The four automations a small team should build first

Start with the four copy-paste leaks: recaps nobody shares, inbox checking, invoice downloads, messages typed twice. The test, the order, what stays human.

Ivar André KnutsenAI systems and workflow automationUpdated September 9, 20269 min read
The four automations a small team should build first: cover

Build the four that leak every day and never show up on an invoice: the meeting recap that never gets shared, the inbox and Slack you check every few minutes, the invoice PDF you download by hand, and the message you type twice. Each has a repeatable input and a result someone can check. Build the smallest, most-used one first: dictation, then the recap, then the digest, and invoices last.

Why these four, and not the clever ones

A proposal assistant, a customer chatbot or a dashboard can all be useful first projects. These four are another starting point: they tackle work a small team can recognise, measure and check without changing how the whole company runs.

So I use a plainer test. A task is a good first automation when three things are true.

It has enough value to justify the work. Frequency helps, but a difficult monthly task can also be worth automating.

It has a clear output. Open one thing, move it to another, tidy it, send it. If the result depends on context, write down which context the system needs.

It has a practical check. Someone can tell whether the output is right, and the workflow knows when to ask for a decision.

Run a normal week through that test and the same four come out:

  • The recap that never gets shared. The notes exist, in a doc. The people who weren't in the room never see them, because sharing took four more clicks than anyone had.
  • The checking. Slack, then email, then Slack again. Not because anything arrived. Because something might have.
  • The invoice download. A PDF lands in the inbox or on a billing page. Someone downloads it, renames it, drags it into the folder. Every vendor, every month.
  • The message typed twice. You dictate it, paste it into a model to tighten it, copy it out, paste it again. Four hops for one Slack message.

Nobody invoices for any of these. That is exactly why they never get fixed.

I was a plumber before I built workflows, and the trade taught me this: find the leak before you buy a bigger pump.

What the checking actually costs

The four are not equal. Three cost minutes. The checking costs attention, and attention is the expensive one.

Microsoft's 2025 Work Trend Index found that employees using Microsoft 365 are interrupted every 2 minutes by a meeting, email or notification. The average worker gets 117 emails a day, most skimmed in under a minute, and 153 Teams messages every weekday. Swap Teams for Slack and the shape holds.

Gloria Mark at UC Irvine has measured screen attention for two decades. Her recent studies put the average time on any one screen at 47 seconds, down from two and a half minutes in 2004, and once you're pulled off a task it takes about 25 and a half minutes to get back, per this UC interview. People interrupt themselves as often as anything external does. The Slack tab you open "just to check" is a self-inflicted one.

So the digest is not about two minutes per check. It replaces two dozen interruptions with two scheduled reads. That's the biggest win on the list, and still not the first build.

The ladder I climb before I build anything

Before I build anything, I walk the task up five rungs. Most stop before the top.

Remove. Does this task need to exist? Example: the weekly status email nobody replies to. Stop sending it and see who notices.

Simplify. Same result, fewer steps. Example: give vendors one address, invoices@yourcompany, instead of five people's inboxes. Now there's one place to watch.

Use a tool you already pay for. Example: Google Meet's "Take notes for me" writes a notes doc after the meeting and can email it to everyone invited, on eligible Workspace editions. If your team reads email, that may be the whole recap fix. Slack's Workflow Builder, on paid plans, runs steps on a schedule or from a link, no code. And Wispr Flow, whose Pro plan is $15 a user a month, already strips filler words and punctuates as you speak, and its Command Mode rewrites selected text by voice.

Automate. Connect tools you have with a small scheduled or event-driven workflow. Example: a trigger at 09:00 and 14:00 that pulls Slack mentions and Gmail threads since the last run and summarises them per sender.

Build. Write something custom, because no tool has the shape you need. Example: dictation cleanup with a style guide made from your own past messages, one key press, a review window before it pastes.

Rung three is where I lose the most sales, and I'm fine with that. If the tool you already pay for does the job, use it, and tell me so I can stop quoting you.

The four side by side, and the order to build them

What triggers each one, what it hands back, and what stays yours. The last column is the honest one.

