Skip to content

AI Use CasesAny business

Start with the work.

Practical examples of AI taking on everyday work, sorted by the part of the business it sits in, including the ones where we would tell you a plain workflow does the job better.

Certified in the Claude Partner Network Microsoft Solutions Partner ISO 27001 ISO 9001 Independent since 1999

Pick a part of the business Talk through your own candidates

Potential AI candidates

Pick the part of the business you are thinking about

Every candidate names the work as it happens today, what changes, what stays with a person and what has to be in place first.

Part of the business

Verdict

12 candidates, across 6 parts of the business.

Client service and support

Drafting replies to routine questions

AI candidate

A first draft in your own wording, with the account details filled in, so the work becomes checking and sending.

How it works today, and what stays human
The work today
The same twenty questions arrive every week, and each answer gets written from scratch by whoever picks it up.
Stays human
Anything involving a complaint, a credit, a contractual position or a client who is already unhappy stays with a person from the first word.

Routing incoming requests to the right team

Plain workflow wins

Nothing, usually. If the categories are stable and the rules are known, a filter and a form do this more reliably and more cheaply than a model will.

How it works today, and what stays human
The work today
A shared inbox gets triaged by hand every morning.
Stays human
All of it. A rule does the sorting, and a person handles anything the rule cannot place.
In place first
Nothing. If someone quotes you an AI project for this, ask what a rules-based filter would cost.

Sales, bids and proposals

First drafts of repeated bid sections

AI candidate

A first draft from your own approved wording, cut to the word count, leaving the bid team the parts that are specific to the client.

How it works today, and what stays human
The work today
Every tender asks for the same company background, methodology and quality sections, rewritten each time to a new word count.
Stays human
Every claim, every figure, every reference and the win theme itself. A bid is a legal document.

Answering client assurance questionnaires

AI candidate

Draft answers from your own documented positions, with anything you have no position on flagged rather than guessed.

How it works today, and what stays human
The work today
A client sends a security or AI questionnaire and it lands on whoever has capacity.
Stays human
A person checks every answer before it is sent. An assurance questionnaire is a representation about your business.
In place first
Documented positions to draw on. This is one of the places where the policy work gets used a second time.

Marketing and communications

Writing in your company voice

AI candidate

Your vocabulary, tone and limits are built into the tool so that drafts arrive ready to use.

How it works today, and what stays human
The work today
Copy comes back sounding like a press release, so it gets rewritten by the one person who knows how the business talks.
Stays human
Anything published in your name and every claim in it.
In place first
A company voice guide. This is built during Adopt, the phase that brings your first team into daily use, so it is the one candidate on this page you already have if you have been through Track.

Choosing which campaign to run next

Plain workflow wins

Nothing worth paying for. If the data is thin, a model will produce confident output from thin data, which is worse than an argument between people who know the market.

How it works today, and what stays human
The work today
A decision made in a monthly meeting from partial data.
Stays human
All of it.
In place first
Nothing. Fix the measurement first, because a decision built on data nobody trusts cannot be defended when someone asks how it was made.

Finance and administration

Pulling the story out of a month-end pack

AI candidate

A first draft of the commentary against the numbers you supply, in the format your board already reads.

How it works today, and what stays human
The work today
Someone writes the variance commentary from scratch under time pressure.
Stays human
Every number, every explanation of why it moved and anything that will be presented as fact to a board or a lender.

Checking supplier documents against what was agreed

AI candidate

Differences between the document and the agreed position get surfaced for a person to look at.

How it works today, and what stays human
The work today
Invoices and renewal notices get compared to contracts by hand whenever someone finds the time.
Stays human
The decision, the dispute and the payment always stay with a person.

Operations and delivery

Writing up a process that only lives in someone's head

AI candidate

A structured draft of the process, built from an interview and from the records the work leaves behind, which the same person then corrects rather than authors.

How it works today, and what stays human
The work today
The person who knows how something works has never had time to document it, and is the single point of failure.
Stays human
The correction and the sign-off that it is right. Because people follow it, a wrong process document is more dangerous than none.
In place first
Very little. This is often the safest first candidate in a business, because the input is knowledge rather than client data.

Finding the pieces of work worth automating at all

AI candidate

This is what Discover, the first stage of Build, is for. Candidates get scored on how often the work happens and how much decision-making it needs, and the ranked list is the deliverable.

How it works today, and what stays human
The work today
Without a ranked list, the idea with the most backing tends to get built first.
Stays human
We agree the ranking with you against your own priorities.
In place first
Nothing. This is usually where a business should start once the rules exist.

People and internal

Answering the questions your team asks most

AI candidate

Your people get answers drawn from your own documented policies, and the tool flags any question those policies do not cover.

How it works today, and what stays human
The work today
One person answers the same policy and process questions one at a time.
Stays human
Anything at all about an individual's employment, pay, performance or wellbeing.

Deciding who to promote or how to performance-manage

Plain workflow wins

Nothing. We would tell you this is not a candidate. The input is people, the stakes are legal and personal and the output would carry a confidence it has not earned.

How it works today, and what stays human
The work today
A decision made by managers who know the people involved.
Stays human
All of it stays with a person permanently.
In place first
Nothing. We would advise against building it at all.

Nothing on this page matches both of those. Try another part of the business or another verdict, or tell us what you had in mind.

By sector

A longer list for your sector

Each sector page carries a longer list, written in the language of that sector.

Criteria

Which AI to buy is the second question

The first is which pieces of your work are worth handing over and what has to stay with a person. These four tests answer it, and they are the ones we would apply ourselves.

  • It happens often Work that turns up every week is worth attention. Work that happens twice a year is almost never worth building for.
  • The inputs are already written down If the knowledge only exists in someone's head, capturing it comes first, and that is often a good candidate in its own right.
  • Someone can tell quickly whether it is right If nobody can spot a wrong answer in under a minute, there is no way to review the output or to trust it.
  • A person still owns what matters Commitments, figures presented as fact, anything about someone's job or pay and anywhere a confident wrong answer costs more than a slow right one.

Sometimes the answer is a checklist, a form or a fixed rule, and we will say so. You are not charged for building something we do not think earns its place. Nothing here gets built before the rules exist, so Govern, where your leadership agrees them, comes first, then a first team, then a permissions review.

When a plain workflow beats AI → How review points get set → See how the route works →

Common questions

About this list

Are these things you have built?

These are categories of work, described as categories on purpose, so you can test each one against your own.

What we can evidence is on the Proof page.

How do I know which of these is right for us?

You probably already suspect, and Discover exists to test that suspicion. Candidates get scored on how often the work happens and how much decision-making it needs. You agree on the ranked list before anything is built.

Could we just start with one of these and skip the rest?

You can start with one piece of work, and each build is scoped and quoted on its own so you can stop after any one. What you cannot sensibly skip is the rules, because what makes a build safe is who is connected and what they can already reach.

Will this replace people in these roles?

Yes, some jobs change. If a job is mostly moving information from one place to another, that is the part a build takes, and the person does more of the rest. We tell you which tasks it covers before we build it, so you can see what is left. People keep the review points, so nothing goes out without someone looking at it. If we think a role changes, you hear it from us.

My function is not on the list.

Then that is a gap in this page. Tell us what you had in mind. We will tell you whether it looks like a candidate, including when it does not.

Read next

Talk through your own candidates

Bring the three pieces of work you have in mind. We will tell you which is worth doing first, and which would be better served by a plain workflow.

Talk through your own candidates See how Discover ranks them