HomeServicesFor DevelopersDocsPricing
Sign InStart free

Introducing Flow v4.0: Proactive AI for Small Teams

What changed in Flow v4.0, why proactive suggestions matter, and how the new release supports faster decision-making.

Back to Blog

Availability note: this product note explains the operating model behind Flow v4.0. Specific capabilities can vary by account, connected services, and rollout stage. Confirm current behavior in your workspace or contact Dealsflow before planning a workflow around a particular feature.

Flow v4.0 is organized around a simple principle: when a system has relevant workspace context, it should help a team notice the next useful step without hiding the source information or the need for human judgment. Proactive assistance is not permission to act without limits. It should make pending work clearer and shorten the path from review to an approved action.

What is new in Flow v4.0

Why proactive AI matters

A team can have the information it needs and still miss the next action because the signal is spread across records, messages, and task lists. Proactive assistance can help by presenting a specific observation with its source and a bounded recommendation. A statement such as “Three opportunities have no next task” is more useful when it links to those records and explains the rule that marked them.

The design goal is not constant interruption. Suggestions should be relevant to a person’s role, easy to dismiss, and clear about whether they are informational, drafted, awaiting approval, or completed. Teams should be able to tune or pause a workflow that produces noise.

Test with a realistic example

Choose one recurring situation where a delayed response creates avoidable work. For example, an opportunity may reach a review stage without a named owner or next task. Define the source records, the condition that should be noticed, and the action that is safe to prepare. Then test complete, incomplete, duplicate, and outdated examples.

A useful result should show why the item was surfaced, link back to the source, and distinguish a recommendation from an action. If the account has insufficient information, Flow should ask for clarification or route the item for review rather than inventing the missing detail.

Add guardrails before enabling action

Common failure modes

How to evaluate the release

  1. Choose one workflow where a delayed next step matters.
  2. Identify the exact records and signal that should be inspected.
  3. Define a safe recommendation and the actions that require approval.
  4. Test the workflow with representative edge cases.
  5. Confirm that source context, status, ownership, and failures remain visible.
  6. Measure whether the workflow makes the intended outcome easier, not merely whether it produces more notifications.

To see the operating model in context, start with the AI business operating system overview and the AI workflow automation guide.