How far should automation go to be sufficient?

This is a question that Sinh Vũ often hears, often accompanied by two extremes: one side wants to automate everything to avoid hiring people, while the other is hesitant to engage. The reality lies in between and depends on the nature of each task, not on technology.

What type of tasks can be automated reasonably?

Automated assistants (AI agents, meaning programs that perform a series of actions according to predefined rules) work well when the tasks meet three conditions simultaneously: they are performed frequently, have clear rules, and the consequences of errors do not directly impact customers or the brand's reputation.

• Remind to approve content on schedule.
• Synchronizing design assets (logo, colors, fonts) across storage repositories.
• Classify customer requests and direct them to the right person for processing
• Draft the initial version for review, not for direct posting.
• Compile periodic reports from existing data.

The value of automation diminishes as the work requires more contextual judgment. This is not Sinh Vũ's perspective but a conclusion from the design operations research by Nielsen Norman Group.

The boundary that people must maintain

Three stages that humans must not let go of, no matter how good the system is:

• Brand voice: machines can mimic intonation, but cannot perceive whether the nuances are appropriate or not for each moment and audience.
• The accuracy of data: AI agents can be factually incorrect while appearing very confident. Content with data, legal commitments, or product information must be verified by a person.
• Strategic alignment: A grammatically correct message can still be wrong in timing, wrong in channel, or contradictory to the overall positioning. This is the judgment of someone who understands the big picture.

In addition to the three stages above, crisis communication management, pricing decisions, and commitments to customers must be finalized by an authorized person, not handled automatically by machines.

Follow steps, do not jump straight into autonomy.

The safety framework that Sinh Vũ applies consists of three levels of increasing autonomy:

There is no reason to jump straight from having no process to Level 3. Automating chaos only makes chaos happen faster.

Common mistakes

• Expecting incorrectly: Buying tools with the hope of completely replacing people, then being disappointed when supervision is still needed. An automated team is a support system, not a replacement.
• Turn on the machine without rules: When internal processes are unclear, the machine will replicate that chaos on a larger and faster scale.
• No logbook: Without an audit log (a record of all machine operations), when errors occur, there is no way to trace the cause.
• Forget the cost of maintaining the system: Monthly subscription software, third-party API fees, and the time of the supervisor are all real costs. The efficiency equation must consider both sides.

The viewpoint of Sinh Vũ

Sinh Vũ operates under the OPC model, utilizing an automated assistant team in real processes, not theoretical consulting. Practical experience shows that automation only adds value after the approval rules and processes have been standardized. Therefore, in the O2 package, the automated team is an optional add-on, placed after the process establishment phase, not as a starting point.

The human-in-the-loop check is not a sign of a weak system. It is the right design for every step that carries brand risk.

Improvado, Human in the Loop AI Marketing; House of Martech, AI Agents in Marketing Governance.

Sinh Vũ does not commit to specific savings figures as those numbers depend on each company's internal operations, not on tools. What Sinh Vũ guarantees is: if you do not have a clear process yet, the first step is not to choose an automation tool.