Skip to main content

Choosing the work

Which tasks should we use AI for?

Choose tasks where you can explain the required result, supply suitable information and check the output. Then decide which parts belong with a person, an AI assistant, existing software or automation. One piece of work may need several of these.

By CURN editorial team Published Updated

Start with a piece of work you recognise

Take a recent research note, spreadsheet update or set of PDFs. Describe what someone needed at the end and what happened between the request and the finished result. Include the time spent finding files, clarifying the request, checking the answer and moving it into the right place.

This gives you something concrete to improve. A request to summarise a PDF might conceal a different need: finding three facts for a decision, comparing several versions or updating a shared record. Those needs call for different methods.

Allocate the steps

Use the table as a starting point. A tool choice still needs a trial with the material your team is permitted to use.

Step in the workA sensible starting pointWhat to check
Interpret a request or decide what mattersThe person responsible for the resultThe question and the decision are clear.
Find or draft information from supplied textAn AI assistant, with reviewThe result matches the sources and keeps missing information visible.
Calculate totals or apply a fixed ruleA spreadsheet formula or existing softwareThe inputs and rule are correct.
Move approved information between systemsAn existing integration or simple automationPermissions, duplicates and failed transfers are handled.
Approve a conclusion or external communicationThe named reviewerSomeone takes responsibility for the final use.

Choose a trial you can judge

Pick work that causes enough effort or delay to matter, with an output someone can assess. A familiar task is useful because you already know what a good result looks like. If nobody can explain how to check it, settle that before relying on a new method.

  • Name the input and the output, including where the finished work belongs.
  • Confirm which files and tools the team may use.
  • Choose an ordinary example and an awkward one, such as an incomplete document.
  • Agree who will review the result and what would make it unacceptable.

A PDF example

Imagine a team comparing a handful of PDF reports. It needs a table of dates, stated figures and relevant passages for a meeting. A person selects the reports and defines the columns. An assistant produces a draft table with page references. The reviewer checks the entries against the PDFs and marks absent facts as missing.

A spreadsheet then calculates any required totals from checked figures. The person preparing the meeting decides what the comparison means. This keeps extraction, calculation and judgement visible as separate steps. A polished summary alone would not tell the team whether those steps had been done properly.

Measure the whole piece of work

Compare the new method with how the team currently works. Include preparation, checking, corrections and the final handover. A quick first draft is only useful if the complete task meets the standard with an acceptable amount of effort. Record errors as well as time, including what the tool missed and what the reviewer had to reconstruct.

Then make a plain decision: keep the method, change it, try another approach or stop. The useful result may be a shared instruction, a better spreadsheet or a clearer process. Only add a build or automation when the work gives you a reason.

Start with the problem.

Bring a decision, a stuck piece of work or something that may need building. We can work out whether CURN is the right fit.

Book a call

30-minute conversation

Choose a time that works.

Tuesdays to Thursdays, UK time.