What to prepare before starting AI

Clear objective first

Before you start any AI work, be specific about the outcome you expect the system to produce. "AI" is an umbrella term: a successful project begins with a concrete, measurable workflow you want to improve (for example: route incoming customer questions to the right team, flag likely errors in invoices, or summarize meeting notes for action items). Define the decisions the AI will influence, who will review its outputs, and how you will measure success. That scope-focused approach keeps early work cheap and testable instead of building a large system that only demos well. See related guidance on when AI belongs in product workflows in our article When AI belongs in your software (and when it does not).

What you need to understand about "what to prepare before starting AI"

You do not need a PhD to begin practical AI work, but you must understand two things: how AI produces probabilistic outputs, and what data the model needs. As Coursera recommends, non-technical stakeholders benefit from a conceptual grounding — for example, Andrew Ng's "AI for Everyone" explains that models find patterns in data, not facts, and that outputs are probabilities rather than certainties. Beyond the concept, plan practical preparation: inventory sources of relevant data, check data quality and ownership, and document access pathways and privacy constraints. If you or your team need technical upskilling, Coursera also outlines an incremental learning path from basic Python and math through machine-learning frameworks that helps match expectations to capabilities (coursera.org).

But I’m about to go to medical school — shouldn’t I be preparing?

If this question is a stand-in for major life or role changes, the short answer from experience is: prepare deliberately, not obsessively. As a personal-planning example, a commentary for incoming medical students recommends using transition time to rest and do a few things you enjoy rather than trying to front-load every possible skill. That same principle applies to AI projects: preparation matters, but overcommitting to learning every tool or rewriting all systems before you have a target use case wastes time. Treat early preparation like getting ready for a new role: get confident in the fundamentals, document what will matter on day one, and leave room to learn on the job (ProspectiveDoctor; DAVRON).

So what should I be doing right now?

1. Define the problem and success criteria. Replace vague goals like "use AI" with workflows and measurable outcomes.
2. Audit your data. List where the data lives, its formats, obvious quality issues, and legal or privacy constraints.
3. Pick a small pilot. Ship a minimal workflow that runs end-to-end rather than a large prototype that only looks impressive.
4. Decide human oversight and verification. Make outputs inspectable and keep people in the loop where quality matters.
5. Build an integration plan. Identify where the AI will connect to existing systems (CRM, finance, internal tools) so the pilot can scale without a full rework.
6. Assess skills and learning needs. For non-technical leaders, take a short course that explains concepts without heavy math; for engineers, follow a staged upskilling path toward frameworks and model evaluation (coursera.org).
7. Prepare compliance controls. Document data retention, consent, and audit trails before you move to production.

These steps are practical to apply in any organisation in Canada or the United States. If you want help turning a pilot into production or choosing which workflow to address first, our AI & Intelligent Automation page explains how teams typically move from problem discovery to a running workflow, or contact us through /contact/ to discuss a scoped discovery.

Guidance current as of 2026.