Enquiry ops: getting leads from first message into the CRM without losing them

Most B2B teams don’t have a lead generation problem. They have a lead handling problem.

An enquiry arrives — web form, WhatsApp, a reply to a rep’s email — and then it sits. Someone eventually reads it, pastes something into the CRM, and forgets to set a follow-up. Three weeks later nobody can say whether that lead was worked or dropped. The pipeline number is fiction, so nobody trusts the forecast, so people stop updating the CRM, which makes the number worse.

That is an operations problem, and it is fixable without buying anything with “AI” in the name.

Map the path before you change it

Write down what actually happens today, not what the process document says:

• Where do enquiries physically arrive? List every inbox — website form, shared email, WhatsApp, LinkedIn DMs, a rep’s personal phone.
• Who sees each one first, and how fast?
• What gets typed into the CRM, by whom, and when?
• What decides who owns the lead?
• What happens if nobody replies?

Nearly every team that does this finds at least one channel with no owner at all. That is usually the biggest single win available, and it costs nothing to fix.

Collect enough context on the first touch

A message that says “Hi, price?” costs five exchanges before anyone can help. Ask for just enough to make a useful first reply:

• Who is asking, and for which company
• What problem they’re trying to solve
• Rough size or scope
• Timeline
• Anything the team should know before replying

For a prefilled message link or a short form, that’s enough. Resist the long form — every field costs you enquiries, and you can qualify properly on the call.

Decide what “in the CRM” means

Most CRM hygiene arguments are really definition arguments. Settle these once, in writing:

• What qualifies as a lead worth creating a record for
• Which fields are mandatory at creation, and which can wait
• Who owns a record before it’s qualified
• What closes a record out, and what reason codes exist
• How a lead that goes quiet gets handled

A team that agrees on this and updates records by hand beats a team with a beautiful automation and no shared definitions. Automating an undefined process just produces wrong data faster.

Track the basics before automating

Before spending anything, measure for one month:

• Enquiries received, by channel
• Median time to first reply
• How many enquiries became qualified conversations
• How many records were created without a follow-up set
• The most common missing information

If the same failure repeats, automate that specific piece. If it doesn’t, keep it manual — that is a real answer, not a cop-out.

Manual, automated, or AI

Three different scopes, routinely confused:

Manual is enough when one inbox receives everything, volume is low, one person owns the reply, and there’s no routing logic.

Deterministic automation is worth it when enquiries must route to different owners, records need creating from a form or webhook, follow-up reminders need to fire reliably, or the same data is being retyped between systems. Most of what teams call “AI automation” is this — and it should be. It’s cheaper, faster, and it doesn’t hallucinate.

AI earns its place when the work needs judgment on unstructured input: summarizing a call into structured notes, matching a message to the right contact when the email doesn’t match, drafting a reply for a human to approve. That is a genuinely different capability, and it belongs behind a human check.

The mistake is pretending those three scopes are the same. They aren’t, and paying AI prices for deterministic plumbing is how budgets get wasted.

Build it so agents can reach it

One design rule worth adopting even if you never buy an AI system: make each step addressable.

An automation that only fires when a human clicks a button in a UI is a dead end. The same step exposed as a clean API, a CLI, or an MCP endpoint is something any agent you adopt later can call — without a rebuild. The plumbing you lay now decides how much optionality you have in two years.

When it’s worth quoting a build

When the mapping shows a repeated, costly failure with a clear owner and a measurable cost, that scoped build is Production Implementation — written scope, acceptance criteria, test plan, handover docs.

If you’d rather see the whole picture first, the AI Systems Diagnostic maps how your work actually runs and where AI pays, and where it doesn’t.

---

Dealing with this in production? The AI Systems Diagnostic is the starting point.