Client type
Fibre Network Group
Key stakeholders
Multiple business lines
Location
South Africa
Project outline
- A leading South African fibre network operator group needed line-level visibility into revenue, network performance, and customer activity across its operating businesses.
Before this engagement, every executive insight was assembled by hand from disconnected source systems, with no shared analytical foundation.
StructureIt was engaged to build the full stack requiring a medallion data lakehouse, an AI-driven analytics and dashboarding layer, and a Model Context Protocol (MCP) server giving the client access to the platform through whichever AI tool their team already uses.
The Challenge
Running a fibre network at scale generates enormous volumes of operational, financial, and commercial data. For our client, that data was spread across dozens of disconnected systems and never reached the executive table in a form ready to act on.
Our client operates one of South Africa’s largest fibre footprints, spanning multiple operating brands, an extensive precinct base, a wide network of Internet Service Provider (ISP) partners, and a continuous build programme. As the business grew, the distance between what was happening on the network and what reached leadership had widened to the point where decisions were being made on incomplete, lagging information.
Four structural problems compounded the situation. Source data was spread across billing, operational and business support systems, network performance platforms, financial ledgers, accounting systems, and partner statements, each with its own schema, cadence, and data definitions. Reporting was ad hoc, which meant when a revenue, connections, or uptake question arose, a finance team member hand-built the answer in a spreadsheet, a process that absorbed significant time before the numbers reached anyone who could act on them. Commercial leadership had no trusted view of precinct-level uptake or ISP-level performance, resulting in decisions about where to direct investment, where to reduce focus, and which partners to prioritise were made on instinct rather than evidence.
There was no foundational analytical layer in place at all. Trend reporting, performance comparison, and exception flagging were not routine outputs; each had to be constructed from scratch by senior finance staff, which meant strategic analysis was rationed by whatever time the team could spare. The result was a senior leadership team running a scaled business without the day-to-day reporting infrastructure that scale demands. By the time the numbers had been pulled together, the window to act on them had often already closed.
Our approach begins by learning and understanding a client’s problems in depth before any solution is designed. The engagement opened with a structured discovery process to map the full landscape of data sources, reporting needs, and stakeholder requirements before a single line of architecture was drawn.
How StructureIt helped
The planning phase brought together our client’s finance, operations, and commercial teams alongside the StructureIt team to map the right tools, methods, and sequence for a reliable build. Rather than applying a generic data architecture, the team designed a platform purpose-built for the Fibre Network Operator (FNO) business model, one where a modern data foundation and an artificial intelligence (AI)-driven intelligence layer sit together as a single system. Both halves are deliberate, where one without the other does not generate the visibility a business at this scale needs.
The build follows a three-layer architecture. At the base, a medallion data lakehouse on Databricks ingests structured and unstructured feeds from network, billing, finance, and partner systems via a contract-driven model that accommodates schema changes without breaking downstream consumers. Data contracts, schema versioning, lineage tracking, and a hash-chained audit trail mean every figure that lands in front of an executive is traceable back to its source system and the moment it was loaded.
On top of that foundation sits a Revenue Integrity Platform designed to reconcile network activity, billing events, and financial postings continuously. Service lifecycle tracking, network-versus-billing reconciliation, product tier verification, wholesale billing validation, and infrastructure utilisation monitoring all run as standard outputs. Finance can see leakage, lag, and exceptions where they happen, rather than discovering them weeks later in a variance review.
The intelligence layer brings a library of named AI agents online progressively, each reading from the same governed data layer and inheriting its lineage, role-based access, and audit controls. The Revenue Integrity Agent surfaces revenue leakage with explanation and root cause. The Uptake Performance Agent maps adoption patterns by geography, ISP, and product. The ISP Performance Agent compares partners across revenue, subscriber growth, churn proxy, and activation times. The Executive Insight Agent synthesises growth and decline drivers for the leadership team.
All four agents are fed from one reconciled data spine. Multiple delivery surfaces sit on that spine, each shaped to how a different audience works. An executive deck delivers performance, insights, and action points on a regular monthly cadence. Business Intelligence (BI) dashboard workbooks give finance the granular cuts it needs for variance analysis and lifecycle questions. The Commercial Intelligence console gives the Chief Commercial Officer (CCO) and commercial leadership team a three-level master-detail view that runs from group performance down to the individual precinct and into the bilateral ISP-by-product detail, with insights, focus lists, status notes, and criticality scoring sitting alongside the numbers.
Alongside these surfaces, the platform’s Model Context Protocol (MCP) server gives the team another way in. The same reconciled data and agent catalogue is exposed to whichever AI tool the team is already using, so executives and analysts can ask plain-English questions of the data without learning a new interface. The MCP server sits inside the client’s own tenant boundary. Every call routes through the same identity, role-based access, and governance controls as an internal one — tenant-scoped anonymisation, a verification gate before any model is invoked, and an append-only audit chain. While the access channel changes, the governance does not.
Most AI failures in finance and operations are not model failures but data failures. A language model cannot reconcile a number it cannot find, or explain a metric whose definition shifts between systems. The output looks credible, but the underlying answer is not defensible. The platform treats this as a single design problem where the lakehouse provides structure, governance, and lineage; the AI layer reads from that governed surface and nothing else. Every answer is grounded in the same numbers the finance team has already signed off on, and every call leaves an audit trail.
“The solution that StructureIt developed for us structures vast volumes of information into clear, accessible reporting, making trends and insights immediately visible. This has fundamentally changed the way we are able to view and interpret our data.”
The Results
Both operating businesses now work from a shared lakehouse dataset, with the first analytics surfaces in active use by finance and commercial leadership. Executives review performance against a single audited source rather than a stack of reconciled spreadsheets, and the platform surfaces reconciliation differences between network, billing, and financial systems as they arise, rather than leaving them to surface weeks later in a variance review.
Questions that previously absorbed significant amounts of senior finance time now resolve against a single governed dataset on a much shorter cycle, freeing up capacity for the strategic analysis the business actually wanted from its finance function. The agent library is being progressively brought online, and the MCP server already lets executives ask plain-language questions of the data from within the AI client they were already using, with answers grounded in the same governed numbers the finance team has signed off on. No BI queue, no parallel spreadsheet, no risk of an unverified figure reaching the board.
On the commercial side, the CCO and the team now have a shared console that lets them navigate from a group view to an individual precinct and into ISP-by-product detail in a few clicks. The platform is being used to manage precinct uptake, surfacing where activations are tracking behind plan and what intervention is required, and to ground ISP partner conversations in numbers both sides agree on. Decisions that previously depended on a senior commercial leader’s read of partial information are now backed by a defensible dataset and an agent layer that can explain its reasoning.
What this engagement demonstrates is that AI agents in finance and commercial operations are only as trustworthy as the data layer beneath them. The work StructureIt did here was to build that foundation first, then build the intelligence on top of it. The result is a platform our client can grow from, with governance and auditability already in place as the business scales.
Collaboration, on your terms
We collaborate with you to develop personalized software solutions that contribute to your business growth in the capital markets
