Most companies using AI for content, proposals, or internal work hit the same wall. The outputs are technically fine. Grammatically correct. Reasonably structured. But they do not sound like the company.
Wrong proof points. Generic positioning. A tone that could belong to anyone. So someone sits down and rewrites it.
At that point you have to ask: was this actually faster than writing it from scratch?
The real problem is not the model
When you open a new AI conversation and ask it to write a proposal, it starts from zero. It does not know your services. It does not know your pricing logic. It does not know your past projects, your differentiators, or even your company name beyond what it can guess.
So it does what any reasonable system would do with no context — it produces something generic.
People blame the AI. They try better prompts. They switch models. They add more instructions to individual messages. None of this solves the fundamental issue.
The fix: a structured knowledge foundation
The fix is what we call a knowledge foundation — sometimes called organisational context or a company wiki. It is a structured set of documents that contain everything the AI needs to represent your company accurately.
This is not a single “about us” paragraph. It is a comprehensive, maintained collection that covers:
- Company identity — mission, positioning, differentiators, history
- Services and capabilities — what you actually deliver, at what level of detail
- Proof points — case studies, client outcomes, metrics, references
- Pricing logic — how you price, standard rates, commercial rules
- Process definitions — how work gets done, methodologies, frameworks
- Brand voice — tone, vocabulary, what you say and what you never say
- Operational context — team structure, tools, partnerships, certifications
- Target audiences — who you sell to, their problems, their language
What changes once it exists
Once this knowledge base exists, every AI interaction draws from it. Every output starts from an accurate, specific, on-brand position.
The proposal AI knows your actual case studies. The content AI uses your real differentiators. The tender response AI applies your actual pricing model. Nobody has to paste context into every conversation.
But the real power is the compounding effect. Each time you add something to the knowledge base, every AI output across every use case gets better simultaneously. Add a new case study once — it is immediately available in proposals, marketing content, sales outreach, and tender responses. Update your pricing — every skill that touches commercial numbers reflects the change.
We built this for ourselves first
We built this approach for ourselves at DigiDuo before delivering it to anyone else. Our own knowledge base feeds our content production, our proposals, and our internal workflows.
Then we delivered it for a client — a professional services firm with 30+ people, working across ITAM, SAM, and ITSM. Their workspace now has 15 authoritative knowledge base files feeding 18 different workflow skills. Every output their team produces through the AI system is grounded in their actual company knowledge.
How to build one
The foundation takes 2-3 weeks to build properly. Here is the honest breakdown:
Week 1: Gather raw materials. Interview stakeholders. Pull existing collateral — proposals, pitch decks, case studies, process docs. Identify gaps.
Week 2: Structure and write the knowledge base files. Each one follows a consistent format. Each one is authoritative — meaning it is the single source of truth for that topic.
Week 3: Test and refine. Run real tasks against the knowledge base. Find where the AI still gets things wrong. Fix the underlying knowledge, not the prompts.
The critical point: this should be built before any AI automation. If you build skills first and add knowledge later, you end up retrofitting. Every skill needs rewriting. Build the foundation, then build on top of it.
The bottom line
If your AI outputs feel generic, the problem probably is not the AI. It is that the AI does not know who you are.
Fix that, and you will be surprised how capable the technology already is.