AI Lead Qualification & Inbox Automation

Five inboxes, one slow reply

A customer sends an Instagram message at 22:40. Another writes to the Facebook page, a third taps the WhatsApp button, a fourth opens the website chat, and a fifth messages on Telegram. Each one arrives in a different place, and each one waits until somebody has time to look. By the next morning the most interested buyer has often already spoken to someone else.

The visible cost is slow replies. A less visible one comes later, when nobody can say how many enquiries arrived last month, which channel produced them, or which deserved a sales call. People repeat the same qualifying questions, follow-up depends on memory, and the business keeps no record it can learn from.

We build the missing layer: one intake flow across your social and web messaging channels that replies within seconds, qualifies the enquiry against rules you approve, and passes it to a person when one is needed.

Who this suits

• Agencies and brokers where the first reply decides whether a lead stays. Real estate, travel, clinics, law firms, and training providers see the effect fastest.
• Service businesses with a shared social inbox and more messages than people to read them.
• Teams paying for ads or content that push traffic into direct messages, who want to know which spend produces qualified conversations.
• Companies that tried a website chatbot and found the real volume sits in WhatsApp and Instagram.

It is a poor fit where enquiries are already under control, where every conversation needs a specialist from the first word, or where the offer changes too often to describe in a knowledge base. We will say so if that is what we find.

What we deliver

Channel capture and storage

Each channel reaches us through its official API, a webhook, or a provider such as Twilio, and every message is written to a database with its channel, timestamp, contact details, and history attached. That record makes reporting possible and stops a returning customer from starting over a week later.

Reply generation

The assistant drafts and sends a first reply within seconds, using the conversation history, the stage the enquiry has reached, and your approved service or listing information. A retrieval layer sits over your own documents, so your team updates price lists, availability, and policy wording in one place and the replies follow from the current version.

The qualification model

Qualification is a small scoring model you can read. We agree the fields that decide whether a conversation deserves a sales call: what the person wants, the timeline they mention, a budget range if they volunteer one, their location, and how they found you. The assistant extracts those into structured columns, scores them against weights you set, and routes the result. A high score goes to a named owner with the summary attached, and a low score gets a useful answer plus a place in a nurture sequence. You set the weights and can change them as the pipeline changes, so the score stays readable.

Where a human takes over

Handover rules are written before launch and stored as configuration, not inside the prompt:

• Price negotiation, contract terms, and anything with legal or regulatory weight go to a person. The assistant acknowledges the question and gives a time for a reply.
• Messages mentioning a complaint, refund, cancellation, or an existing contract are escalated with a notification instead of answered.
• When the assistant cannot find an answer with enough confidence, it says a person will confirm, records the question, and notifies the inbox owner.
• High value enquiries, set by score or a named account list, are paused for approval: the draft goes to a manager, and only a person sends, edits, or discards it.
• A customer who asks for a human gets one, with the conversation history attached so they do not repeat themselves.

Every escalation writes to the same database, so slow handovers appear in the weekly numbers instead of disappearing into someone's private inbox.

What the reply may claim

Automated replies are statements your business makes, so their claims are limited by rule. The assistant works from an approved knowledge base that you sign off, and it states only what that material contains:

• No prices, discounts, delivery dates, or availability that a person has not approved and that the system cannot re-read from your own data.
• No promises about outcomes, approvals, eligibility, or timelines that depend on a third party.
• No invented names, references, case studies, or credentials.

When an answer is missing, the assistant says a person will confirm and creates the task. We keep these guardrails beside the workflow with test questions that must produce an escalation, and re-run that set before switching to a new model version.

How the engagement runs

Week one is discovery. We map the channels, read a sample of real conversations, and agree the qualifying fields, the escalation rules, and the claims the assistant may make. Week two builds the capture layer, the database, and the drafting flow against a staging inbox your team can test. Week three adds the timing rules, the approval path, the notifications, and the reporting view, then runs a supervised period with drafts reviewed before sending.

Integration usually decides the timeline, because the flow touches your CRM, your messaging providers, and your internal systems. If the content that fills these inboxes should also be scheduled across those channels, we handle the publishing side under AI trend and content automation.

What it costs

Work runs at our flat rate of $39 per hour, or as a fixed scope once we have seen the channels and the qualification rules. A build of this size typically falls inside our published typical order range of $1,800 to $3,900 across one to three weeks, covering discovery, two channels, the qualification model, the approval path, and reporting. Each extra channel, a second language, or a two-way CRM sync moves the figure and the timeline, and a connector for an in-house system is usually the largest single driver. Teams that want continuing capacity instead of a finished project use a monthly plan from $2,699; the pricing page explains both models and lists the current typical order figures. Ongoing knowledge base updates and monitoring usually continue on an hourly basis or through website maintenance and support.

Proof

Automated Sales Workflow is an n8n, PostgreSQL, and OpenAI system we built for a real estate agency that was answering inbound Instagram enquiries by hand. Messages are captured, stored with their history, answered from current listing and policy material, and followed up on a schedule, while managers approve sensitive replies before they leave. It proves the pattern on live social traffic, including the approval step that keeps commercial control with the agency.

SmartAIChats is our conversational platform, used by businesses to answer visitors and qualify them before a person is involved. It proves we can ship the conversation layer as a product with setup a non-technical team completes on its own.

Vasilkoff.info is our own estimator and chat application, and it proves the intake half: structured detail collected from an open question, answers grounded in domain knowledge, and a summary a person can act on. We run our own enquiries through it.

Related services

• AI development for the model integration, retrieval, and evaluation work behind the system
• API development and integration for connectors to your CRM, messaging providers, and internal systems
• Bilingual AI support bot when the assistant also has to serve existing customers and answer in more than one language
• Secure AI backend and API key protection for moving model calls and provider keys out of browsers and apps into controlled endpoints
• Website maintenance and support for the ongoing care of the site that feeds your inbox

Next step

Send us the channels you use and a week of real conversations with personal details removed. We will reply with the fields we would extract, the rules we would set, and a fixed scope with dates.

Contact us with a short description of where your enquiries arrive, or run the workflow through the Vasilkoff.info estimator for a first cost and timeline range.