Orbit Lead-Intake Completeness Automation: No E-mail, No Phone, No Lead

A prospecting record with a company name, an industry, and a stated reason for interest reads as complete at a glance — every field a person skims looks populated with something real. The gap only shows up later, and specifically: when a nurture sequence has no e-mail to send to, or when the salesperson assigned the account has to admit they're calling a number with no idea who will pick up, because nothing forced a real, checkable contact method to exist before the record was allowed to exist at all.

Centriu's own daily prospecting routine researches real companies — industrial manufacturers, medical and aesthetic clinics, local commerce and services with a weak digital presence — and registers qualified leads directly into Orbit's own pipeline through a single database function every AI agent calls. On 2026-09-05, the owner's own review of that day's output found three companies registered with no e-mail address on file at all: a real gap between "found a company that looks promising" and "found a way to actually reach that company" that nothing in the tool had been checking for.

The fix rewrites the lead-creation function itself so the refusal happens before a single row is written, not after. A submitted e-mail is checked against a real address-shaped pattern (something@something.something) and rejected if it doesn't match; a submitted phone or WhatsApp number is stripped to digits only and rejected if it comes out shorter than ten characters — deliberately accepting either field as the qualifying phone number, since a single valid number commonly serves as both. Both checks have to pass before the function proceeds to look up or create the company, the contact, and the pipeline deal underneath it. Whichever check fails first returns a specific, actionable refusal — naming the exact places a person or an agent can actually go find the missing information (the company's own site, Google Maps, LinkedIn, Instagram) — rather than a bare validation error that explains nothing about what to do next.

The same day's second, closely related owner instruction went further: contact information has to be genuinely complete, not merely present. The tool's own contract text (the description every AI agent reads before calling it) and the daily prospecting agent's own standing instructions were both rewritten in the same migration to state the rule twice more, independently of the database check — so an agent reading its own instructions before ever calling the tool already knows what will be required, rather than discovering the rejection only after attempting the write.

A prospecting agent finds a manufacturer that fits the ideal profile — real signals of digital neglect, a plausible reason to reach out — but the only detail confirmed is the company name and city; no e-mail or phone turned up in the first search pass. Before the fix, that record enters the pipeline anyway, and it sits there until a person eventually notices there is no way to actually contact it. After the fix, the tool refuses the write immediately and names exactly where to look next — the company's own site, Google Maps, LinkedIn, Instagram — so the agent either completes the research in the same pass or the company is correctly left out of that day's count, rather than silently occupying a pipeline slot nobody can act on.

Every lead that reaches Orbit's pipeline now carries, by construction, at least one confirmed way to reach it by e-mail and one confirmed way to reach it by phone or WhatsApp — closing a gap where a real, same-day audit found three leads with no e-mail on file, and removing the possibility of a salesperson being assigned an account with genuinely no way to make contact.

This automation governs what data has to exist before a lead record is created at all — it does not verify that a submitted e-mail or phone number is currently correct or reachable, and it does not replace a person's own judgment about whether a given company is actually worth pursuing.

Telling an AI agent, in its own instructions, what a complete lead record should contain is real guidance — and it is also exactly the kind of instruction a model can misread, deprioritize under volume pressure, or simply forget across a long working session. Centriu's fix keeps that same instruction in the agent's own text AND enforces the identical rule at the one place nothing can route around: the database function itself, which refuses the write regardless of what any calling agent believed the rule to be that day.