The order matters more than the ambition.
Data, analytics, BI and AI are one chain. The value that reaches the end of it is the product of every link, not the best one. Most organisations invest heavily in the last link and wonder why so little arrives.
01 / The argument
Losses compound. They do not average.
If your data layer is trusted two thirds of the time, and your analytics layer covers two thirds of the decisions that matter, and two thirds of your AI output reaches a workflow, you have not lost a third. You have lost seventy per cent.
This is why starting at stage three is expensive. The programme is not competing for the value on the table. It is competing for whatever survived two stages of leakage nobody addressed.
02 / The sequence
Four questions, in this order, every time.
Each one is only answerable if the one before it is settled. Skipping ahead does not save time, it defers a cost and increases it.
Can you trust the numbers?
If two systems disagree and nobody arbitrates, everything downstream inherits the argument. This is an ownership problem far more often than a technical one.
Do the numbers reach a decision?
Reporting that has never changed a decision is a historical record. Useful occasionally, expensive always, and it will not support anything built on top of it.
Should a machine make that decision?
Sometimes yes, often partly, frequently not at all. This is the question everybody starts with and it is third for a reason.
Can you prove it worked, and defend it?
A baseline captured before go live and a governance trail that survives a customer, an insurer or a regulator asking. Neither can be added afterwards.
03 / Where I come in
Three groups, one chain.
I work across all of it rather than selling one layer, because the layer somebody sells you is rarely the layer that is failing.
Data and platform
The foundation: trustworthy, owned and affordable. Settle it badly and every layer above inherits the settlement, usually without knowing.
Analytics and AI
Turn the foundation into decisions people actually make, then sequence AI onto something solid rather than something hoped for.
Operate and govern
How the whole thing is run, and how you defend it to a customer, an insurer, an auditor or a regulator before any of them asks.
04 / In detail
What each one is, and when it applies.
Find the row that sounds like this quarter. If more than one does, the earliest in the chain is usually the one to start with.
| Service | You need this when | What you are left with | Typical shape |
|---|---|---|---|
| 01 Data and platform | |||
| Data modernisation | Three systems disagree, nobody arbitrates, and every number becomes a negotiation. | Estate map with named owners, measured quality baseline, costed migration sequence, decommissioning list. | 4 to 12 weeks |
| Platform optimisation | The cloud bill grows faster than the value, and no team sees the cost of its own workloads. | Cost attribution model, ranked savings list with effort and risk, licence position for renewal. | 3 to 6 weeks |
| Legacy modernisation | Business critical logic lives in Alteryx, Access or a spreadsheet nobody will admit to. | Inventory of what actually runs the business, a migration path, and the manual hours removed. | 6 to 16 weeks |
| 02 Analytics and AI | |||
| Analytics and BI | You have more reporting than you can use and less insight than you need. | Usage backed inventory, retirement list, governed metric layer, reporting operating model. | 4 to 8 weeks |
| AI transformation | Twelve ideas, all sponsored, none sequenced, competing for the same three people. | Prioritised portfolio with stops marked, honest build or buy, value cases with baselines. | 3 to 8 weeks |
| Agentic workflows | You want systems that act, not just advise, and nobody has defined what happens when one is wrong. | Authority boundaries, tested rollback, named accountable owner, instrumented usage. | 6 to 12 weeks |
| 03 Operate and govern | |||
| Target operating model | Everything takes longer than it should, nobody can refuse anything, and four people believe they own the roadmap. | Decision rights on one page, demand governance with an agreed basis for saying no, a funding recommendation and a sequenced target shape. | 6 to 10 weeks |
| Governance and assurance | A customer, insurer or regulator could ask tomorrow and the answer would be a scramble. | Inventory, classification, policy, model risk register, audit trail that holds up. | 4 to 8 weeks |
| The 7D Baseline | You are not sure which of the rows above is actually yours. | All seven gates scored, prioritised order of work, value cases, a plan leadership can act on. | 3 weeks |
Durations are indicative and depend on estate size and how many people I need to speak to. Everything is fixed scope and fixed fee, agreed in writing before it starts.
Which link is yours?
Seven questions, four minutes, no email. It will tell you which part of the chain is actually holding you up, which is usually not the part being blamed.