AI Guides › Step-by-step guides
By Nigel Guy · 9 min read
The usual morning ritual is six tabs: payments, bank, inbox, calendar, the sales spreadsheet, the ad account. You glance at each figure, feel informed, and close them without having decided anything, because a number on its own says nothing about whether it is good, bad or normal. Asking Claude to "summarise my business every morning" automates the same glance, only faster and with more confident wording.
The rule: write down which numbers matter and what would make each one worth acting on before you connect anything. The brief reads that card back to you; the connectors only fill it in.
| You need | Detail at time of writing |
|---|---|
| A paid Claude plan | Scheduled tasks and connecting apps to an artifact both need Pro, Max, Team or Enterprise. Pro is listed at $20 a month ($17 a month billed annually) and is charged in local currency; in the UK that has worked out at roughly £18 to £20 a month including VAT. Check the figure at checkout. |
| Claude on web or desktop, plus the mobile app | You build on web or desktop; voice mode is designed to work best on your phone. |
| Logins for each data source | You authorise each connector with your own account, and Claude can only see what that account can see. |
| One afternoon | Mostly spent on Step 1 and on testing, not on clicking. |
One change to know about: from 6 October 2026, new Cowork tasks on Pro and Max run in the cloud, and the "Only on your computer" option is being removed. Scheduled tasks already work only with your connectors and files saved to your Claude account, never a folder on your laptop. If your sales figures live in a spreadsheet on your desktop, move them somewhere a connector can reach first.
Open a blank note and make a table with five columns. Keep it to five to seven rows; a brief with fifteen numbers is a tab you have to read.
| Number | Source | Exact definition | Compared with | Act if |
|---|---|---|---|---|
| Yesterday's takings | Stripe | Gross successful charges, UK time, midnight to midnight | Same weekday last week | Down more than [X]% |
| Unpaid invoices over 30 days | QuickBooks | Count and £ total | Last Monday's figure | Total above £[X] |
| Today's meetings | Google Calendar | Count, first start time | None | Any clash or no gap before noon |
The "Act if" column is what turns checking into understanding. If you cannot write a threshold for a number, leave it off the card: you do not act on it, so you do not need to hear it every morning.
The dashboard is the version you look at when the spoken brief flags something. Turn on Cloud code execution and file creation in Settings > Capabilities first, because artifacts need it. Then paste this, replacing the bracketed parts with your number card and brand colours:
You are building a one-screen business dashboard for a small UK business owner who checks it before 8am.
My number card is below. Each row has a source, an exact definition, a comparison and an "act if" threshold:
[PASTE_NUMBER_CARD]
Build an artifact that pulls each number live from my connected apps when opened. For every row show: the current figure, the comparison figure, the change, and a plain flag ("Act" or "Fine") based on the threshold. Put "Act" rows at the top. Show the time each figure was fetched. Use £ and UK date formats. Style: [BRAND_COLOURS_OR_"plain"].
Rules:
- Use read-only tools only. Do not create, edit or send anything in any connected app.
- If a source returns an error or no data, show "Not available" for that row. Never estimate or fill a gap.
- If any definition on my card is ambiguous, ask me before building.
Before you finish, check every row against my card and confirm each figure on screen comes from the named source.
Fill in: your card from Step 1 and a colour preference. The first time the artifact uses a connected app, Claude lists the apps and tools it will touch and asks you to approve them; you can switch individual tools off there. Connector tools that require approval for every action are not available to artifacts, which suits a read-only dashboard.
This is the prompt the schedule will run. Run it manually in a normal conversation three mornings in a row before you schedule it.
You are preparing my spoken morning business brief. I will listen to it, not read it, so write for the ear.
Number card:
[PASTE_NUMBER_CARD]
Steps:
1. Fetch each number from its named source using read-only tools.
2. Compare it with the figure in the "Compared with" column.
3. Sort rows into "Needs a decision" (threshold crossed) and "Normal".
4. Write the brief: open with how many items need a decision, then each of those in one or two sentences with the figure, the comparison and the likely reason if the data shows one. Then cover the normal items in one short sentence in total.
Format: under 200 words, plain sentences, no tables, no bullet symbols, amounts rounded to the nearest pound.
Rules:
- If a source fails, say which one and that its number is missing. Do not guess or reuse yesterday's figure.
- Do not send, draft, edit or file anything.
- Do not give financial, tax or legal advice; describe what the numbers show.
Before finishing, check each stated figure against the fetched data and remove any sentence you cannot trace to a source.
Fill in: your number card. Tighten the wording until you would be happy hearing it unchanged.
There are two routes, depending on which version of Claude you see.
/schedule in any Cowork task.Each run becomes its own session, and runs happen even when your computer is asleep. The Scheduled page shows upcoming and past runs, and lets you pause, edit, run on demand or delete.
Voice mode is in beta on all plans and works on iOS, Android, desktop and web. In the merged Claude experience, voice works in any conversation, including one where Claude ran a task, so open the morning run on your phone and tap the sound wave icon next to the microphone. Ask: "Read me this morning's brief, then wait for questions." Voice mode can use your connected tools, so "why is Stripe down?" can get a fresh look rather than a guess.
If you are still on the split Chat and Cowork layout, voice mode is not available inside Cowork (dictation is). Read the run's output, or start a voice conversation in a normal chat and ask Claude to fetch the figures there. Use push-to-talk if you listen in the car or kitchen; hands-free mode struggles with background noise.