The interesting question is not what the AI can do. It is what it may not.
An AI that can search your leads, draft a pitch and launch a campaign is an AI that can spend your money and message your market while you are in a meeting. Rechlio’s answer is a gate: the model can propose a consequential action, and only a person can execute one. This page explains exactly how that gate works, because a safety claim nobody can inspect is marketing.
Where AI genuinely helps
Three jobs, all of which are judgement-at-volume — the thing that is unbearable by hand and unreliable by rule.
Scoring and segmenting leads
You describe your company and what you sell once. Every lead is then scored against that description and segmented, so a list of five hundred arrives sorted rather than alphabetical. ₹2.00 per lead.
Drafting an opening line
Written for that specific business rather than a merge field. Treat it as a first draft — the model has a business listing to work from, not your relationship with the market.
Classifying replies
Interested, not now, not interested. ₹0.10 per reply. It decides the order you read things in; it does not answer for you.
The approval gate
Rechlio has a chat agent that can use tools. Some of those tools are harmless — reading a number, summarising a list. Seven of them are not, and those are classed as consequential: finding B2B leads, finding businesses without a website, scanning a document directory, creating an outreach campaign, changing a campaign’s state, configuring an integration, and sending a WhatsApp message — which is the one that debits your wallet and calls Meta.
When the agent asks for one of those, it does not get an effect. It gets a proposal row — a description of what it wants to do and the exact arguments it wants to do it with — and a human sees it. There is exactly one route in the product that can turn a proposal into the real thing.
The five conditions
All five must hold before anything executes. Each closes a specific way the gate could otherwise be walked around.
- It belongs to your workspace. No proposal from one tenant can be executed in another.
- You are the person it was proposed to. A colleague cannot approve an action the agent raised for you — and does not even see it.
- It is still pending. Approval is single-use. The status changes before execution, guarded so two simultaneous approvals cannot both run the action.
- It has not expired. Proposals live fifteen minutes. A stale approval is not consent.
- The arguments still match. Each proposal stores a SHA-256 hash of its payload. Approving a message to ten contacts cannot be redeemed for a message to ten thousand.
That last one is the condition that matters most and is most often missing elsewhere. Without it, “approve” means approving an intention rather than an act, and the difference between the cheap version and the expensive version of the same intention is a number nobody re-read.
What stays human
- Launching a campaign. Always, regardless of size.
- Sending a WhatsApp message that costs money.
- Changing integration credentials.
- Deciding whether a lead score is right. It is a sort order, not a verdict.
- The pitch. The draft is generated; the claim about your product is yours.
There is also a hard limit worth naming: the AI cannot make cold outreach lawful. Consent, content and timing are governed by Meta’s policies, TRAI’s rules and India’s data protection law, and no amount of automation changes who is responsible for them. See the acceptable use policy.
What the AI costs
| Scoring and segmenting one leadAgainst your company profile | ₹2.00 |
|---|---|
| Classifying one incoming replySmaller model, higher volume | ₹0.10 |
| Enriching one company record | ₹0.50 |
| Approving a proposalThe gate itself | ₹0 |
Charged per action from a wallet, on every plan including the free one. See pricing.
Questions
- Can the AI send messages without me?
- No. Sending a billed WhatsApp message, creating or launching a campaign, changing a campaign’s state, running a lead search and rewriting integration credentials are all classed as consequential. Asking for one produces a proposal, not an effect, and only a person can turn a proposal into the real thing.
- Why do proposals expire after fifteen minutes?
- Because a stale approval is not consent. If you walk away from a proposal and come back an hour later, the context that made it a good idea has gone — the list has changed, the conversation has moved on. Fifteen minutes is long enough to read carefully and short enough that nothing gets approved out of forgetfulness.
- Does the AI write my messages?
- It drafts them. Each lead gets a suggested opening written against that specific business rather than a mail merge, and it is explicitly a first draft — editing it is the expected workflow, not a sign something went wrong. The instructions it runs under tell it to stay close to what you actually wrote and not to invent claims about your product.
- What model is behind it?
- Claude, from Anthropic. Reply classification runs on a smaller, cheaper model than lead segmentation does, which is why one costs ₹0.10 and the other ₹2.00.