Why Most Data & AI ‘Best Practices’ Are Just Expensive Mistakes in Disguise

How many times have you heard this phrase “best practice?”

I’ve lost count of how many times I’ve sat in meetings, read RFPs, or listened to keynotes where the phrase “best practice” gets wheeled out like some kind of silver bullet. This post was inspired by a post that Adrian Smith did last week.

I do get it, well sort of. On the surface, best practices seem like a foundational promise, a proven path where only those who know tread are in the know, a safe bet on the poker table, or as some people might refer to it as an “industry standard”. Most just use the phrase because it’s what everyone else is doing!

There is one thing I’ve often noticed about the phrase, is this, most of what’s touted as a best practice in the world of data and AI is nothing more than a well-dressed mistake. Even worse, it’s often someone else’s mistake, packaged, dressed up and marketed, sold to everyone as THE gospel.

If you’re serious about making your data and AI strategy real, not theoretical, not a “we ticked the box” initiative, then now is the time to question everything. Especially the things you’ve been told are “best.”

Because chances are, they’re not.

The Comfort of Consensus vs. The Cost of Compliance

There is always a deep-rooted comfort in doing what others are doing. If everyone’s building data platforms, spinning up COEs, hiring Chief AI Officers, or adopting frameworks from the latest Big 4 whitepaper, then surely it must be the right move?

Except that’s not strategy. That’s just plain and simple compliance.

The problem with compliance is that it’s reactive. You’re not making choices based on what your business uniquely needs, you’re copying a blueprint from a company that has completely different problems, dynamics, and ambitions.

You’re not operating from first principles. You’re operating from FOMO.

The Seven Best Practices That Often Go Wrong

Let’s call out a few of the sacred cows. Not because they’re always wrong, but because they’re often misapplied, misunderstood, or implemented without any critical thinking.

1. Build a Data Lake / Lakehouse / Warehouse: you know the one. “Where you get to modernise your data infrastructure.” Sure. But instead of asking why, many jump straight into building. Huge capital expenditure. Long timeframes. And no clear commercial outcome. Then two years later, people are still emailing spreadsheets around.

2. Stand Up a CoE (Centre of Excellence): sounds great in theory. In practice? It often becomes a silo of clever people working on things the business doesn’t understand, value, or use. If it’s not embedded in the core of the business, it’s just an ivory tower with a Jira backlog.

3. Hire a Chief AI Officer: because, of course, we now need a new title to solve an old problem. One that is almost always a proxy for: “We don’t know how to do this, but we hope someone will figure it out for us.” Titles don’t solve operating model gaps.

4. Follow the MLOps Playbook: yes, we need good process and rigour. But companies are layering in “best practice” tooling, pipelines, and controls before they’ve even figured out how to get the first model to production, let alone make it useful. Don’t automate what you can’t yet demonstrate works.

5. Separate the ‘Data People’ from the Business: this one’s subtle but deadly. A lot of companies love to do this. Whether it’s by org structure, funding lines, or physical location, when you create separation, you create translation overhead. That overhead more or less kills delivering value. Data needs to be in the bloodstream, not a separate organ.

6. Buy Another Tool: this is one of my favourites! You’ve got a tool problem, right? Nope, Nada, Non, you are wrong. You’ve got a thinking problem, a design problem, a cultural problem. Throwing another SaaS license at the situation is just kicking the can down the road and confusing the hell out of your frontline teams. Don’t do it!

7. Benchmark Against Competitors: actually, reflecting on it, this is actually one of my favourites, because there is always some special person who thinks they are clever bringing this up! This one always gets applause in the boardroom and results in…guess what…nothing. Because you’re looking sideways instead of forward. You don’t know what your competitor’s strategy is. You don’t know what they’re actually achieving behind the press releases, and their constraints aren’t yours. So why are you copying their moves?

So, What Do You Do Instead?

Start from zero. Not from some templated strategy or borrowed playbook. Wipe the slate clean and ask: what are we really trying to do?

Here’s what that looks like:

1. Start With Use Cases That Matter: now you know me, this is the path to a better future. But, not all use cases are created equal. Find the ones that generate revenue, improve customer retention, or reduce cost in a measurable way. Don’t chase AI for AI’s sake. Chase outcomes that matter.

2. Design the Operating Model Around Decisions, Not Pipelines: how does a decision get made in your business today? Who makes it? With what data? How often? If you don’t know that, you’re not ready for AI. Start by mapping the decision loop, understand the processes and then infuse data into it, end to end.

3. Build a System That Can Learn: data and AI need feedback loops. If there’s no way to measure what works and what doesn’t, you’re just building in the dark. Every model, dashboard, or recommendation should feed into a learning cycle. Otherwise, it’s noise. I’ve designed these feedback loops from strategic objective to outcomes and there are embedded loops in all the stages.

4. Kill the Vanity Metrics: no one cares how many pipelines you’ve built. Or how many petabytes you’ve stored. Or how many people signed up to your dashboard. If it doesn’t drive a business result, it’s theatre. Focus on commercial impact.

5. Empower the Frontline: if the people closest to the customer, product, or operations don’t have the tools and access to act on data at the moment it matters, what’s the point?

6. Invest in Thinking, Not Just Tech: too many executives want to buy their way out of complexity. The simple point here is that you can’t. You must think your way through it. That means slowing down, asking better questions, and building the capability to solve hard problems with clarity and not just cash.

7. Get Comfortable Being Uncomfortable: a real data and AI strategy isn’t about best practices. It’s about bold bets, hard trade-offs, and custom solutions that don’t fit into neat little boxes. It won’t always feel safe. But it will move you forward.

Wrap Up

If you’re following “best practices” without challenging whether they actually fit your business, you’re not leading, you’re outsourcing your strategy to the herd.

This hurts, because most of the herd aren’t going nowhere fast.

You don’t need better practices, you need better thinking, and a model that’s built for your future, not someone else’s past.