How it works
Each piece of work is shaped by the organisation and the task, so its length and steps vary. Most follow roughly the order below, and work can stop after any step.
-
A first conversation
The organisation describes the task by email. If the project may be able to help, a conversation follows, by video call or in person in London. It covers what the task involves, how often it comes up, who handles it, where it slows down, and anything already tried.
What is askedSomeone who knows how the task is done day to day.
The first conversation covers Video callorIn person, London- What the task involves
- How often it comes up
- Who handles it
- Where it slows down
- Anything already tried
-
Looking at how the task is done
Before any tool is discussed, the project looks at how the task is actually done: where the information comes from, who handles it, how long each part takes, where it waits and how often it is redone. If the tool is chosen first, it is hard to tell afterwards whether it helped.
It helps to see the work itself rather than a description of it. The most frustrating part of a task is not always the part that takes the most time. Work done in small pieces through the day is easy to overlook. Much of the effort often goes on coordination, such as chasing, checking and moving information between people and systems, and that needs a different kind of change from producing documents faster.
This is done through conversation and from examples the organisation chooses to share, such as sample documents or exported spreadsheets with personal information removed. Any access to the organisation's own systems is agreed case by case and kept to what the work needs.
What is askedExamples of the work, prepared by the organisation.
A task, as it is done now- Enquiry arrives by email
- Details copied into a spreadsheet
- Waits for someone to check it
- Reply written
- Missing details chased Often redone
Illustration, not a real organisation's task. -
Recommendations, and the decision
The organisation receives recommendations: whether AI or automation would help with the task, which approach would suit, what setting it up would involve, what it would cost to run, and what could go wrong. Leaving the task as it is, or making a simpler change that involves no AI, is always among the options considered.
The organisation decides what, if anything, to do next, and is free to act on the advice without the project's involvement.
Recommendations cover The organisation decides- Whether AI or automation would help
- Which approach would suit
- What setting it up would involve
- What it would cost to run
- What could go wrong
Always among the options
- Leaving the task as it is
- A simpler change that involves no AI
- AI or automation
-
Designing and setting it up
Where the organisation wants to go ahead, the project helps design the system and set it up. The work usually involves some of the following:
- an AI model used through its API, usually Anthropic's Claude;
- a workflow tool such as n8n, to connect the steps;
- a Notion or Google Drive database, to hold the records;
- GitHub, for workflows triggered from a code repository.
Tools are chosen to suit each case. Running costs are estimated before anything is set up. Everything is set up in accounts the organisation owns, and the organisation pays the providers directly.
Changes are made one at a time, so that each can be judged before the next begins.
What is askedOne person in the organisation who will look after the system once it is running.
Accounts, set up one at a time Owned by the organisation- AI modelClaude, through its APIIn the organisation's name
- Workflow tooln8nIn the organisation's name
- DatabaseNotion or Google DriveIn the organisation's name
- Code repositoryGitHubIn the organisation's name
Tools are chosen to suit each case. Running costs are estimated first, and paid by the organisation directly to the providers. -
Handing the system over
When the system is working, the organisation receives the account details, a short written explanation of what the system does and what it costs to run, and instructions for changing it or switching it off. From then on the organisation runs it. The project does not maintain or monitor systems after handover.
- Account details
- Written explanation
- Instructions
When AI is not the answer
The project recommends against using AI where it would not help. The main reasons are these.
The information is not in usable form
If it lives in people's heads, in inconsistent spreadsheets or in scanned documents that have never been organised, that needs sorting out first. An automation fed poor information produces poor results faster.
The task does not happen often enough
Something that comes up four times a month will rarely repay the work of setting up an automation and looking after it.
The delay is somewhere else
If work waits on one overloaded person, or on an approval that takes days, speeding up the steps before it only lengthens the queue.
A mistake would cost too much
Some work cannot tolerate a confident wrong answer: decisions about people's safety, regulated advice, or anything where an error would be found by someone outside the organisation. Either a person checks every result, and that time counts against any saving, or the task stays manual.
To ask about a piece of work, email