From disparate spreadsheets to a secure, scalable AI platform in 6 months
There’s plenty of posts about enterprise AI programmes stalling somewhere between the demo and production. The smoke and mirrors of a shiny PoC joining the graveyard of vanity projects, neglected by the frustrated team.
This blog will share the 4-month journey of partnering with a critical national infrastructure operator taking enterprise AI ambition from pipedream to platform.
The starting point was pain, not ambition
Our work began with a team drowning in spreadsheets. This team was managing partner relationships in Excel, preparing high-value bids by manually assembling content from dozens of documents, and tracking market intelligence across regulatory filings and news sources for multiple territories. We identified 3 high-value, shippable use cases to focus on.
The platform needed to be secure and scalable, yet seamless
The AI platform we built is a production-grade system housed on Azure. Its value comes from enabling purpose-built agentic workflows to process the organisation's data, linked with other tools such as web search to automate the research, analysis, and document generation.
This foundational capability initially enables 3 use cases, which can become more. Currently in build:
The Relationship Manager is driving increased partnership conversion.
In 3 months, we built a system for systematically storing vital data, with analytics obtained by MCP (Model Context Protocol) directly from the source data to track engagement across every territory.
Using AI, the user types "summarise my meetings with this partner" in the chat interface and the system queries the CRM, pulls meeting records, synthesises themes across multiple conversations, and returns strategic context: what the partner cares about, where the relationship stands, what is still open. Ask "what should we do to win this opportunity?" and it produces a structured strategic recommendation grounded in actual meeting history, generating downloadable briefing documents in seconds.
The Bid Preparation System allowed our customer to move fast.
The AI system that generates complete proposals in minutes, in the right format. Weak or incomplete answers are flagged so the team knows exactly where to focus before submission.
The Market Intelligence Hub differentiates bids.
The first live version provides deliberate and process competitive analysis reports on demand. It fetches live regulatory data, parses official filings, and builds competitor profiles across financial strength, track record, geographic positioning, partnership capability, and strategic intent. Each competitor is rated for threat level and partnership potential. The output is a structured analytical report.
It replaces roughly eight hours of manual work per reporting cycle with an agentic workflow that a human reviews and adds strategic judgement to in a fraction of the time.
Open, honest collaboration is vital for the system to be valuable
Focused collaboration with the end users allowed us to create a gold-standard level before production code was written: quantified acceptance criteria, defined accuracy targets, user acceptance testing plans, data classification, guardrails, and human-in-the-loop checkpoints. A specification rigorous enough that an engineer can build from it without questions and a tester can verify against it without ambiguity.
All three run on Azure within the client's security perimeter, deployed through CI/CD with observability, monitoring, and cost attribution and tight role management, requiring complex architectural design - cloud, enterprise and software.
These use cases aren't items you hand to an off-the-shelf chatbot. Systems that create market-differentiating outcomes require live external data ingestion, deterministic protocols, audit trails, MCP orchestration, CRM & other application integration, and bespoke multi-agent workflows leveraging a multi-model approach. That’s the difference between a tool for individual convenience and a platform for the work that decides whether you win bids.
Every AI-generated output follows a guardrail framework: claims cite their source, training-data knowledge is never used for factual assertions, uncertainty is stated not guessed at, and nothing reaches a record until a human approves it. The AI governance framework integrates with the parent organisation's enterprise policies.
Custom, compliant, production-grade AI is hard
Three lessons to take when we look back at the success of this engagement.
The biggest blockers were organisational, rather than technical. Governance processes built for nine-month waterfall cycles don’t accommodate the pace of AI delivery where you can deliver a working product in 1-2 months.
Adoption is a delivery problem, and ensuring that users feel comfortable and confident with the solution is vital. When usage dipped, we treated it as a product defect, not a change-management escalation.
The specification is the product. Spec-driven development, where a rigorous specification is written before code and a working prototype is built alongside it, meant engineers understood intent viscerally and built the right thing the first time. It also meant every AI feature had testable accuracy criteria, defined guardrails, and a clear human-in-the-loop model from day one. Speed went up and quality didn’t go down.
Electricity isn’t the only industry to benefit from this capability
These use cases at their core share foundational capabilities from the AI platform that allow for:
automated, intelligent monitoring of regulated, public information sources with AI qualification of opportunities;
RAG-based generation of complex proposal documents from a practically ‘indexed’ organisational knowledge base with compliance checking;
a relationship intelligence layer that enriches CRM records from external data.
These applications could quickly embed in other regulated industries, formal tendering processes and long-cycle partnership development.
Energy & Utilities. Offshore wind developers, water utilities, gas networks and heating and hydrogen sectors all navigate complex regulatory requirements from multiple sources. Similarly to electricity, an automated AI platform to manage all research, monitoring and response generation would be enormously beneficial.
Infrastructure & Construction. Contractors bidding into National Highways, Network Rail, or HS2 handle multiple complex requirements and stakeholders, similarly to the electricity sector.
Defence & Government Contracting. Suppliers monitoring tender portals produce highly structured, compliance-driven proposals. Sensitive data handling and tender compilation, as in this example, are highly relevant.
Financial services. Investment firms including private equity and venture capital companies, and law firms, track competitors and partners to analyse the market and generate responses. AI-automated monitoring and response generation speeds up document drafting to just minutes.
Healthcare & Life Sciences. Pharmaceutical and medtech companies tracking regulators and providers need partner intelligence and CRM enrichment.
Telecoms. Spectrum auctions, Ofcom consultations, and fibre tenders replicate the regulated opportunity landscape.
Real Estate. Real estate developers can reuse geographic and stakeholder mapping to track planning applications, council positions, and local political sentiment.
