Building an AI-assisted content workflow at KPMG Portugal
A knowledge base in Claude Projects that cut the time from first draft to final approval from weeks to days.
Context
At KPMG Portugal, I produce a steady flow of content: newsletters, emails, LinkedIn content and website pieces. All of it has to meet strict brand guidelines and pass marketing compliance review before publication.
The challenge
Getting content from first draft to final approval took 2–3 weeks. The delays came from several sources:
- Gathering information for each piece: positioning, personas, market context
- Writing briefs from scratch every time
- Review cycles, including several rounds with marketing compliance
There was also a consistency problem. Different pieces were written by different people, and each one sounded different. For a firm trying to build a credible, recognizable voice on AI, that worked against us.
Strategy
My hypothesis was that most of the delay came from context that had to be rebuilt for every piece. If I centralized that context in one place, together with the compliance rules, every draft would start much closer to final and need fewer correction rounds.
Execution
I built a knowledge base in Claude Projects with all the information that shapes KPMG’s AI content:
- KPMG brand guidelines and brand voice
- Marketing compliance rules
- My own research on competitors and the market
- The buyer personas, customer journey and positioning I had developed for the AI offering
Now, whenever I create a piece of content, the workflow is:
- First draft: the project writes a first draft grounded in the brand voice, personas and positioning.
- My edit: I work from that draft, shaping the argument, angle and message.
- Compliance check: the project reviews every version against the compliance rules before it goes to formal review.
- Approval: I review and approve the final version.
Results
- Fewer revision rounds with marketing compliance, because issues are caught before formal review
- A more consistent voice across pieces, because every draft starts from the same guidelines, positioning and personas
Learnings
This project developed my skills in designing AI-assisted content workflows. It’s less about the tool itself than about the system around it: deciding what context the AI needs, where in the process it adds the most value, and where human judgment has to stay.
This transfers directly to any content team under pressure to produce more without losing quality or consistency, especially in regulated or brand-sensitive environments where every piece goes through review.