Workflow
My AI workflow
AI gets used ad hoc, so output quality is inconsistent and nothing compounds.
The outcome
AI sits inside defined tasks with saved prompts and a human check, so quality is predictable and time saved is real.
The steps
Pick tasks, not tools
List the recurring writing, summarising and classifying tasks in the business first.
Write the brief once
Each task gets a saved prompt with context, tone rules and an example of good output.
Draft with ChatGPT
Speed matters more than polish at this stage.
Edit with Claude
Rewrite into plain UK English and strip anything that reads like AI.
Automate the boring half
Make calls a model mid-scenario to classify enquiries or summarise long threads.
Human sign-off
Nothing customer-facing publishes or sends without a person reading it.
Where this usually goes wrong
- Publishing unedited output
- Buying four AI tools before using one properly
- No policy on what data may be pasted in
- Letting AI speak to customers unsupervised
Where my companies fit
Declared openly — these are businesses I benefit from.
- Keystone AI
Where AI touches customer communication for trades — missed calls, enquiry capture, follow-up — Keystone AI handles it as a managed system rather than a DIY prompt library.