Reality Check

AI Project Reality Check

10 questions to find out if your AI project is heading for production or heading for the bin. Brutally honest. Free. No email required.

Your score: 0/30

Answer the questions below to find out if you're building something real or just keeping the cloud providers happy.

1.

Do you know what business outcome your AI project delivers?

Not "AI-driven insights platform." An actual outcome. "Reduce customer churn by 5%." "Cut manual review time by 60%." If the sentence contains the word "leverage," start again.

2.

Have you measured where your AI model is wrong?

AI is a model of reality — not reality. There's always a gap between what it predicts and what actually happens. If nobody's quantified that gap, you're selling magic beans to the business.

3.

Is your ML platform justified by the number of models you run?

If you've built an elaborate KubeFlow or Vertex AI platform for one model that runs once a week, you don't have an ML platform. You have a science fair project with a very expensive poster board.

4.

Could your team run the AI pipeline without the vendor who built it?

The vendor's consultants optimised for their platform's capabilities, not your team's. If they left tomorrow, could your engineers deploy, debug, and modify the pipeline? Or would they just stare at it?

5.

Do you know what your AI infrastructure actually costs?

Not the monthly cloud bill. Cost per prediction. Cost per pipeline run. Cost per environment. If it's all just "the cloud budget" then you can't optimise it, and the DBU pricing matrix is eating you alive.

6.

Is your data governed, or just stored?

Data in the lake isn't data you can use. If your ML model is trained on data that was cleaned by someone who left the company, you don't have a data asset. You have a liability with a Databricks bill attached.

7.

Did you design for your security and compliance environment from day one?

VPC Service Controls, private endpoints, audit logging, data residency. If you started in an open sandbox and you're now discovering what's blocked in production — welcome to the 2-3x development time multiplier nobody warned you about.

8.

Are your AI coding agents amplifying good direction, or confident nonsense?

AI agents are brilliant at the first 80%. They'll scaffold your project, write your tests, and knock out features faster than your team can review them. But they don't know what matters. They amplify direction — good or bad.

9.

Is your "multi-cloud AI strategy" a strategy, or an accident you named?

Running AI workloads on three different clouds isn't multi-cloud strategy. It's three different teams who made three different choices and someone in management drew a circle around it on a slide.

10.

Who designed your architecture — your engineers or an LLM?

Someone had an idea, asked Claude about it, got the requisite attaboy, and now there's a Jenga tower that passes for architecture. Before you know it, the LLM is writing Jira tickets and the people who've honed their craft for years are just implementing its design. When it falls over — and it will — who's carrying the bag?

Scored lower than you hoped?

Don't worry — most organisations do. Book a free 90-minute Architecture Office Hours session and let's talk about which gaps matter most for your situation.

Book Architecture Office Hours