Skip to content

checklist

CRM automation: what to wire up first

Irfan5 min read

What should you automate first in a CRM?

Automate in dependency order: capture, then hygiene, then routing, then follow-up, then reporting. Capture ensures every enquiry reaches the CRM at all. Hygiene ensures the records are trustworthy. Only then is routing and follow-up worth building, because automated sequences running on duplicated or incomplete records send the wrong message to the wrong person and do so at scale.

Most CRM automation projects start with follow-up sequences, because that is the visible win. It is also the point at which existing data problems become systematic, customer-visible and fast.

This is ordered by dependency. Each stage assumes the one before it works. Build in this order and each layer sits on something trustworthy; skip ahead and the automation faithfully amplifies whatever is already wrong.


Stage 1 — Capture

If an enquiry does not reach the CRM, nothing downstream can act on it. Your pipeline is a subset of reality and the reporting will not say so.

  • Every website form writes to the CRM, not only to an inbox
  • Phone enquiries are captured — logged during or immediately after the call
  • WhatsApp and messaging enquiries reach the CRM. In many markets this is the largest channel and the one most often left out
  • Email enquiries to shared inboxes are captured, not left to whoever reads them
  • Source is recorded on every record — channel, campaign, referrer
  • Capture failures alert someone. A form integration that breaks silently costs enquiries for weeks before anyone notices, and the CRM shows a quiet period rather than an error
  • Test each channel end to end, from a real device, on a schedule

The alerting item is the one skipped most often and the one that costs most. A broken form does not look like a bug — it looks like a slow month.


Stage 2 — Hygiene

Automation on untrustworthy records produces confident, systematic errors.

  • Deduplication at the point of entry, matched on a normalised email or phone — not cleaned up afterwards
  • Required fields enforced at creation, kept to the minimum that is genuinely needed
  • Consistent formatting for phone numbers, country codes, company names
  • Ownership assigned on every record. Unowned records are invisible work
  • Stale records identified rather than accumulating indefinitely
  • A defined merge process for the duplicates that get through
  • Opt-out status respected across every channel, and it must survive a merge

Duplicates are the most customer-visible CRM failure. The same person receives the same sequence twice, from two owners, and concludes you are disorganised — which is a reasonable conclusion.

Capture and hygiene are unglamorous and they determine whether everything after them works, which is why

the CRM automation work that starts here.


Stage 3 — Routing

Now that records are complete and trustworthy, they can be sent to the right place.

  • Assignment rules defined and written down — territory, service line, account size, whatever genuinely applies
  • A fallback owner for anything the rules do not match. Unassigned records disappear
  • Response-time expectation set per route, and measured
  • Escalation when a new enquiry is untouched past that threshold
  • Round-robin or load-balancing where routing is by capacity rather than fit
  • Out-of-hours handling defined, including what the enquirer is told
  • Reassignment on absence — holiday and sickness cannot strand pipeline

The escalation item is where automation earns most of its return. A lead that sits untouched for two days is usually lost, and nobody is watching for it manually.


Stage 4 — Follow-up

Only now. Sequences built before stages 1–3 send the wrong message to the wrong person at scale.

  • Immediate acknowledgement to the enquirer, honest about what happens next
  • Internal notification to the owner, with the enquiry content included
  • Follow-up sequence with a defined end. Sequences that never stop become spam and damage the sender reputation you need for the messages that matter
  • Sequences halt on reply — every channel, not just the one they were sent on
  • Stage transitions automated only from unambiguous events — proposal sent, contract signed, payment received
  • Judgement-based transitions left manual — "qualified" is a decision, not a form submission
  • Nothing automated is presented as personally written. Recipients notice, and the trust cost exceeds the time saved
  • A person can intervene at any point without breaking the automation

The seventh item is a judgement call worth stating plainly: an automated message that reads as individually typed is a small deception, and it is the kind prospects detect. Automated and useful beats automated and pretending.


Stage 5 — Reporting

Last, because reporting on unreliable data produces confident wrong decisions.

  • Source attribution survives from capture through to closed revenue
  • Pipeline stages reflect reality, not what is convenient to report
  • Response time measured from enquiry to first human contact
  • Conversion measured per stage, not only end to end — the leak has a location
  • Lost reasons captured in a closed list, not free text
  • Reporting reconciles with finance. If CRM revenue and accounting revenue disagree, one of them is being ignored, and it will be the CRM

What not to automate

Do not automateWhy
Personal-sounding outreachDetected, and it costs more trust than it saves time
Judgement-based stage changesProduces a tidy pipeline that no longer forecasts
Deletion of anythingArchive instead. Automated deletion has no undo
Sending on unverified dataAn automated message to a mistyped address is permanent
Anything with no failure alertSilent failure is worse than no automation

Before you start

Two conditions. Neither is technical, and both are more predictive of success than the platform choice.

Do reps actually use the CRM? If deals are managed in an inbox or a personal spreadsheet, automation cannot see them. Building workflows on partial adoption produces a system that appears to work and quietly covers a fraction of the business. That is an adoption problem, and automation makes it harder to detect rather than easier.

Is the current process worth encoding? Automating a stage model nobody believes in makes it permanent. As covered in the framework for deciding whether to automate at all, a bad process running faster is not an improvement — and the mapping that precedes this work is the step most failed automations skipped.

Working through this list?

Most teams get through the audit and stall on the remediation. That is the part we usually take on.

Key takeaways

  • Order is capture, hygiene, routing, follow-up, reporting. Skipping ahead multiplies existing data problems.
  • Every enquiry channel must write to the CRM, or your pipeline is a subset of reality.
  • Deduplication before sequences. Automated follow-up on duplicates is visible to the customer.
  • Automate stage transitions from evidence, not from someone remembering to update a field.
  • If a rep works outside the CRM, fix that before automating — automation cannot see what it does not contain.
FAQ

Questions about this

Anything that reads as personal but is not. Automated messages presented as individually written are noticed, and the damage exceeds the time saved. Also avoid automating stage changes that require judgement — a deal is not qualified because a form was submitted, and a pipeline that says otherwise stops being useful for forecasting.

Related services

More on AI Automation

TopicsCRMAutomationWorkflow automation
Chat with us