Data analytics and BI

The typical organisation has more reporting than it can use and less insight than it needs. Both problems have the same cause, which is that reporting gets built on request and retired on nobody's.

Analytics is the least glamorous link in the chain and the one that decides whether anything above it works. AI sequenced onto an analytics layer nobody agrees with inherits the disagreement, at greater cost and with more confidence.

What usually needs fixing

The dashboard graveyard. Dozens or hundreds of reports, a handful genuinely opened, several still refreshing nightly at real compute cost for an audience of nobody. Counting this is uncomfortable and immediately actionable.

Competing versions of the same number. Finance and operations both right, using different definitions, in the same meeting. The argument is not about data. It is about which definition the business runs on, and that is a decision nobody has been asked to make.

Self service that nobody uses. The tool was rolled out without the definitions, the training or the trust. Self service without a governed metric layer does not democratise analysis, it multiplies the number of conflicting answers.

Reporting instead of decision support. The pack describes what happened without telling anyone what to do about it. If a report has never changed a decision, it is a historical record, and it should be priced as one.

How I approach it

Start by counting. How many reports exist, who opens them, how often, and what decision each one supports. That single exercise usually justifies the engagement, because the answer is worse than anyone expects and the remedy is obvious once it is visible.

Then settle the definitions. A governed set of metrics with named owners is worth more than any tooling decision, and it is what makes self service work rather than multiply confusion. Tooling comes last and matters least.

If a report has never changed a decision, it is a historical record. Useful occasionally, expensive always.

Where I am careful

Rationalisation makes people nervous, because a report nobody opens is still somebody's work. It has to be handled as a conversation rather than an audit, or the next inventory gets quietly obstructed.

I am also wary of replatforming as a first move. Moving a broken reporting estate to a better tool produces a broken reporting estate with a better licence fee.

What you get

• A full inventory of the BI estate with real usage data attached, not assumptions.
• A retirement list, which is usually longer than the build list, with the cost of keeping each item.
• A governed metric layer: definitions, owners, lineage, and where the truth lives when systems disagree.
• A reporting operating model covering who builds, who approves, who maintains and who retires.
• A decision support pack for the leadership team that replaces description with direction.
• A self service readiness view: what can safely be opened up, to whom, and what has to stay governed.

How this connects

This sits between data modernisation below it and AI transformation above it. Skipping it is the most common and most expensive sequencing mistake I see, because the cost does not appear until the AI programme is already running.

The free gate check will tell you in about four minutes whether this is where you are stuck.