A finance director I know signed up for a machine learning course at a top business school. Twelve weeks, roughly £9,000, taught by a professor who researches reinforcement learning. She passed. She then went back to her team and still couldn’t automate a single reconciliation task, because the course had taught her to tune neural networks rather than to fix a reporting workflow.
The course wasn’t bad. It was built for someone else. That mismatch is where most corporate and personal AI training budgets quietly disappear, and it’s the thing worth solving before you compare syllabi or prices.
Start with the task you want to stop doing
Before you open a course catalogue, write down two or three things you’d like to hand to a machine by next quarter. Summarising supplier contracts. Drafting first-pass responses to support tickets. Pulling numbers out of scanned invoices. Cleaning a messy spreadsheet of customer records.
That list does more work than any brochure. A procurement manager who wants to extract clauses from PDFs needs retrieval, prompt design and data hygiene. A data analyst who wants to forecast demand needs Python, SQL and a decent grounding in statistics. Those are different courses, and a general “intro to AI” module serves neither of them well.
Once you know the task, you can rule out about 70% of what’s on the market in an afternoon.
The four formats, and who each one actually suits
Most AI courses for professionals fall into one of four buckets. They cost wildly different amounts and demand different levels of self-discipline.
- Self-paced video libraries (Coursera, Udemy, vendor academies). Cheap, usually £15–£80 on sale. Great for picking up syntax and definitions. Terrible if you need accountability, because roughly 10–15% of learners finish them.
- Cohort-based online programmes. Six to twelve weeks, live weekly sessions, £800–£4,000. You get deadlines, peers and someone who notices when you go quiet. Best for people who already know they procrastinate.
- University and executive education. £5,000–£15,000 for a short course, often with a certificate carrying a recognisable name. Worth it mainly when your employer is paying and the credential itself opens doors internally.
- Vendor and platform certifications. A few hundred pounds per exam. Narrow, practical and increasingly listed in job adverts. If your company runs Salesforce, Microsoft or SAP, this is usually the shortest route to visible value.
There’s a more detailed breakdown of how to judge the online options in this guide to online AI classes that actually teach you something, which is worth reading before you hand over a card.
What separates a real course from a PDF with a certificate
Half the “AI courses” advertised to professionals are slides, a multiple-choice quiz and an automated certificate. That isn’t necessarily worthless, but it shouldn’t cost £1,200. Look for three signals.
Check the capstone, not the module list
Every syllabus says “prompt engineering” and “ethical AI”. Very few say what you’ll build in week ten. Ask for a sample of last cohort’s final projects. If the answer is a case study write-up, you’re buying reading material. If it’s a working tool that pulls from a real database and handles errors, you’re buying skills.
Find out who marks your work
Feedback is the expensive part of teaching. A teaching assistant who returns comments on your code within four days is worth more than a famous lecturer who never sees your submissions. Ask the ratio of learners to reviewers, and whether the review is human or automated.
Confirm the tools are current
AI tooling turns over fast. A course recorded in early 2023 will still teach you sound fundamentals of regression and classification, but its sections on language model APIs may be two generations out of date. Check the last update date, and look for anything mentioning retrieval-augmented generation, evaluation and cost control, which are the three things practitioners spend most of their time on now.
There’s a fuller checklist in this piece on how to tell which AI training courses are worth your money.
Free platforms are often the fastest route, not the cheapest
Professionals tend to assume free means thin. In AI, some of the strongest beginner material is free and hands-on. LearnAI.org, for instance, is built around getting you to ship small working projects rather than watch someone else do it, which is exactly what a busy adult needs in the first three weeks.
A sensible sequence for many people is: two or three weeks free to confirm the subject holds your interest, then paid and structured once you hit the point where you need feedback to progress. Paying first is fine if your employer is footing the bill. Paying first with your own money, before you’ve tested your appetite, is how unopened subscriptions happen.
Domain-specific training beats generalist courses
If you work inside a large enterprise system, general AI courses will leave you stranded between theory and your actual toolset. Someone in an SAP-heavy organisation gains far more from platform-specific training than from a generic deep learning module, simply because the data, permissions and deployment paths are already defined for them. This overview of the SAP Learning AI Hub is a good example of what that looks like when it’s done properly.
The same logic applies to data platforms. If your team already runs on Databricks, a certification tied to that stack turns theory into something your manager can approve a budget line for. The costs, exam structure and realistic time commitments are set out in this look at Databricks Academy courses and certifications, and the honest answer is that the exam fees are modest while the study hours are not.
Fitting eight hours a week around a fifty-hour job
Every professional course claims to be designed for busy people. The realistic number is six to ten hours a week for a demanding technical course, and you should assume the upper end. Two weekday evenings plus a Saturday morning is a schedule that survives contact with a real job; five one-hour slots scattered across the week usually doesn’t.
Use AI itself to compress the boring parts. Summarising documentation, generating practice questions, explaining an error message at 11pm: that’s a legitimate use of a study partner, and treating the model as a tutor rather than an answer machine roughly halves the time lost to dead ends. There’s a decent walkthrough of using Meta’s AI as a study partner if you want a starting point.
One more practical note: tell your manager what you’re learning and ask for a small, real problem to apply it to. Courses that stay theoretical fade in about six weeks. Courses attached to a live task don’t.
What you should be able to show at the end
By the final week, you want artefacts, not just a certificate. A working script or small application. A written note explaining a trade-off you made and why. A short demo you could give in a team meeting in under five minutes without notes.
If you finish a twelve-week programme and can’t point to at least two of those, the course didn’t fail you — it was simply the wrong purchase. Ask about the outputs before you enrol, and you’ll avoid the expensive version of learning that leaves you exactly where you started, only with a PDF to show for it.