Entflo · How we work

§ 01 · The question

How does this actually work? And how do I know it won't break in week three?

Most of the founders who land here have already bought an AI tool that didn't quite work. The pattern is usually the same.

Failure 01Three tools, one in use

The accountant with three AI tools.

An accounting firm buys three AI tools over a year. One for invoice classification, one for client emails, one for tax-letter drafts. Six months in, only one gets opened. The partner has spent more time copy-pasting between the three than the manual work they replaced. Best case, they save an hour a week.
Failure 02Bot ahead of CRM

The agency with the lead bot.

An agency wires up an AI lead-qualifier. It runs around the clock and qualifies cleanly. The CRM it hands to syncs once a day, and nobody asked it to. Hot leads still get re-typed by hand the next morning. The bot is fast. The handoff is the same as it was last year.
Failure 03Drafts in nobody's voice

The proposal writer that didn't help.

A firm builds an AI proposal writer. The drafts come out in a tone that doesn't sound like anyone at the firm. Partners rewrite from scratch before sending. The fastest part of the workflow ends up being the part that gets thrown away first.

None of these businesses had bad AI. They had a disconnected cycle. The fix is almost never the model. The work is the integration.

§ 02 · The trust ladder

Conservative on day one. Autonomous when earned.

Every AI build runs the same risk. It looks great in the demo and silently fails three months later. Our way of avoiding that is to never give the system more autonomy than it has earned. We run every engagement through three phases.

Phase 01Conservative defaults

Day 1 to about week four. Everything drafts.

The system drafts every output. Emails land in your Gmail as drafts. Quotes land in your CRM as drafts. Posts sit in a review queue. Your team reviews everything and decides what goes out. Nothing irreversible runs without a named human clicking Send. We watch what gets approved, what gets edited, what gets rejected, and the system learns from all of it.
Phase 02Earned autonomy

System by system. Quality-based, never time-based.

The automations that have earned trust start running on their own. Other automations stay in draft mode. New automations always start at Phase 01, even when older ones are already running independently. The move from draft to auto-run is quality-based, never time-based.
Phase 03Weekly digest

The system runs. You read the summary.

Once an automation has been running cleanly for long enough, it moves to weekly-digest mode. The system runs most of the cycle. You get a weekly summary of what ran, what got flagged, what needs your eye. Edge cases route to a human. Nothing important happens silently.

What happens if quality slips

If the system's quality drops at Phase 02 or Phase 03, that automation auto-reverts to Phase 01 the same hour.

We diagnose, fix, re-test before promoting it back. You hear from a named human inside 24 hours.

Some things stay human, every phase.

Pricing changes above a threshold you set per engagement.
Legal commitments and signed contracts.
Client termination and significant relationship decisions.
Money out, suppliers, refunds, above a threshold.
Anything publishing publicly under your name without prior approval.

§ 03 · The data position

Your data. Your account. Your keys.

Three things about how we handle your data, because this is usually the question the CFO asks before the audit gets booked.

01Per-client API keys

Every client gets their own account, their own keys, their own bill.

Every client gets their own Anthropic account, their own API keys, their own bill. Your usage is fully isolated. Your data never mixes with another client's. If you cancel, the keys stay with you.
02The model doesn't learn from you

Foundation model APIs. No training on your data, by default and by contract.

We use foundation model APIs, Anthropic mostly. They don't train on your data, by default and by contract. Most APIs route through US data centres, which is fine for SMBs under the Australian Privacy Act. For clients in regulated industries, healthcare and regulated finance, we can route through AWS Sydney instead. Same model, Australian residency.
03If you cancel, you keep everything

The code. The SOPs. The configuration. The keys.

You keep the code, the SOPs, the documentation, the configuration, and your own API keys with the model provider. The system runs in your cloud account, not ours. The day you cancel, nothing stops working. The architecture is yours.

The honest part

The system you have on the day you cancel keeps running. The system you have a year from now is the one you built with us.

The base is yours. The compounding takes the partnership.

§ 04 · The audit

A 30-minute call. A one-page diagnostic. Yours to keep, with or without us.

Top three opportunities. Ranked by cash and capacity.
The assumptions. So you can check the math.
Build order. What unlocks what.
A "build it without us" version. In case you would rather.

Book the audit ⏎

If there's nothing worth fixing, we'll tell you.