FBT, provisional tax, ACC levy, BAS. Use the exact name from the list.
4. Flag and summarise advice
This is the high-value pattern. If you define what "advice" means for your firm, the AI can surface every moment in the call where advice was given and pull it into its own section of the summary. Useful for engagement-letter compliance, for billable-work justification, and for building the audit trail every accounting firm wishes was already there.
Prompt snippet:
Identify every moment in the call where the team member gave the client advice. For this firm, advice means [INSERT YOUR DEFINITION]. Summarise each piece of advice under a heading "Advice given", as a numbered list, with one bullet per piece of advice and a one-line follow-up suggestion under each.
The placeholder is the most important thing you'll write. Every firm draws the "advice" line in a different place: some include any verbal commentary on a return, others only count formal recommendations with a written follow-up. Be explicit. The clearer the definition, the better the AI's flagging.
Pulling action items into FYI with FYI AI (Elite tier)
If you're on FYI's Elite tier, the call summary stops being read-only. FYI AI can read the note Cradle writes against the client record and act on it.
The setup assumption is that your Cradle is already connected to FYI per Setting up the Cradle integration with FYI, and that the call summary is landing on the client in FYI as a note (see Benefits of the Cradle and FYI integration for the integration in general).
With FYI AI configured, the same note can drive:
- Tasks for the assigned team member. The action items you recapped at the end of the call become tasks against the client in FYI.
- Pre-filled timesheet entries. Based on the call duration and the advice your prompt flagged, FYI AI can stage a draft time entry for review.
- A draft follow-up email. If you said "I'll send you a summary," the email is already drafted by the time you're back at your desk.
- Other FYI AI workflows your firm has set up. FYI AI is configurable on the FYI side; whatever workflows you build there can read the Cradle note like any other source.
The upshot: you hang up the call. By the time you've put the headset down, FYI AI has the tasks lined up, the timesheet entry waiting for your sign-off, and the draft email ready to send. You review, hit send, and move on.
FYI AI is part of FYI's Elite tier. If you're on a lower tier, the rest of this article still applies (a sharper prompt is still worth the time), but the FYI-side automation isn't available. Check with FYI for the current tier details.
Iterate
Your prompt isn't set-and-forget. Spot-check a handful of summaries each week. If the AI is missing things, tighten the prompt with more explicit instructions. If it's making things up or over-reaching on the "advice" pattern, tighten the prompt the other way (more specific definition, narrower scope). The Writing a good call summary prompt article has more on this.
The whole exercise is worth doing precisely because it compounds: every minute you spend tuning the prompt is a minute saved on every call from then on.