Automation Trigger What it hands back What stays human
Dictation cleanup A hotkey on the text you just dictated The tightened version in a review window, one more key to paste The review. Anything confidential. What you meant when the transcript didn't.
Meeting recap to Slack The recap doc landing in Drive A shortcut, a picker, one click to post to a channel or a person Reading it first. Choosing who sees it. Fixing what the transcript got wrong.
Twice-daily digest A schedule, 09:00 and 14:00 in your time zone Who reached out, what they want, what they're waiting on, with a link to each thread Replying. It never answers for you.
Invoice PDFs to Drive A PDF by email, or a scheduled pull from a billing tool Renamed by date, vendor and amount, filed in one Drive folder, duplicates skipped Checking it's real. Approving it. Paying it.

Read the last column twice. It's what makes the other three safe. Now the order.

Your team's order may be different. A focused AI assessment can identify which recurring task is worth addressing first, show the assumptions behind its value and turn it into a testable automation brief.

Dictation first, because it's the smallest and you feel it twenty times a day. One hotkey, one review window, one more key to paste. It's also where the style guide gets written, from twenty of your own past messages, and the recap and the digest reuse that guide. Build it in a day, use it by lunch. How I did it is in clean up your dictation with Wispr Flow.

The recap connects spoken notes to Slack. In the current Chimpy workflow, Jari dictates in Wispr Flow. The recap follows his preferred structure and tone, then is automatically pasted into his Slack account. Details in Wispr Flow recaps to Slack.

The digest third, because it's the biggest and needs tuning. It gives attention back, not just minutes. It goes third because it touches two inboxes and the summaries need a couple of real weeks before you trust them. One honest caveat: LinkedIn messages don't come through, because LinkedIn's official API doesn't open the inbox to ordinary developers, so the digest picks up LinkedIn's own email notifications instead. The full build is in the twice-daily Slack and email digest.

Invoices last, because a mistake here costs the most. It reads, renames and files. It never pays, never replies, never deletes. But the sender allowlist that keeps fake invoices out is better built from a month of real invoices than from a guess, so it goes last, and it's the easiest to check once it runs: open the folder on Friday and count. The build is in auto-save invoice PDFs to Google Drive.

What they're worth, estimated

These are estimates. Nothing has been measured yet, and I'll keep saying so until it has.

Recap to Slack: say four meetings a day and three minutes each of open, copy, find the channel, paste, tidy, send. That's 12 minutes a day, about 1 hour a week.

The digest: say you check Slack and email every 20 minutes across an 8-hour day. That's 24 checks at 2 minutes, 48 minutes a day. Two digest reads at 5 minutes cost 10 of those back. Call it 38 minutes a day, just over 3 hours a week, before counting recovery time.

Invoices: say 40 invoices a month and 3 minutes each to find, download, rename and upload. That's 2 hours a month, about half an hour a week.

Dictation: say 20 dictated messages a day and 45 seconds each of paste, wait, copy, paste. That's 15 minutes a day, 1.25 hours a week.

Total: 1 plus 3 plus a half plus 1.25, roughly 5 to 6 hours a week for one person, estimated. If your numbers are half of mine, it's 3. Run yours, not mine.

I'm building against that estimate right now, for a team lead at Chimpy, Europe's largest powerbank rental service, founded in Zürich in 2013. All four workflows, on Google Workspace, Slack and Wispr Flow. Not finished, not measured. After 30 days in use, counted numbers replace the estimates, and I'll publish the difference. The build is in the Chimpy case study, and the shape of the work is on the implementation page.

Decide where a person should review the result

The review point depends on what a mistake would cost. A draft summary can be checked before it is shared. A filing workflow can put uncertain matches into a review folder. A chatbot can answer routine questions using approved material and hand unusual requests to a person.

Commercial decisions deserve clear boundaries. Decide who can approve a discount, commit to a delivery date or authorize a payment. The system can prepare the information and suggest a next step without being given every permission in your business.

Frequency is one part of the calculation. A monthly task that takes a day may be more valuable to automate than a daily task that takes seconds. Count the effort, the review and the cost of errors before deciding.

Information handling matters too. Decide which sources the system can read, where outputs go and who can access them. Those requirements belong in the scope, alongside the feature you want built.

