Jan 12, 2022
Date
Elena Voss
Head of Product
Read time
Read time
One of the most common concerns teams raise when evaluating AI-powered automation is control. Specifically, the fear that automating support actions means losing oversight. What if the system processes a refund it should not have? What if it sends the wrong message to a high-value customer? These concerns are valid, and they shaped a core part of how we designed Dispatch.
The Engineering Challenge
Our answer is the Approval Gate: a programmable checkpoint that sits between the AI's decision and the execution of any action. It allows teams to define granular rules about what can auto-execute and what requires a human to approve before it goes through.
The engineering challenge was building this without introducing latency. If every automated action had to wait in a queue for approval, the speed advantage of AI would disappear. Customers would still be waiting, just for a different reason. We needed a system that could distinguish between low-risk and high-risk actions in real time, and only pause execution when it genuinely mattered.
The Policy Engine Approach
The approach we landed on uses a policy engine that evaluates each action against a set of configurable rules before execution. These rules can be based on:
Action type (refund, account change, escalation)
Monetary value (threshold-based holds)
Customer tier (enterprise vs. standard)
Account age (new accounts flagged for review)
Custom CRM fields (any attribute you track)
A $15 refund on a standard account might auto-execute. The same refund on a flagged account might require manager approval. The rules are yours to define.
The Approval Experience
When an action is held for approval, Dispatch routes the request to the appropriate person with full context attached: the original ticket, the customer profile, the proposed action, and the policy that triggered the hold. Approvers can accept, reject, or modify the action directly from the notification, without opening the ticket separately.
The Feedback Loop
We also built a feedback loop into the system. When an approver consistently approves a certain type of action, Dispatch surfaces this pattern in the Insights dashboard. This helps teams gradually expand their automation coverage with confidence, based on real data rather than guesswork.
For example, if a team lead approves 98% of refunds under $50 over a three-month period, Dispatch will flag this as a candidate for automatic execution. The team can then update the policy with a single click, knowing the decision is backed by their own approval history.
Speed and Control Are Not Opposites
The goal was never to automate everything blindly. It was to automate confidently, with the right safeguards at the right points.
Speed and control are not opposites. They just require thoughtful architecture. The Approval Gate is our proof that you can have both, without compromising either.
Read also
