The data challenges of a large organisation
A large organisation does not lack data. It produces it in every business application, every sensor, every customer journey. What it lacks is more prosaic: an identical revenue figure in the finance report and in the sales report, a customer reference base without duplicates, a quick answer to the question “where does this value come from?”. A data expertise engagement almost always starts from this observation, rarely from a lack of technology.
The symptoms are similar from one organisation to another. Committees spend their time reconciling figures instead of deciding. Business teams rebuild their own extracts in spreadsheets, for lack of confidence in the official reports. AI projects remain at the demonstration stage because the data feeding them is neither documented nor reliable. The platform bill increases without anyone knowing how to link it to a use.
Regulatory constraints add to these internal difficulties. The GDPR, Regulation (EU) 2016/679, requires you to know which personal data is processed, for what purpose and for how long. The AI Act, Regulation (EU) 2024/1689, subjects high-risk AI systems to requirements on the data used to train, validate and test them. An organisation that cannot describe its data struggles to demonstrate its compliance.
Our view: what makes a data programme fail or succeed
Data rarely fails all at once. It gets bogged down, workstream after workstream, for four reasons found in most organisations.
The platform comes before the uses. A complete technical foundation, delivered without a priority use case, waits for its users. Architecture choices made without knowing the expected refresh frequency, the volumes or the sensitivity of the data are then paid for in redesigns.
Nobody is accountable for the data. As long as no business owner takes responsibility for the definition and quality level of a piece of data, every anomaly goes back to the technical team. That team has neither the authority nor the business knowledge to decide.
Quality is dealt with at the end of the chain. Correcting in the warehouse data that was badly entered in the source application means starting again with every load. The cost moves, it does not disappear.
Production release is underestimated. A useful dashboard or model must be monitored, documented, maintained and funded. Without organised operations, the value demonstrated during the pilot erodes.
Programmes that last show the opposite traits. They start from a small number of business decisions or processes to be improved, and size the foundation accordingly. They appoint business owners before choosing a catalogue tool. They measure the actual use of what they deliver and retire what is not used.
Above all, they treat data as a system. A platform without governance produces data that nobody dares to use. Governance without a platform produces documents. Analytics without quality produces debates about the figures.
Data expertise: how do we work?
Our work follows a four-stage approach: Frame, Prove, Industrialise, Operate. Framing sets the priority uses and the decisions they must inform. Proof is obtained on a limited scope, with real data and real users. Industrialisation extends what has worked. Operation puts in place monitoring, support and cost tracking.
Five domains structure this work. Each one has a detailed page.
The technical foundation. Our consultants contribute to defining the architecture, ingestion pipelines, modelling and operation of your data platform. The choice between warehouse, lake, lakehouse and data mesh is dealt with there on criteria of use and organisation.
Responsibilities and rules. Data governance establishes who is accountable for each critical piece of data, how it is documented and who can access it. It is also the framework within which GDPR compliance is handled.
Reliability. Our work on data quality and reference data covers measurement, business rules, automated controls and the management of shared reference data.
Reporting and analysis. Business intelligence and analytics turn data into indicators defined only once, dashboards that are actually consulted and analyses accessible to the business.
Industrialisation of models. MLOps and LLMOps take predictive models and applications based on language models from prototype to a monitored and maintained service.
You can call on us for just one of these domains. We then check its dependencies with the others, because they often determine the result: a semantic layer presupposes agreed definitions, a quality control presupposes an owner to handle the anomaly, a model in production presupposes stable input data.
In all cases, we work with your teams and not in their place. A platform or a governance that your staff have not built does not long survive the departure of those who designed it.
The link with our other areas of expertise
Data is not an isolated domain. Three links deserve to be made explicit from the framing stage.
With artificial intelligence. An assistant that answers from your documents is only as good as those documents: if they are out of date, contradictory or poorly classified, the answers will be too. A predictive model trained on biased data reproduces those biases. Data workstreams therefore directly condition our work in artificial intelligence, from strategy to agents.
With cloud and platforms. A data platform is a workload that is demanding in terms of compute and storage, subject to the same questions as the rest of the information system: hosting, sovereignty, security, automation, cost. We deal with them using the methods of our cloud and platforms expertise.
With transformation management. Management Square has its roots in project, programme and portfolio management advisory. A data programme is first and foremost a programme: it has sponsors, dependencies, resource decisions and working habits to change. We apply the same management discipline to it.
Where to start?
Three questions are enough to position an organisation and choose the first workstream.
- Which decisions are currently taken on disputed figures? The answer identifies the indicators to be made reliable first and the business areas to involve.
- Who is accountable for the data feeding these decisions? If nobody has been appointed, governance is the first workstream, before any technical investment.
- What does the current platform cost, and for which uses? If the answer does not exist, you need to measure before modernising.
We suggest starting with a short assessment, conducted with your business and technical teams. It leads to a roadmap that fits on a few pages: the priority uses, the foundation workstreams they require, the order in which to undertake them and the decisions expected from senior management.
The first use case is then taken through to production, on a deliberately limited scope. It is this that proves the method, reveals the real difficulties and trains the teams. To discuss it on the basis of your situation, get in touch with us.