The short answer
Strip away the hype and AI automation is a simple idea. Ordinary automation follows rules: when a form is submitted, create a record and send an email. It works well when the input is tidy. AI adds the ability to cope with input that is not tidy, such as a rambling email, a voicemail or a scanned document. Put the two together and you can automate tasks that used to need a person to read something first.
Automation and AI are not the same thing
| Ordinary automation | AI step | |
|---|---|---|
| What it does | Follows fixed rules | Interprets, summarises, classifies or drafts |
| Input it needs | Structured: form fields, dates, amounts | Unstructured: free text, speech, documents |
| Predictability | The same result every time | Usually right, occasionally wrong |
| Typical cost | Lower | Higher, with usage fees |
| When it fails | It stops and tells you | It can be confidently mistaken |
That last row is the important one. A rule either runs or it does not. An AI step can produce a plausible answer that is wrong. So the design question is always the same: what happens when it is mistaken, and who would notice?
Eight practical examples
1. Sorting incoming enquiries
An AI step reads each new email or form message and tags it: new enquiry, existing customer, supplier, complaint or spam. The right person is notified. Nothing is deleted on the AI's say-so.
2. Drafting first replies
For common questions, the system drafts a reply using your own wording and information. A person reads it, edits if needed and sends it. The time saved is in the typing, not the judgement.
3. Summarising calls and meetings
A recorded call becomes a short summary with agreed actions, saved against the customer's record. Participants must know they are being recorded, and someone should check the summary before relying on it.
4. Turning voice notes into tasks
A site manager or salesperson records a voice note between appointments. The system transcribes it, pulls out the tasks and files them. This is one of the most popular uses with teams who are rarely at a desk.
5. Extracting data from documents
Invoices, application forms and certificates are read and the key fields pulled into your system for a person to confirm. Accuracy varies with the quality of the document, so the confirmation step is not optional.
6. Answering routine website questions
An assistant on the website answers factual questions from your own published information, such as opening hours, service areas and how to book, and hands over to a person for anything else. We compare the options in AI chatbot vs human receptionist.
7. Qualifying leads before a call
The system asks a new enquirer two or three questions and summarises the answers, so whoever calls back is prepared. It should not reject anyone on its own.
8. Producing first drafts of routine documents
Proposals, reports and follow-up emails that follow a pattern can be drafted from the notes on record. A person owns the final version.
Where a person must stay in charge
- Anything involving money, contracts or legal commitments.
- Decisions about individual people, such as eligibility, credit or recruitment.
- Advice a customer would expect a qualified person to have given.
- Sensitive conversations: complaints, bereavement, debt.
- Any message sent in your name without review, until you have watched it perform for some time.
What you need before you start
AI works on the information you give it. If enquiries are scattered and records are incomplete, sort that first. The plain automations, such as capturing enquiries in one place and following them up, usually deliver more for less and create the clean data that makes AI useful later. Our list of processes to automate before hiring is the place to begin.
Check where your data goes. Know which services process it, what they keep and whether that fits your obligations to customers. Do not put confidential client information into tools you have not assessed.
What it costs
An AI step adds build time, testing and a usage cost that rises with volume. At Storm, automation on its own starts from £999 per workflow, and AI work is scoped on top of that depending on what it has to handle. Our guide to what business automation actually costs gives wider ranges, including when an AI project is not yet worth it.
How to judge a proposal
- Ask what happens when the AI is wrong, and who would notice.
- Ask to see it working on your own messy examples, not a tidy demo.
- Ask what it costs per month at your volume, not only to build.
- Ask which parts could be done without AI. Often that is most of it.
Common questions
Will AI replace staff?
In most established businesses it removes repetitive tasks, not roles. The realistic gain is that the same team handles more without the admin growing at the same rate.
Is it safe to use with customer data?
It can be, with the right tools and settings. It depends on which service is used and how it is configured, which is why this belongs in the scoping conversation.
How do we choose a first project?
Pick a task that is frequent, low risk if it goes wrong and currently needs someone to read something. If you would like a second opinion, a free 24-hour audit can include where automations would help most.
Want this applied to your website?
George or Nicholas will record a personal walkthrough of your site within 24 hours.
Not ready for an audit? See Automations and AI systems.
