Most 'AI for business' pitches sell a future that isn't here yet. The AI that pays for itself today is unglamorous: it removes the repetitive work that quietly costs you sales. Start there.
Start with the repeatable layer, not the moonshot
Don't start by replacing a department. Start with the tasks done the same way every day — follow-ups, data entry, reporting. That's where an LLM plugged into your existing tools pays back fastest, and where you can actually measure the result.
Three places AI pays back first
- Follow-ups & recovery — abandoned quotes or bookings chased automatically
- Lead capture — a screenshot of a WhatsApp or DM lead turned into a clean CRM entry
- Reporting — the week's numbers compiled and emailed on schedule instead of by hand
Integrate, don't train
You don't need to train a model. The realistic path is wiring existing models (Claude, GPT) into the tools you already use — Stripe, Google, your CRM, WhatsApp. Training a foundation model is a different, far more expensive business, and almost never what an SME needs.
What to check before you spend
- Is the task actually repeatable and rule-ish?
- Do you have the data (even messy) it needs?
- Can you measure the before and after?
- Are third-party costs (API tokens) itemized?
- Who owns it when it breaks?
A realistic first project
Pick one painful, repeatable workflow. Automate it end to end, measure the time or revenue recovered, then reinvest into the next one. This is exactly how the automations running live on Marrakech Private were built — one system at a time, each anchored to a real result.
FAQ
- Do I need my own AI model?
- Almost certainly not. For an SME the value is in integrating existing models (Claude, GPT) into your current tools. Training a model from scratch is a separate, far costlier undertaking most businesses never need.
- Where should I start?
- With one repeatable, measurable task — follow-ups, lead capture or reporting. A narrow first project that pays back is worth more than an ambitious one that stalls.
- How are AI costs billed?
- Setup once, then optional maintenance, with third-party API/token usage itemized. Costs scale with usage, so a small automation stays small on the invoice.