TLDR: As Addesu grew, our internal knowledge turned into digital noise. Essentially too many documents in too many places. By shifting AI’s role from writing guides to verifying claims, we cut documentation time from 90 minutes to just 10. The result is a democratised knowledge hub where internal expertise is verified, accessible and ready to share to the entire team regardless of seniority.

Background

Growth is always the goal. However, as our team scaled, we hit a frustrating paradox: the more expertise we gained, the harder it became to actually leverage it effectively.

It all came down to the mass volume of internal documents. Best practices, platform insights, industry updates and tactical guides were proliferating across channels. When your Best Practices live in three different folders and four different formats, they just become digital noise. Ultimately, essential business intelligence lived in different storage locations, inconsistent file formats and varying levels of quality and completeness.

The real problem was fragmentation.

And the solution? Standardizing Business Intelligence in a way that maintains accuracy, governance and senior oversight.

The Problem: The Human Bottleneck

Before we automated this, our documentation process was, frankly, too slow. A senior lead would spend 90 minutes drafting a guide, only for another senior lead to spend 30 minutes fact-checking it.

There was a lack of order, tonal inconsistency and operational guardrails which made documentation slower and riskier to produce, difficult for new members to use and hard to rely on for client-facing knowledge assets.

The Solution: The Constrained AI Agent

Hence, we built a highly constrained AI agent designed with the central purposes of accuracy, clarity and auditability.

Our educational shift consisted of instead of asking the AI to write a guide, we taught it to treat every input as a claim that needs verification.

How the Agent works:

• 1. Seamless Ingestion: You can feed it raw notes, a messy URL, or even just a handful of screenshots.
• 2. Evidence-Based: It treats all inputs as claims, not facts, and verifies them.
• 3. Validating control: It flags where a screenshot is missing and visual confirmation is required, or where some highlights need further human evidence to be showcased.
• 4. The Guardrails: We programmed strict structure, formatting and terminology, as well as next steps checklists that go through us, keeping it all under our control.
• 5. Publication-Ready: It doesn’t output a draft, but rather a standardized, formatted document ready for the team to use immediately.

The Results: From 90 Minutes to 10

The shift from manual documentation towards an AI-standardized workflow was immediate and clearly measurable as seen below:

Key Takeaway

Our built-in AI Agent led to an 85% increase in time saving (our seniors’ calendars really thank us). Yet, even more impactful, the same constrained agent is now used to fact-check existing client-facing documentation, ensuring that as platforms like Meta or Google change their UIs, our guides don’t become relics of the past.

The real value is that expertise is now fully integrated into learning content for everyone in the office, making it accessible, verifiable and consistently reliable regardless of seniority.

These democratised information hubs are where we believe AI contributes the most with its standardised and formalised skills. We have all seen a lot of businesses recently opting for AI over human knowledge. We wish to stress that it is by enabling AI rather than using it as a replacement where real long-term growth comes.