How to prepare your AMS or CRM before adding AI automation

AI automation works better when the system already has clear stages and owners. If every opportunity is marked differently, automation will struggle to route work correctly. Start by standardizing lead status, quote status, renewal status, producer owner, and next action fields.

Pick a short list of statuses people will actually use. New, attempting contact, missing information, quoted, bound, lost, and not proceeding is enough for many shops. Twenty custom stages that mean the same thing will be ignored, and the automation will follow the mess.

Retire unofficial statuses that live in note text. If producers write “hot” in a note while the official status stays “new,” reminders will fire at the wrong time. The cleanup is a behavior change as much as a field change.

Print the official list at the desk or pin it in the AMS. When a CSR asks which status to use, the answer should be on the wall, not in a Slack argument. Consistency is the preparation AI actually needs.

Long notes are useful history, but they are not the same as an actionable task. A CRM should make the next step visible without forcing the team to read every note. AI can summarize history and suggest next actions, but the CRM still needs a place to store ownership, due dates, and status.

Create a next-action field or task that is required when status changes. “Call back about driver list by Thursday” can be completed. A three-paragraph call note cannot be assigned to a weekend duty producer.

When you add summarization, keep the original note. Summaries help scanning. They should not become the only record. A reviewer should be able to open the source if the summary looks wrong.

Require a next action when status changes so weekend duty producers are not asked to interpret a three-paragraph call note. “Call back about the driver list by Thursday” can be completed. History can stay in the note. AI can summarize that history for a reviewer. It should not replace the task field the rest of the office needs to see.

Before adding automation, decide where a record should live and which system owns each piece of information. This prevents duplicate records, conflicting statuses, and confused reporting. The cleaner the operating model, the more reliable the automation will feel to the team.

A common split is AMS for policies and documents, CRM for pipeline and follow-up. That only works if people know which status to trust for a new quote request. Write it down. If both systems have a “quoted” flag, pick one for automation triggers.

Stop dual entry where you can. If CSRs update the AMS and producers update a spreadsheet, AI will not reconcile that for you. Pick the system the first workflow will read, and make that the place people work.

Write which system owns status for a new quote request. If both AMS and CRM have a “quoted” flag, automation will fire twice or never. Duplicate records and conflicting stages are not an AI problem. They are a source-of-truth problem you should solve before any routing workflow goes live.

Every open record needs an owner. Unowned is a first-class problem, not a blank to ignore. Round-robin, by product, or by producer book are all valid rules if they are written and followed when someone is out.

Due dates should match the workflow. New web leads might be due the same business morning. Missing documents might be due two business days after the last request. Renewal document chases follow the cadence you chose. Dates that are all “today” train the team to ignore them.

Decide how duplicates are matched: email, phone, named insured, or a combination, with a human merge when it is unclear. Automation that creates a second record for an existing client will send the wrong follow-up and pollute reporting.

Unowned should be a first-class status the owner reviews daily during cleanup, not a blank to ignore. Due dates should match the workflow: same business morning for new web leads, a short window for missing documents, and the renewal cadence you already chose. Dates that are all “today” train the team to ignore every reminder, including the ones AI will later create.

Owners need a list of unowned new leads, overdue next actions, and renewals with no activity. If those lists disagree with what producers see, nobody will use the automation. Rebuild the report from the cleaned fields, not from a parallel spreadsheet.

Keep the report small. A trustworthy short list beats a dashboard with charts nobody can explain. When a status is wrong, fix the record, not the chart.

Do not ask AI to invent conversion rates from dirty data. If half the “quoted” records were never quoted, the math is theater. Cleanup first, then report.

Rebuild the owner list from the cleaned fields: unowned new leads, overdue next actions, and renewals with no activity. If that list disagrees with what producers see, nobody will trust the automation later. Fix the record when a status is wrong. Do not fix the chart. A short trustworthy list beats a dashboard of unexplained widgets.

Freeze new custom fields, new pipelines, and new marketing forms until the core statuses are stable. Adding fields during cleanup is how agencies never finish. Finish the short list, train the team, then consider extras.

Freeze unofficial tools that duplicate the CRM. Group chats used as a pipeline will undermine the cleanup the week after you finish it. If a chat must exist, it should link to the CRM record, not replace it.

You can still take new business during cleanup. You just should not launch routing automation until owners and statuses mean the same thing on Monday and on Friday.

Freeze new custom fields, extra pipelines, and unofficial chat-as-CRM habits until the short status list is in use. Adding fields during cleanup is how agencies never finish. Group chats that duplicate the pipeline will undo the week after you finish. If a chat must exist, it should link to the CRM record, not replace it.

Week one: agree on statuses, owners, and the source of truth. Map current values to the new short list. Close or re-own anything that has been sitting with no activity longer than the agency is willing to defend.

Week two: update the live records the first workflow will touch—new leads and near-term renewals—rather than boiling the ocean of old history. Train the team on the next-action habit. Run one daily stand-up against the new lists until people stop reverting to notes-only work.

Only then connect AI. Drafts, reminders, and summaries should read the cleaned fields. If you connect first, you will automate the old nicknames, the blank owners, and the duplicate phones, then blame the model for the routing.

If week two shows producers still living in spreadsheets, stop and finish the behavior change. An automation that reads the CRM while the real pipeline is in Excel will look broken even when the mapping is correct. The cleanup is done when the Monday list matches what people are actually working.