AI automation
Useful where it is dull. Nowhere else.
AI automation, done properly, takes repeated admin off a person's desk without touching the judgement calls that earn the money. TDM Advisory builds it for sorting enquiries, drafting replies, reading documents and chasing quotes, and always with a person reviewing anything a customer will see.
Where AI automation earns its place in a small business
The tasks worth automating share a shape: they happen most days, they follow the same rules every time, and getting one wrong is recoverable. Sorting an inbox qualifies, because an enquiry can be read, categorised and routed while the urgent ones are flagged rather than buried. Drafting a first reply qualifies, because you still read it and press send. Reading supplier invoices and job sheets into structured records qualifies, because the source document is right there to check against. Chasing an unanswered quote qualifies, as long as the chasing stops the moment somebody replies. Pricing a job, deciding whether to take on a customer, or handling a complaint does not qualify, and no automation here will try.
Sorting the inbox
New enquiries read, categorised and routed to the right person, with the urgent ones flagged rather than buried under newsletters.
First-draft replies
A reply written for you to edit and send. You always press send.
Reading documents
Supplier invoices and job sheets turned into structured records instead of retyped.
Chasing quietly
A polite follow-up if a quote has gone unanswered, stopping the moment the customer replies.
Summarising the week
What came in, what went out, what is still open, in one short digest instead of five reports nobody opens.
How an automation gets built and tested
Every automation goes through the same four stages, and it can be abandoned at any of them if the task turns out not to suit it. First the repeated task is identified, which means something that happens most days and currently sits with a person. Then the rules are written down in plain language, because vague rules produce vague automation. The build stays deliberately small: one task, drafting only, nothing sent without review, run alongside the existing way of working for a couple of weeks. Only once it is boringly reliable do the low-risk parts run unattended, and anything a customer will read keeps a person in the loop permanently.
- 01
Find the repeated task
Something that happens most days, follows the same shape every time, and currently sits with a person.
- 02
Write down the rules
What counts as urgent, what a good reply looks like, what must never be sent without a human. Vague rules produce vague automation.
- 03
Build it small
One task, drafted only, nothing sent automatically. You review the output for a couple of weeks.
- 04
Let it off the lead
Once it is boringly reliable, the low-risk parts run on their own. Anything customer-facing keeps a person in the loop.
Questions about automation
How do you decide what is worth automating?
- By how often a task happens, how long it takes and how much damage a mistake causes. Something done twice a day that takes ten minutes is worth looking at. Something done twice a year usually is not, however annoying it is. If the payback is not obvious within a few months, I will tell you rather than build it.
Does automation always mean AI?
- No, and often it should not. Most of the time the answer is connecting two tools you already pay for so a record stops being typed in twice. AI only earns its place where the work involves reading or writing something messy, such as sorting enquiries or pulling detail out of a document, and a person still checks it before a customer sees it.
Will anything reach a customer without me seeing it first?
- Not unless you decide it should. Automations start in draft only: replies are written for you to edit and send, and nothing goes out on its own. After a couple of weeks of you reviewing the output, the low-risk parts can run unattended. Anything customer-facing keeps a person in the loop by default.
What happens when an automation gets something wrong?
- The rules get rewritten. Vague rules produce vague automation, so most errors trace back to an instruction that was never pinned down: what counts as urgent, what a good reply looks like, what must never be sent without a human. Fixing the written rule fixes the behaviour, and the draft-only stage exists to surface those cases before they cost anything.
How long before an automation runs on its own?
- Until it is boringly reliable, which you judge rather than me. The sequence is the same each time: find the repeated task, write down the rules, build one small version that only drafts, then review its output for a period. Only when the drafts stop needing correction does anything come off the lead.
Last updated:
A free review and sample redesign of your website
Send your address and I will write up what is costing you enquiries, where you sit in search, and then design a sample version of your homepage so you can see the difference rather than read about it. No call required to receive it.