ai sme

A Copilot licence, or a built workflow?

AA
Aurum Avis Labs Author
8 min read

Microsoft Copilot for business: licence or built workflow?

Two purchases show up in the same conversation: seats for Copilot or ChatGPT Team, and a workflow that runs on its own every Monday. The licence helps with drafting inside Microsoft 365 or a company chat. The build takes over the glue between mail, a spreadsheet and your line-of-business tool. When the question changes every week, buy seats. When Monday looks like last Monday, you need a defined transfer more than another seat.

This is for a managing director, office lead or finance lead in a Swiss SME on Microsoft 365, often after someone said “get Copilot”. It is a choice between an AI licence and a fixed-price build, not a tool ranking.

QuestionLicence (Copilot or ChatGPT Team)Build (workflow)
What changes?Wording and draft on each requestInput, output and weekday stay the same
Where it runsIn apps the person already opensBetween systems, even when nobody is typing
Typical spendMonthly per seatOne-off fixed price after the workflow is defined
When to stopWhen transfer is the bottleneck and drafts are already good enoughWhen half the cases are exceptions or nobody owns the workflow
  1. Name the job

    Write down input, output and weekday.

  2. Check repetition

    Same columns and recipient? Lean towards a build.

  3. Data rule

    Which text may go into which company account?

  4. Pick the price shape

    Monthly seats or one package, then compare.

When the licence is enough

Microsoft Copilot in a company works inside the Microsoft 365 files that person can already open: Word, Outlook, Excel, Teams, and search across what their account is allowed to see. ChatGPT Team does the same kind of job in a company chat: a draft, a summary, a one-off question.

Sketch, not a client story. Someone in sales writes quotes. They paste the last customer mail, ask for a shorter reply, and send it after they have read it. Someone in operations needs a clause from a contract filed two years ago and would otherwise hunt through folders. The question is new each time. The file they touch is the one already on screen.

In that situation you do not need a build. Buy the licence, write the data rule, and stop. Customer files, named quotes, and payroll lists do not belong in a private Free chat. That rule is one sentence long here; the traffic light is in ChatGPT and customer data.

A licence is also the right spend when you are still unsure the work repeats. If you cannot name the input, the output, and the weekday, there is nothing to automate yet. Seats let people try. A project would freeze a process you have not described.

When a built workflow fits

Input and output stay the same every week. Several systems have to meet. The same column, the same recipient, the same approval, often the same person doing the copying because they are the only one who knows the order.

Second sketch. Every Monday, finance exports the CRM, opens two spreadsheets, pulls a PDF from a portal, and pastes figures into a workbook the management team reads at 10:00. Copilot in Excel can help clean the sheet that is already open. It does not log into the CRM, fetch the PDF, apply the same filters, and send the workbook without someone starting it. The figures sit in five places with five permission models. A chat window does not hold those logins, and it should not.

Invoice capture is the same shape when the gap is outside the accounting tool: PDFs in an inbox, a portal download, a check against a purchase order that lives somewhere else. The accounting software may already propose bookings. The leftover work is the glue. The filter for whether that glue is worth a project is in which process to automate first: it repeats, it hurts, it has a clear start and end, and the data is already in mail, a file, or a tool you use.

A built workflow is also the right shape for quote follow-up when the failure is rhythm, not wording. Copilot can draft the reminder. It does not keep the list of open quotes, wait seven days, and show who has gone quiet. If that list is the same columns every week, the list is the project. If every reminder is a judgement call about a relationship, leave it with a person and a licence.

ChatGPT Team versus a built workflow

ChatGPT Team and Copilot often get framed as rivals. For this decision they sit on the same side. Both are seats: a company account, a contract, an allow-list, people drafting inside it. Neither one is the Monday glue.

Pick the seat from the tools you already have. If the work happens in Outlook, Excel, and Teams, Copilot sits where the files are. If the work is a shared company chat and people are not in Microsoft 365 all day, ChatGPT Team can be the allow-listed place to draft. You can own both seats and still have the report done by hand. The licence answers “where may people type”. The workflow answers “what happens when nobody is typing”.

Neither seat replaces an integration service. Power Automate or a similar tool moves data between programs. How Make, n8n and Power Automate differ is in Make vs n8n vs Power Automate: pick the tool after the process.

The data rule does not flip with the vendor. A company plan is the place for anything that is not public. A private account on a phone is not. Details, including the Free versus company distinction, stay in the data-protection article linked above. This page adds no legal opinion.

What still needs a person

A workflow that moves files does not decide a customer complaint, a discount, or a figure you would not put in a management report without a look. Build the transfer. Leave the judgement with someone named.

If the real complaint is “the team has no shared rules, and client text already sits in private chats”, a flow will automate a habit you have not agreed. The next step there is a workshop, not a connector: workshop or implementation.

Skip a build when any of these is true:

  • You cannot point at one repeating input and one repeating output.
  • Half the cases are exceptions, and the exception is the actual job.
  • Nobody is allowed to say the workflow is finished. A model will not invent that owner.
  • The pain is policy (what may be pasted where), and the copy-paste only exists because the allow-list is missing.

Skip extra Copilot seats when the bottleneck is the transfer itself and people already draft well enough. More seats speed up the sentence. They do not create the Monday file.

What you pay for, roughly

A licence is a monthly cost per person. Microsoft and OpenAI set those prices by plan, and the numbers move, so this article does not print a seat price. Check the vendor page when you buy. What matters for the comparison is the shape: you pay it every month, for as many people as you seat, whether or not Monday gets shorter.

A bounded build is a one-off fixed price after the workflow is defined. Guide sizes on our offer run from a small package of a few days (about CHF 3,200 to 4,800) through larger packages. They are listed there as guide prices, not a quote. Hosting afterwards is a guide of CHF 50 to 150 a month and can be cancelled. Token or model cost in production sits with you.

The Process Check that picks the size is free. Those figures come from the offer, not from a survey of other firms. The final size is decided on your workflow, on the offer page, not by matching a blog row.

Compare the two spends only after you know which job you have. A licence for eight people, every month, for a year, can cost more than a small package and still leave the transfer manual. A package for a workflow that changes every week builds a pipe for a job that will not repeat.

If you are unsure which case you are

That is what the Process Check is for. Thirty minutes, one workflow, a written case, and a package size if a build makes sense. The half hour is allowed to end with “the licence is enough”. You then have that in writing, and you do not buy a package to be polite.

Bring the actual Monday: where the file starts, who touches it, where it lands, and how often. If you cannot fill that in, the Check will say so, and the right spend is still the licence plus a one-page data rule.

ai sme process automation process check
AA

Written by

Aurum Avis Labs

Builds and ships at Aurum Avis Labs. Writes here about what we learn working with founders and SMEs in the DACH region.

Customize your cookie preferences. Essential cookies cannot be disabled as they are required for the website to function properly.

Essential Cookies

Required for basic website functionality, security, user authentication, and error tracking.

Always active

Analytics Cookies

Help us understand how visitors interact with our website to improve user experience. Includes Google Analytics and Microsoft Clarity session recordings.

Marketing Cookies

Used to track visitors across websites to display relevant and engaging advertisements.