Two models, two different promises
Lead providers promise volume: more interested people, more requests, more calls. A custom automation promises something else: that no request falls through the cracks. It doesn't bring new people; it gets more out of the ones you already have, including yesterday's customers.
They answer two different questions. The trouble starts when the first is used to answer the second.
The comparison, point by point
| Purchased leads | Custom automation | |
|---|---|---|
| What you get | Names and requests | A system that works on all your contacts, new and old |
| Who owns it | The provider: stop paying and the flow stops | You: workflows, data and logic stay in-house |
| Exclusivity | To be checked: the same lead may reach others too | Works only on your history and your customers |
| After first contact | Outside the scope of the service | The core of the job: reply, qualification, follow-up |
| Data and consent | Origin and consent to be checked case by case | You know what every step does; GDPR and AI Act built into the design |
| Over time | Value doesn't accumulate | The system improves every month: data, replies, templates |
Where leads really get lost
When I map a request's journey with a company, the gaps are almost always the same:
- Response time: the request arrives Friday evening, someone sees it on Monday.
- Quotes with no follow-up: the offer goes out, nobody calls back those who don't answer.
- Forgotten customers: those who bought two years ago never hear from you again.
- Scattered channels: email, WhatsApp, forms and phone end up in different places.
- Single-person dependency: if the person handling leads goes on holiday, the process stops.
More leads in a process that loses them just means losing more of them.
What "orchestrating AI tools" means
Many businesses already pay for two or three AI services, and everyone uses them separately. Orchestrating them means connecting them into a directed flow: a language model reads and classifies the request, the system drafts a first reply (which you approve), records the contact in one place, sets reminders and tells you every week who to call back.
I do it with n8n, an open-source tool that can even run on a small in-house server. It's the same approach I use for my own systems, such as the automatic social media publisher or the newsletter form that validates the email, adds it to the right list and alerts me on Telegram if something goes wrong.
When lead generation makes sense
To be fair: buying leads is not always wrong. If you're entering a new market, have no history or nobody knows you, an external flow of contacts can help. The problem is when it becomes the only answer to "we're not selling enough", before looking at what happens to the leads that already come in.
The sequence I recommend is simple: first fix the leaks, then, if needed, get a bigger bucket.
Where to start
Try answering three questions: how long between a request and your reply? Who follows up with those who don't answer a quote? When did you last write to last year's customers? If even one answer makes you uncomfortable, that's where the first piece of automation belongs.