Skip to content
be-human.ai
← Back to the journal

Business workflows

AI Lead Routing: Set Territory, Product and Deal Size Rules

Lead routing rules an AI agent can apply by territory, product and deal size, with a named sales owner for every exception and a clear audit trail.

Be Human typographic cover with the title: AI Lead Routing: Set Territory, Product and Deal Size Rules
Conceptual illustration created with AI · Be Human

AI lead routing means an agent reads each new lead, applies written rules for territory, product interest and deal size, and assigns the lead to the right sales owner in the CRM. The rules decide the outcome, not the model, so the work starts with a short set of conditions that a person has approved. Anything the rules cannot settle, or that conflicts with them, goes to a named owner who makes the call.

Key takeaways

  • Write routing rules as a short table of conditions approved by the sales lead, not as a loose instruction in a prompt.
  • Let the agent apply territory, product and deal size rules, and let it stop when two rules conflict or when none applies.
  • Give every exception type one named sales owner with a backup and a decision deadline.
  • Log the rule that fired for each assignment so a manager can audit it later.
  • Replay past leads and run the agent in a proposal mode before it writes assignments on its own.

Why routing breaks when the rules live in people's heads

Most sales teams route leads by habit. A senior seller takes the large accounts, a regional lead handles anything from their area, and a product specialist receives requests that mention a particular module. Those habits work until a form arrives from a company outside the usual region, asking about a product line the team sold recently. The lead then sits unassigned, or it lands with whoever happened to check the inbox first.

An agent does not fix this by guessing better. It fixes it by applying rules that someone wrote down, consistently, to every lead, and by sending anything the rules cannot settle to a person. That is a narrower job than making sales decisions, and it is the reason the approach can run every day without surprising the team.

Write the routing rules before you configure anything

Start with three dimensions because they cover most assignments: where the buyer is, what they want to buy, and how large the opportunity looks. Write each as a condition with one outcome. A territory rule might send companies headquartered in a given region to the regional owner. A product rule might send requests for a specific module to the specialist for that module. A deal size rule might send an estimated value above a threshold set by the owner to a senior seller.

Keep the table short. A rule that needs a footnote is usually two rules. Before any configuration, ask the sales lead to confirm each row, and ask what should happen to a lead that matches none of them. Many teams find that the qualification questions on their form decide whether routing works at all, which is why AI lead qualification without losing the human touch is worth reading alongside this method.

Use this decision guide for each incoming lead:

  • If one rule matches and its owner is available, assign the lead and record the rule name.
  • If two rules point to different owners, do not choose between them. Send the lead to the exception queue with both rules listed.
  • If no rule matches, such as a missing region or an unrecognized product, send the lead to the exception queue.
  • If the company already has an active owner in the CRM, keep that owner and flag the new inquiry for their review.
  • If the estimated value is missing or looks inconsistent with the form, hold the lead until a person confirms the value before any size rule applies.

A step-by-step plan for the first version

The first version should be small enough that the sales lead can read it in one sitting. The steps below follow the order in which mistakes are cheapest to catch.

1. Ask the people who assign leads today to describe the last few exceptions they handled and what they did with each one. 2. Write the rules table with one owner named for each outcome, and get the sales lead to approve it in writing. 3. List the fields the agent needs, such as region, product interest, estimated value, company name and existing owner. Mark the fields that are often blank in your forms. 4. Give the agent read access to the records it needs and write access only to the owner field and a short routing note. Keep the permissions narrow, following the principle of granting only what the task requires. 5. Replay past leads through the rules and compare each proposed assignment with what the team actually did. Investigate every mismatch before changing anything. 6. Run the agent in proposal mode for a period the owner chooses. The agent suggests an assignment, and a person accepts or rejects it. 7. Turn on automatic assignment only for rules whose proposals are consistently accepted. Keep exceptions, and any lead above the top threshold, on manual review.

