An agent that qualifies and routes every inbound lead

It starts with your ideal customer profile, written down as criteria a machine can check. In the first stage of every engagement, Map, we sit with your sales lead and split it into 2 lists.

Hard rules are facts a lead meets or fails: a country you serve, a minimum headcount, an industry you do not sell to. These run in plain code before any model is called, because code is cheaper and gives the same answer every time.

A language model is worth using on the soft signals. "We are replacing our current vendor in Q1" reads differently from "send me a price list", and a Head of Operations at a 200-person firm reads differently from a student. The model reads the message, the enriched company record and the page the lead came from, then returns a score, a tier (A, B or C) and 2 or 3 reasons in plain English. The reasons appear in the CRM and in Slack, so a rep can disagree in seconds.

Before launch we score 200 to 500 of your past leads where the outcome is known. If the A tier misses deals you actually won, we rewrite the criteria before touching the threshold. The same test runs whenever the scoring prompt or the model changes, which is LLM evaluation in its simplest useful form. Read how we measure automation results for the method.

The build is a fixed price between $12,000 and $25,000. The low end covers one CRM, up to 3 lead sources and one routing table. The top end covers Salesforce territory rules, chat transcripts, several brands or languages and a backfill that scores your open leads. Most builds take 4 to 6 weeks, with a demo of working software every week.

That is the same price band as the AI tier of sales and CRM automation, which runs 5 to 8 weeks because it also covers pipeline hygiene, handoffs and a dashboard. This blueprint is the lead intake part on its own. It is a workflow with one model step, so it costs less than a custom AI agent, which starts at $15,000 and takes 6 to 12 weeks.

Running costs sit on your own accounts and are billed at cost:

At 1,000 leads a month the model line is tens of dollars. Enrichment is where budgets get surprised, so the workflow only enriches leads that pass the hard rules. That also keeps personal data to a minimum, which matters if you sell into the EU or UK, where enrichment needs a lawful basis such as legitimate interests (ICO guidance (opens in a new tab)). Changes after handover are $175 an hour, or covered by an automation care plan.

As a rule of thumb, below about 40 inbound leads a week a person reads them faster than a build pays back. A HubSpot or Pipedrive workflow with round robin assignment and a template reply does most of the job, and as a workflow integration it costs $1,500 to $6,000.

The same applies when leads already arrive structured. If your demo form makes headcount, country and use case required dropdowns, rules can score it and a model adds little. We will say so in the scoping call and build the rule version, because it is cheaper to run.

Two more cases where we would wait. If nobody has agreed what a good lead looks like, the agent will apply a definition your sales team disputes. Agree the ICP first; it usually takes one working session with sales leadership. If CRM ownership is a mess, with duplicate accounts and departed reps still owning records, routing will be wrong on day one. Fix ownership and data first through sales and CRM automation, then add the agent.

Finally, if the real problem is that reps never follow up, better scoring will not change that. Look at the sales call follow-up blueprint or a plain SLA alert.