The client and their challenge
Jadarah Consulting is a mid-sized consulting firm in Saudi Arabia that participates regularly in government and corporate tenders. Before our engagement, their proposal team of three people was spending 60% to 70% of their working hours on RFP responses — reading requirements, structuring responses, writing content, and formatting final documents. They were turning down approximately half of all incoming RFPs due to capacity constraints. Every missed RFP was a missed revenue opportunity.
The solution architecture
We built a system with four components. First: a document ingestion engine that accepts PDF, Word, and Excel uploads and converts them into structured, searchable content using OCR and our Arabic NLP pipeline. Second: a requirements extraction module that identifies and categorizes every stated requirement, evaluation criterion, and submission guideline — typically 95% accurate without human review. Third: a proposal generation engine powered by GPT-4 that, given the extracted requirements and the company's capability profile, generates a full proposal draft section by section. Fourth: a review interface where proposal managers can edit, regenerate specific sections, and export the final document in formatted PDF or Word.
The results — measured over 6 months
Average time per proposal: from 3.2 days to 5.8 hours — a 78% reduction. Monthly proposals submitted: from an average of 4 to 11. Win rate: improved by approximately 15 percentage points, attributed partly to higher quality proposals (less rushed, more thorough) and partly to the ability to bid on more opportunities. Team satisfaction: the proposal team reported significantly lower stress levels and more time for strategic activities like client relationships and solution design.
What surprised us during the build
Two things surprised us. First, the Arabic bilingual output was more useful than the client initially expected. Having proposals automatically generated in both Arabic and English eliminated a translation step that previously added half a day to every submission. Second, the requirements extraction module surfaced requirements that the human team had consistently missed when reading long RFP documents — sometimes critical eligibility criteria buried on page 47 of a 60-page document. The AI's systematic reading proved more thorough than human reading under time pressure.
What the system does not do
We want to be transparent about the system's limitations, because we see too many AI vendors overpromise. The system generates draft content — it does not finalize proposals. Senior staff still review and refine every output before submission. The system does not replace proposal strategy — it executes the writing, not the thinking. And the system requires an initial setup period of two to three weeks to ingest the company's existing proposal library and build its capability knowledge base. There is no instant magic — there is a well-engineered process.