Automation arrives as a single proposition and is really a series of separate decisions, several of which should be answered no. Some parts of marketing improve when nobody touches them. Others degrade the moment they run unsupervised, and the damage is usually invisible for weeks. The useful exercise is deciding which is which before buying anything.
Automates well
Campaign construction
Audience building, keyword research, ad set structure. Mechanical, rule-governed, and the part where human effort produces the least differentiated result. Ideal for automation.
Creative production
Generating variants at volume. A person produces two; a tool produces twenty and you choose. The judgement stays yours, the labour does not.
Performance syncing
Pulling numbers from platforms on a schedule so the data is there when you look. Nobody should do this by hand.
Alerting
Telling you something happened — a lead arrived, spend spiked, an ad stopped delivering. Automation is strictly better than a human here because it never sleeps.
Automates badly
Constant campaign editing
Counter-intuitive but real. Every edit to a live ad set restarts the platform's evaluation of it. Continuous rewriting in the name of optimisation is interference, not improvement.
Publishing without review
AI writes the exact claims the CAP Code restricts. Auto-publishing removes the only compliance check you had, and the ASA holds you responsible, not the software.
Replying to customers
An automated reply that gets it wrong costs more than a slow human one. Acknowledge automatically; answer personally.
Deciding what to sell
Tools optimise toward what you tell them to want. If that target is wrong they will pursue it efficiently, which is worse than pursuing it slowly.
💡 On the first one: an ad set adjusted repeatedly over a single morning can stop delivering for the rest of the day while the same account serves thousands of impressions on another campaign, with nothing in the interface explaining it. "Continuous AI optimisation" as a selling point therefore deserves real scepticism — the useful version acts on evidence after a learning period, not daily.
Lead response is where automation pays most
For a UK service business the automation worth buying is rarely the campaign — it is what happens in the first ten minutes after an enquiry lands. Speed of reply decides conversion more reliably than targeting does, and the binding constraint for a small firm is cover: most enquiries arrive outside the hours anyone is at a desk. Automating acknowledgement and routing, while leaving the actual conversation to a person, is the change that shows up in revenue.
Three things worth automating here, in order of value:
- 1An alert that reaches you, not an inbox. A WhatsApp or push notification with the name, number and what they asked about, the moment it lands. You decide in seconds whether to stop what you are doing.
- 2Scoring, so you can triage. Rating each enquiry on declared budget, timeline, how they arrived and what they asked for lets you call the £4,000 job before the £40 one. The scoring must evaluate budget in your own currency — a threshold set in one currency and applied to another is wrong by whatever the exchange rate is, which is how foreign enquiries end up scored as worthless.
- 3An automatic acknowledgement so nobody is met with silence. Not an answer — an acknowledgement. There is a real difference and customers can tell.
If alerts are sent by email rather than messaging, watch deliverability rather than open rates. Automated notifications sent from a shared mailbox to a small internal team routinely land in spam after a few weeks, and the failure is silent — the alerts stop arriving and nobody notices until an enquiry goes unanswered for a day.
Consent and data, briefly
Not legal advice, but three things worth holding in mind as you automate. Lead forms collect personal data, so you need a lawful basis and a privacy notice the person can actually reach. Automated marketing messages to individuals have consent requirements that differ from messages to businesses. And a tool that stores your customer data somewhere you cannot see or export is a problem waiting for a subject access request.
The practical test: can you export everything you hold about a person, and delete it, without asking the vendor? If not, that is a risk you have taken on without noticing.
A sequence that works
First
Automate the alert. Cheapest to set up, highest return, no downside. Do this before anything else.
Second
Automate campaign construction and creative. The labour goes away, the approval stays with you.
Third
Automate reporting and performance syncing so you can look without assembling.
Last, carefully
Automated budget shifting between ad sets — and only with explicit bounds on how much can move and a floor it cannot go below.
What to ask before you let software act for you
The question that matters is narrower than ‘is it automated’. Ask what the software does on its own authority, and whether that can be switched off. Under UK consumer protection rules the trader remains responsible for what is published and promised, so a tool that writes claims, changes budgets or messages customers unsupervised is acting in your name and at your risk. A vendor who cannot answer that clearly is selling you exposure you have not priced.
Automate the build. Keep the approval.
Adyft automates the campaign build and alerts you the moment a lead arrives — and publishes nothing without your approval. Fourteen-day free trial.
Start free — no card needed