Powerful but general
AI models are statistical systems trained on vast amounts of the world's information. That makes them powerful but general. They don't know your business, your staff, or your experience.
Dottie understandsyou
Your projects, your people, your language.
How Dottie uses AI
Two solutions
There are two ways to fix this: fine tuning or Retrieval Augmented Generation (RAG). Fine tuning rewires the model itself. It's powerful, but expensive, and once your data is in, it stays there.
Why RAG
RAG works differently. The AI searches your data in real time and pulls in what's relevant. It's faster to set up, more affordable, and you stay in control.
How Dottie works
Dottie is built on RAG. Upload your files and Dottie categorises and indexes them automatically. Over time, you build a powerful, curated knowledge base for Dottie to draw on.
The secret sauce
Data access alone isn't enough. Without instructions, AI gets confused. So we built Dottie to know not just what your files contain, but how to use them.
In practice
Ask Dottie to draft a proposal and it searches the RFP for specifications, your technical files for relevant experience, and past proposals for how you like to pitch.
Your data stays yours
The AI never learns from your data, so you can remove it at any time. Your competitors may use similar systems. Your data is what makes you unique.
Security without complexity.
Dottie uses Microsoft Azure and Azure OpenAI.
The same trusted AI foundations as Copilot.
| Part of the stack | How Dottie handles it |
|---|---|
| User authentication | Your account is protected through Microsoft's secure sign-in platform. |
| Data storage | Your documents and project data are stored in Microsoft data centres in Australia. |
| Data transfer | Your data is encrypted in transit and at rest, using Microsoft's enterprise-grade security standards. |
| AI processing | Handled through Azure OpenAI on Microsoft's secure global network. |
| AI model training | Your data is never used to train the underlying AI models. Microsoft strongly commits to this here ↗. |
Dottie vs
ChatGPT
We ran three AI workflows across multiple firms, each producing a draft proposal section.
The chart shows the average similarity of these drafts to real submitted proposal sections.
The higher the score, the less editing between AI draft and final.
Average benchmark scores: RFP plus simple prompt 34 percent, RFP plus past proposal plus simple prompt 35 percent, and Dottie 62 percent. The real submitted proposal section is 100 percent.
Method
We tested three AI drafting approaches against real proposal sections previously submitted by each organisation.
- A simple prompt and the tender documents. The most basic AI drafting workflow.
- The same simple prompt and tender documents, plus one example of how the organisation has responded to similar work before. More context, but no guidance on how to use it.
- Dottie, which uses a broader organisational knowledge bank combined with sophisticated prompts. This is designed to produce a draft that is closer to how the organisation actually writes.
Each output was scored against the real human-written section that was actually submitted. The score measures two things:
- Semantic similarity: whether the draft is talking about the same ideas as the real proposal.
- Structural similarity: whether the draft has the same shape, depth and length as the real proposal.
We weight structure slightly more heavily, because a draft that's the right length and format is faster to edit into something usable.
The benchmark isn't trying to prove AI can replace a proposal writer. It can't. It's measuring how much editing effort each approach is likely to leave behind.