Agree record rules before adding AI.
What identifies an organisation? How should two contacts at the same group be represented? Who decides that a deal is actually closed? Automation makes these questions urgent because an incorrect choice can spread quickly. Start with field ownership, record identifiers and permitted status transitions.
AI may help read an enquiry, propose a category or summarise sales notes. Deciding which record to update needs verifiable matching rules. Similar company names do not justify merging organisations; a positive message does not authorise marking a deal as won. Keep interpretation and record changes separately reviewable.
Worked scenario: from inbound enquiry to assigned task.
Illustrative scenario: a distributor receives availability questions and technical requests. The workflow extracts the supplied details, searches existing contacts and prepares a task for the appropriate owner. Multiple possible matches enter a review queue, with the original enquiry retained for context.
Scroll the table to compare all columns.
| Step | Useful rule | Exception |
|---|---|---|
| Identify the contact | Use approved identifiers and existing records | Shared mailbox or duplicate contacts |
| Classify the enquiry | Propose a category supported by the original text | Several needs or insufficient information |
| Assign the task | Apply agreed territory, product and availability rules | Owner unavailable or no applicable rule |
| Update the stage | Write only an approved transition | A colleague has already changed the deal |
A successful connection is only the first check.
Technical discovery covers available objects, custom fields, permissions, associations and the actual subscription plan. HubSpot’s contact APIs, for example, distinguish properties and associations to other records. Connecting a CRM is therefore a set of data operations with separate requirements.
The workflow also needs to handle repeated requests, concurrent edits and partial failures. If a record is created but its confirmation is lost, an uncontrolled retry can create a duplicate. Recognise the original request and inspect the destination state before deciding which step to repeat.
What a first implementation can cover.
Map the data
Identify source systems, fields to read and write, precedence rules and test records. Make fields that must remain untouched explicit. For teams in several markets, agree language and territory rules rather than inferring them from a name.
Simulate historical enquiries
Compare proposed categories, matches and tasks with a commercial owner’s decisions. Include ambiguous contacts and duplicate requests alongside complete examples.
Release to a limited team
Start with one intake channel and one sales group. Record outcomes, corrections and source references, with a practical way to pause processing and handle queued work manually.
Measure changes in sales administration.
If salespeople interpret pipeline stages differently or required fields are routinely ignored, repair the process first. Compare native CRM functionality with a custom integration, including the future work needed when fields, ownership rules or sales processes change. A larger activity count is not itself a better commercial result.
- Elapsed time from receipt to a genuinely assigned task.
- Duplicate records, unnecessary tasks and updates subsequently corrected.
- Required fields still missing after processing.
- Enquiries waiting in the exception queue and their assigned owner.
A template to work from.
CRM data rules
Rules for each CRM object covering identification, editable fields, value sources and a conflict resolution path.
Download the Markdown templateSystem integration map
One connection record containing field mappings, permissions, duplicate handling and a reconciliation check.
Download the Markdown templateAutomation exception register
An exception queue with status, priority, ownership and closure evidence; repeated issues become inputs to workflow improvements.
Download the Markdown templatePractical questions
Can AI automatically email customers?
It can prepare drafts within an agreed scope. Sending is a separate action requiring defined recipients, conditions and exclusions. A CRM data project does not implicitly include campaigns or automatic outbound messages.
Can we keep our current CRM?
We assess the features available in the product and subscription you actually use. Documentation, authorised access and often a test environment are needed. Integration availability is a discovery outcome rather than an assumption.
References and method
Official example of CRM properties and associations to inspect; no universal compatibility or vendor partnership is implied.