AI CRM: which sales tasks should you automate first?

Start with the task that keeps a new enquiry from going unnoticed or a promised follow-up from slipping. Learn when a simple rule is enough, where AI can help and how to test one workflow before expanding it.

AI CRM: which sales tasks should you automate first?

Start with the step your team most often misses: assigning new enquiries or following up when promised. If messages sit in a shared inbox because everyone assumes someone else will reply, start by recording each enquiry and assigning it to a named person. If the team replies promptly but forgets to check back after a call or a quote, turn agreed follow-up dates into internal tasks. Neither job needs AI when the input is clear. Save it for work that involves interpreting text, such as finding unanswered questions in a long email thread or preparing a reply someone will check.

Find where enquiries or follow-ups get missed

Take a small batch of recent enquiries and trace each one from arrival to the last recorded action. Include deals that went quiet, not just successful sales. At each point, ask what a colleague taking over could see: who was responsible, what the customer was waiting for and when someone was due to act. If the answer lives in a private inbox or in somebody’s memory, identify the specific detail that never made it into the shared record. If an enquiry never reached the CRM at all, fix how it is captured before adding follow-up reminders.

Look for a task with a clear trigger, a result you can check and an error you can undo. Creating a reminder when an agreed date is saved fits those criteria: you can check that it appears in the right person’s task list and correct it if the date changes. Automatically treating a brief reply as acceptance of a quote does not. A customer saying “That looks good” might be commenting on the proposal without agreeing to buy. Updating the deal stage on that basis could mislead anyone who works on the account next.

Before setting up a rule, agree on which details people should record and where their colleagues should look for them. The guide to what a CRM does and when a small team needs one can help you decide what belongs in the shared record.

Use rules for decisions you’ve already made

A rule is useful when the team can explain what should happen without interpreting a message. Enquiries from a dedicated service form might go to whoever is on intake duty. An agreed callback date might put a task on the assigned salesperson’s list. For each rule, decide what should happen when a required detail is missing. An enquiry without a clear owner should go to a named reviewer or a queue someone checks regularly, not sit unnoticed in a list of unassigned records.

AI can help when information arrives in prose instead of structured fields. It can turn a long email thread into a short briefing, identify questions that still need answers or suggest a reply based on call notes. Ask it to separate confirmed details from tentative plans and open questions. That makes the briefing easier to review, but it does not make it a reliable substitute for the original correspondence. If a detail will affect the price, timing or terms, check the source before relying on it.

During the first pilot, keep AI output internal until someone has reviewed it. The salesperson checks the facts, revises the draft as needed and sends the message. Changes to the sales stage should also stay with the salesperson when they depend on interpreting what the customer meant. A suggested update is not evidence that a deal has moved forward.

A three-person team tests one follow-up workflow

Imagine a small office plant supplier with an owner and two sales coordinators. They already log enquiries and assign each one to a coordinator. The trouble starts after the first call. One coordinator asks a prospective customer for office photos and agrees to check back on Thursday. She puts the date in her notebook. On Thursday she is out visiting a site, and her colleague can see only that the call took place. He has no way to tell what the customer agreed to send or whether anyone promised to follow up.

The team pilots a rule in one sales stage. After the first call, the coordinator records the next action and its due date. The rule creates one internal task for that coordinator. If the date is missing, the record goes to a review list so someone can check whether a date was agreed but not entered, or whether it still needs to be arranged. If the photos arrive early, the coordinator checks whether they are sufficient and then completes or reschedules the task. Changing the date should update the original task rather than create a second one. Closing the deal should clear any reminder that no longer applies. The team also agrees who checks due tasks when a colleague is away.

Once the reminders are working, the team tests an AI briefing on longer email threads. It should show what the customer has supplied, what is missing and what the coordinator agreed to do. In one draft reply, Tuesday appears as a confirmed visit date even though the customer merely asked whether Tuesday was possible. The coordinator checks the thread and corrects the sentence before sending it.

The team can run this pilot with the information it already records in the CRM. If it later connects its mailbox, it should test that integration separately. For instance, can a reply be linked to the right enquiry when the same company has two offices under discussion? The guide to keeping a useful customer history with Google tools and a CRM covers what to record and where colleagues can find it.

Use this checklist before switching on the pilot

Write down the pilot in enough detail that a colleague could run it while you are away. “Remind us about open deals” is too vague. Specify which records qualify, whose list the task appears on and what makes it go away. Start with one source of enquiries or one sales stage so that, if something goes wrong, you can work out which change caused it.

  • Define the trigger, the required fields and the action the workflow should take.
  • Name the person who reviews incomplete records, including those with no assigned salesperson or no agreed date.
  • Test a normal record, a missing field, a duplicate submission and a customer reply that arrives before the task falls due.
  • Check that rescheduling updates the existing task and that closing the deal clears it when appropriate.
  • Give any AI step only the information it needs, and check who can see both its source material and its outputs.
  • Agree who can pause the workflow and how the team will handle enquiries manually while it is paused.

Use sample records before relying on live enquiries. If the workflow depends on a form or mailbox connection, verify that the information reaches the correct record, does not create a duplicate and is visible only to the people who need it. Check where a failed action shows up and whether the person responsible for it knows where to look. A shared inbox with a named person checking it is easier to manage than a workflow that silently drops enquiries. Before considering a different CRM, see whether your current one can handle the pilot and its exceptions. If it cannot, use the written workflow to compare alternatives.

Measure missed follow-ups, not the number of tasks created

Before the pilot, record how many agreed follow-ups fell due in a typical working period and how many happened late or not at all. Compare those figures with a similar period after launch, taking workload and staff availability into account. Look at the reasons as well as the counts. A task that never appeared points to a problem with the rule or its inputs. A task that appeared while its owner was away may point to a gap in cover. Flag duplicate and unnecessary reminders, too: a fuller task list tells you little about whether customers heard back when promised.

For AI-assisted writing, compare the time needed to check the source and correct the draft with the time needed to write the message from scratch. Distinguish stylistic edits from errors a customer might rely on, such as an unconfirmed visit date presented as booked. If a short list of unanswered questions helps but a complete email takes too long to review, try keeping the list and dropping the draft. In the next trial, check whether the list actually helps coordinators prepare quotes faster and whether they still catch relevant questions the AI leaves out.