Do this this week: count the copy-paste minutes

My four are a guess about your team. A week of counting turns the guess into a fact.

Monday to Friday, one sheet of paper. Four headings: recaps, checks, invoices, retyping. Every time you do one, one stroke. Don't estimate the time yet. Just the stroke.

Friday afternoon, multiply. Say 3 minutes a recap, 2 a check, 3 an invoice, 45 seconds a retype. Add the four lines. That's your weekly number. It's probably bigger than you guessed on Monday, because nobody counts small things.

Then run the test on each column. Enough value? A clear output? A practical check? Start with the task that offers the best useful return for the effort. If the answer is a chatbot, a proposal assistant or something outside these four examples, that is worth discussing too. The examples help you find a starting point; they do not define the limits of the work.

Questions people ask

What should a small team automate first?

Four useful candidates are invoice filing, a Slack and email digest, meeting recaps and dictation cleanup. Choose the one with the clearest benefit and easiest checks. For this example the order is dictation, recaps, the digest and invoices; another team may start elsewhere.

How do I know if a task is safe to automate?

Ask what a correct result looks like, what happens if it is wrong, and who handles exceptions. Frequent, easy-to-check tasks are convenient pilots. Tasks involving judgment can also be useful, with the right information, permissions and review points.

Do I need custom code, or will the tools I already pay for do it?

Try the tools first. Google Meet can generate and share a notes doc on eligible Workspace editions, Slack's Workflow Builder runs scheduled steps on paid plans, and Wispr Flow already removes filler words. A custom build earns its place when you need a shape those features don't offer, like a channel picker in two clicks or a house style in one key press.

How much time do these four automations save?

My estimate for one person is roughly 5 to 6 hours a week, built from worked examples: say 4 meetings a day, a Slack and email check every 20 minutes, 40 invoices a month and 20 dictated messages a day. It's labeled an estimate because nothing has been measured yet. After 30 days in use, counted numbers replace it.

When does an automation need human review?

Review should match the consequence of a mistake. Pricing exceptions, payments and sensitive customer situations usually need clear approval boundaries. Routine customer answers or occasional document jobs can still be suitable when the scope and checks make sense.

Sources

  1. 1.Microsoft Work Trend Index 2025: Breaking down the infinite workday · Interrupted every 2 minutes; 117 emails a day; 153 Teams messages per weekday; survey of 31,000 knowledge workers, February to March 2025
  2. 2.University of California: Can't pay attention? You're not alone (Gloria Mark) · 47 seconds average on a screen, down from two and a half minutes in 2004; about 25 and a half minutes to return to an interrupted task; people self-interrupt as often as they are interrupted
  3. 3.Slack: Guide to Workflow Builder · Starts from a link, a schedule, a Slack action or a webhook; available on paid plans
  4. 4.Google Meet Help: Take notes for me · Notes doc generated after the meeting, saved to the organiser's Drive, shared by email with chosen invitees; eligible Workspace editions only
  5. 5.Wispr Flow: Pricing · Pro plan $15 a user a month monthly, $12 annual; unlimited dictations
  6. 6.Slack API: chat.postMessage · Posts to a channel ID, or to a user ID for a direct message; about one message per second per channel
  7. 7.Microsoft Learn: Getting access to LinkedIn APIs · Open permissions cover profile, email and posting; nothing reads the inbox without partner approval
  8. 8.Chimpy: About us · Europe's largest powerbank rental service, since 2013, more than 10 million rentals with reusable solar-charged powerbanks
Ivar André Knutsen

Written by Ivar André Knutsen

I build and run AI systems, internal tools and workflow automation. You work directly with me from the first conversation through implementation and support. About Ivar

Want this looked at in your business?

We look at where you want the business to go, what is slowing you down and where AI could make a useful difference. You get a clear recommendation: a tool to try, a focused automation, a broader system or a closer look at the process. Any build is scoped and quoted before work starts. Free, no obligation.

Single automations are quoted on the call: a setup fee plus a monthly retainer to run them. Full systems start at $4,500, fixed scope, fixed price.

Book a free 30-minute call