.png)
How I built an AI-supported workflow to remove content bottlenecks
Designed and implemented an AI-supported content workflow that removed speed and bandwidth bottlenecks, enabling faster reactive content while keeping human judgment and quality control in place.
INVOLVEMENT
Full Time
CREDITS
BRAND / COMPANY
B2B Finance
YEAR
2025
PROJECT SUMMARY
Content creation was constrained by speed and limited bandwidth, especially for reactive moments such as events, announcements, or founder-led topics. I introduced an AI-supported content workflow designed to accelerate execution without outsourcing thinking or compromising quality. The system made it easier for the team to respond quickly while keeping content aligned with brand context and platform-specific formats.
I designed the system end to end. This included defining the workflow, selecting and connecting tools, writing and refining prompts, and setting clear guardrails around how AI could and could not be used.
I trained the marketing coordinator on how to use the system effectively and responsibly, positioning AI as a support layer rather than a replacement for human judgment. I remained accountable for quality and alignment, reviewing output rather than producing everything myself.
CHALLENGE
The main bottlenecks were speed and bandwidth. Creating content quickly for reactive moments required context switching and hands-on involvement from a small number of people, which slowed execution. As a result, opportunities to respond in the moment were often delayed or missed.
At the same time, fully automating content or outsourcing judgment to AI was not an option. Maintaining control over quality, tone, and positioning was critical.
APPROACH
I built a workflow centred around how the team already worked. A dedicated Slack channel allowed prompts such as “create a post about us attending X event” to trigger a sequence automatically.
The system first used Perplexity to gather relevant context on the topic. That information was then passed into an OpenAI API instance trained on company-specific knowledge, ensuring outputs reflected internal context rather than generic responses. The final step generated platform-specific copy tailored to different social formats. All outputs required human review and editing. There was no auto-posting. Prompts were carefully designed to guide structure, tone, and format, and the system was positioned as a productivity tool rather than a decision-maker.
In parallel, I created Canva templates that allowed content to be produced in batches, freeing up time and attention for truly reactive posts when needed.