Pretend you already have all the data in the world. If you don't know what to do with it, no tool will fix that

Requests arrive phrased as a solution, not as a problem. Build us a data lake. Set up a machine learning engine. In this episode there is one counter-question for this that is worth stealing: pretend you already have it in place, with everything that exists in it, and tell me what business problem you are solving right now. Without that answer you are looking for a needle in a haystack you piled up yourself. A technology strategy is not a data strategy, and a tool has never been the answer to a strategic question, even though that is where companies most often start. The same mistake comes back every few years under a new name, so it is easy to recognise: the request names a class of system, not a sentence about what the company can't decide today.

The hard part of this work isn't configuring the tools, it is that every department defines the same concepts its own way. The episode gives two examples that make this vivid. A bank in which one system kept accounts at branch level and another at the level of the whole bank, so the data couldn't be reconciled, even though reporting had to be done at the lower level. And an organisation in which two sales divisions assign the same product to a different business unit and a different legal entity, so the sales figures simply can't be added up. Hence a distinction that comes back several times in the conversation: information architecture is not data architecture. Before you describe the tables in a system, you have to know what a client means in this company and what relationships it has in the business context. I write this with some embarrassment, because we ourselves use the word strategy as if everyone understood it the same way, and this episode shows that the assumption doesn't hold even inside a single organisation. There is no standard definition for data strategy, for governance or for a catalogue, so the first step is to agree within your own company what these words mean for you.

The rest is about execution and will sound familiar to anyone who has worked with objectives. A strategy without a focused execution plan is a nice thing on a slide, and carrying out tasks without a plan will only take you so far, so the two have to feed each other. Instead of boiling the ocean, doing twenty per cent of a hundred things at once, it is better to choose five and deliver them in full, because only a finished thing delivers value. Nor is there a single indicator that tells you how you are doing: each pillar, from data quality to privacy, is measured separately and in proportion to where you are on a five-stage maturity curve. And the most important decision is organisational. Data strategy and business strategy are not two parallel tracks with points of contact, but one conversation, so the person responsible for data should sit among the stakeholders of the company's strategy, not receive its results to carry out.