Case studies / Revenue operations

How a revenue team stopped researching the same lead three times

Before Murmurator, an inbound lead was researched by whoever picked it up, with whichever AI tool they preferred, against whichever qualification criteria they remembered.

Every inbound lead now gets the same research from the same sources, scored against criteria written in the workflow, with routing decided by conditions rather than by a model.

Steps in the workflow

7

One calls a model. The rest fetch, check or route.

Where the qualification criteria live

The definition

Conditions on values, reviewable in a diff, changed with a version and a summary.

Leads a model routes on its own

None

The model describes the company. The workflow decides what happens next.

HubSpotHTTP APIsSlackPostgreSQL

This is an example scenario, not a customer. We wrote it to show how this kind of work is put together in Murmurator. The workflow is real and you can build it; the team, the people and the situation are invented. Nothing on this page reports a measured result, because we do not have one to report.

The problem

Qualification by whoever got there first

Inbound leads arrive with a company name and a work email. Turning that into "is this worth a call" means reading a website, working out roughly how big they are, and checking whether anyone here has spoken to them before. Three people were doing that three ways.

The same lead, once per rep

Ana her prompt → a chat app → reads the site →her sense of fit
Ben a browsing agent → its own tools → whatever it found →its own scoring
Chen a spreadsheet → manual search → checks the CRM →his notes, in his notes
  • The qualification criteria are described differently in every prompt
  • Nobody checks the CRM first, so existing customers get researched from scratch
  • A model is asked to both describe the company and decide what to do about it
  • Two leads with the same profile get routed differently depending on the day
  • The research is in a chat window, so the CRM record still says nothing

The workflow

Research once, decide with rules

The model's job is narrowed to the part that needs reading: what this company appears to do, and roughly at what scale. Whether that is a lead worth a call is a business rule, and business rules belong in the definition where the team can argue about them.

  • The CRM is checked before anything is researched, so existing accounts short-circuit
  • The model describes; it does not route, score against revenue targets or assign owners
  • Everything it found is written back to the CRM, so the next person starts where this run ended

1 step of 7 call a model. The rest are the definition doing exactly what it says.

Trigger Deterministic

A form submission arrives on a webhook, or a new contact appears in HubSpot.

Check what we already know Deterministic

The CRM and the product database are read first. A known customer, an open opportunity or an existing contact ends the run here with a note, before a single token is spent.

Gather public context Deterministic

The company's own site and a small set of named sources the team agreed on, fetched through an HTTP connection with an allow-list. The workflow chooses where to look, so every lead is researched against the same sources.

AI assessment AI

One model call reads what was fetched and answers in a fixed shape — what the company appears to sell, who it appears to sell to, rough size band, and which of the team's named segments it best matches, with the evidence for each.

Validate the output Deterministic

A schema with an enum for the segment. A model that invents a segment fails the step instead of writing an unroutable value into the CRM.

Route on the rules Deterministic

Segment, size band and region are read by conditions in the definition. The routing table is visible in a diff, changes with a version and a one-line summary, and is the same at 9am on Monday as at 5pm on Friday.

Write it back and notify Deterministic

The findings, the evidence and the segment go onto the CRM record, and the owning rep gets a message with the summary. The research exists outside the chat window that produced it.

A model reasons here The workflow decides, the same way every run

Why it is built this way

Why the model describes and the workflow decides

It is tempting to hand the model the whole job — "read this company and tell me if it is a good lead". That question has two halves, and only one of them is about reading.

AI

Working out what a company actually does

A homepage is marketing prose. Turning it into "sells payroll software to mid-market construction firms" is reading and inference, it varies wildly by company, and no amount of parsing produces it.

Deterministic

Deciding which segment counts as a priority

That is a commercial decision the business makes, revisits quarterly, and should be able to see. Written as a condition it appears in a diff. Written into a prompt it changes silently, and nobody finds out until the pipeline looks strange.

Deterministic

Checking whether we already know this company

This is a lookup with a correct answer. Doing it first also means the cheapest possible outcome — an existing customer — costs nothing, which a model-first design gets backwards.

Deterministic

Assigning an owner

Ownership follows territory and workload rules that already exist. Asking a model to apply them adds variance to something that is, by design, not supposed to vary.

The model is at its best describing an unfamiliar thing. It is at its worst quietly holding a business rule nobody can find.

Before and after

What actually changed

Before

Everyone's own AI process

  • Criteria described differently in each person's prompt
  • The CRM checked last, or not at all
  • One model call asked to both research and decide
  • Routing that depends on who is looking and when
  • Research that stays in a chat window and never reaches the record

What this would change

What the team gets out of it

The research itself is not obviously better than what a careful rep would produce with a good prompt. It is the same for every lead, it happens for every lead including the ones nobody would have got to, and it is visible afterwards.

  • Every inbound lead is researched, not just the ones that looked promising at a glance
  • Changing what counts as a priority segment is one edit with a diff and an author
  • The evidence behind a routing decision is attached to the record
  • A rep joining the team inherits the criteria instead of reverse-engineering them
  • Spend is bounded per run, and existing customers cost nothing to disqualify

No numbers here, on purpose. Time saved, error rates and volumes depend entirely on the team running the process, and we have not measured this one. When a customer has measured theirs and is willing to publish it, their figures will appear here with their name on them.

Turn an AI process into a workflow.

14-day free trial with $5 of built-in AI included. No card, no per-run fees, cancel anytime.