The first question people ask about handing work to software is “what if it does something stupid?”. The honest answer: that depends on what you told it never to do. Before Datify runs anything for you, spend ten minutes writing the rules. Here are the five that matter, in the plain English you’d use with a new hire.
Rules are how you delegate
You already do this with people. Nobody briefs a new office manager with a flowchart — they say “never promise a same-day slot on a Friday” and trust them with everything else. Rules are the shape delegation actually takes: a small set of hard boundaries, and freedom inside them. Software should be briefed the same way.
The five to write in week one
1. Contact frequency
The single most protective rule you can set. It makes over-messaging — the classic automation failure — structurally impossible, across every campaign at once, including the ones you haven't thought of yet.
2. The discount ceiling
Win-back offers work, and they're also how margin quietly disappears. A ceiling means the system can still be generous on your behalf — just never more generous than you decided on a calm day.
3. Protected accounts
Your biggest client, the one in a complaint, your mother-in-law. Every business has contacts where only a human should ever appear. Name them once and they're stripped from every list at the policy layer — before a draft even exists.
4. Quiet hours
Trivial to state, easy to forget, and the difference between looking like a professional and looking like a bot. Weekends count too, unless your trade genuinely works them.
5. The spend ceiling
Every plan is priced before it runs anyway — this rule just decides which ones can start without you. Small jobs flow; anything big waits for a yes.
Written in English, enforced in code
This is the part that makes the exercise worth ten minutes: the rules aren’t a prompt the AI promises to keep in mind. They compile into the policy layer that filters every list and gates every send. A rule can’t be charmed, rushed or forgotten at 2am, because the model that plans the work never gets to vote on whether the rules apply.
A rule the model can break is a suggestion. Yours are enforced after the model has finished talking.
When a rule gets in the way
A good sign your rules are real: sometimes they block something you wanted. The plan will say so — “118 contacts removed by the 90-day rule” — and you decide, then and there, whether to loosen the rule or keep it. Either way it’s a decision you made once, in writing, rather than a judgement call the software improvised on your behalf.
Start with these five, run a fortnight, and add the sixth the first time you think “it shouldn’t have needed to ask me that”. That’s the whole method.
Ten minutes of rules buys you months of not checking. Frequency, discounts, protected names, quiet hours, a spend ceiling — written like you’d brief a person, enforced like a firewall.
Writes about what actually happens when you put an AI in charge of real business operations — including the parts that go wrong.