Name a sales owner for every exception

An exception needs a person, not just a queue. A queue with no owner becomes a place where leads wait while everyone assumes someone else is watching. Name one sales owner for each exception type: unmatched region, conflicting rules, a dispute over an existing account, and a missing or implausible value. For each, record the primary owner, a backup, and the deadline for a decision.

The deadline should reflect how quickly a lead loses value in your business, and the sales owner sets it. The agent's job is to make the exception visible with the right context: the lead details, the rules that fired, the reason for the hold, and the deadline. The owner decides and the agent records the decision. The same owner, reason and deadline pattern applies to other AI workflows, as described in exception queues for AI workflows.

If a manual decision reveals a gap in the rules, the change should go back to the rules table and the sales lead should approve it. The agent should not quietly learn a new rule from one decision.

Illustrative scenario: a lead that matches three rules

Illustrative scenario: a software company receives an inbound form from a business headquartered in the northern region. The prospect asks about the analytics module and gives an estimated value above the senior seller threshold. The territory rule points to the northern owner, the product rule points to the analytics specialist, and the size rule points to a senior seller.

The agent does not pick one. It creates an exception record that lists all three rules, places the lead in the queue owned by the sales director, and sets a same-day decision deadline. The director assigns the lead to the northern owner with the analytics specialist as support. Operations then reviews whether a rule for this combination should exist, and the sales lead approves it before it goes live. The log keeps every step, so the next review can show why the lead moved and who decided.

What the agent may write and what it must never decide

The agent can read the lead, compare it with the rules, write the owner field and a short note naming the rule, and notify the assigned owner. It should not infer a region from an email signature it cannot verify, change a deal value, reassign an account that already has an active owner, or contact the prospect on its own. Those limits belong in the permission set, not only in the instructions, because instructions can be misread or overridden.

Lead forms are untrusted input. A free-text field can contain text written to look like instructions, so the agent should treat everything in a form as data to evaluate, never as a command. The OWASP guidance on prompt injection describes this risk and the controls that reduce it. Test these limits on purpose by sending unusual leads through the replay before the agent is trusted with assignments.

How Be Human's managed approach handles lead routing

Be Human begins with the routing table, the named sales owners and the exception records, because those define what the agent is allowed to do. A first engagement defines the one role, the CRM fields and tools it needs, the approval boundaries, and an observable success criterion, such as the share of leads that reach a named owner on the same day, measured against the sales team's own records. In managed operations, Be Human runs recurring evaluations on replayed leads, reviews agent permissions when the CRM changes, and handles incidents when an assignment goes wrong. Rule changes return to the sales owner for approval before they take effect. The free Be Human AI diagnostic can help you find the first task worth handing to an agent.

Frequently asked questions

Can an agent reassign a lead that already has an owner? It can flag the lead and propose a change, but it should not overwrite an active owner without approval. Existing account relationships are a common source of friction, so the rule for those cases should be written by the sales lead before the agent runs. The agent's note should explain what it found so the owner can decide quickly.

What should happen when two routing rules conflict? The agent should stop, record both rules, and send the lead to the exception queue for the named owner to decide. Choosing between conflicting rules is exactly the judgment that the sales owner should keep. Over time, repeated conflicts show where the rules table needs a clearer row.

How often should the routing rules be reviewed? Review them when the team changes, when a product line or region is added, and on a regular schedule set by the sales lead. A short review of the exception log usually shows which rules need attention. Changes should go through the same approval as the original rules.

What if the CRM data is incomplete? Incomplete data should lead to a hold, not a guess. The agent can list the missing fields in the exception record so the person handling the lead knows what to confirm. Improving the form fields is often the fastest way to reduce these holds.

Your next step

See where an AI agent fits in your company first.

Take the free Be Human AI diagnostic. A few questions about your work show which task to hand to an agent first, what it needs to know and where a person approves.

Be Human editorial guidance. Examples are illustrative unless explicitly attributed. Sources provide background; implementation choices depend on your workflow.

Less busywork. More room for you.

Bring us the work you wish you had time for.

Book a call to try a personalized AI agent free for 7 days ↗Or take the free AI diagnostic →