AI experts in Medellín: how to choose and what to ask
How to tell who has put AI systems into production from who only gives talks, what to ask before hiring, and what a first project should cost.
Digital Marketing, Medical Marketing & AI Consultant · btodigital
The question that separates an AI expert from someone who talks about AI is one: what do you have running in production today, and who uses it? If the answer is demos, pilots and presentations, there is no real experience yet.
Medellín has a strong ecosystem, with universities, software companies and an active startup scene. It also has, like everywhere, plenty of people who changed their profile title last year.
Which profile do you need?
They do not all do the same thing, and hiring the wrong profile is the most common reason a project goes nowhere.
| Profile | What they do | When you need them |
|---|---|---|
| Data scientist | Custom models on your data | Prediction, segmentation, forecasting |
| Applied AI engineer | Integrates existing models into your operation | Agents, automations, assistants |
| Strategy consultant | Decides what to automate and in what order | When you do not know where to start |
| Speaker or trainer | Raises awareness and teaches | To move the team, not to build |
The most expensive confusion: hiring a speaker expecting an implementation, or a data scientist for something solved by connecting an API.
What should you ask before hiring?
Six questions. The answers show immediately:
- What do you have in production today and who uses it daily?
- What does it cost to run per month? Someone who has done it knows the figure. Someone who has not generalises.
- What has broken, and how did you fix it? The most revealing answer of all.
- What happens to my data? Where it lives, who accesses it, whether it is used for training.
- What stays in my company when you leave? Documentation, access, the ability to maintain it.
- What part of what I want would you NOT do? Someone who says yes to everything is not thinking about your problem.
What should a first project cost?
Less than most people assume, and that is the most frequent surprise.
The model cost is almost never the problem. To give a concrete figure of my own: a WhatsApp agent I run in production costs between USD 5 and 8 a month in total infrastructure, model, database and hosting included. It qualifies leads, sends documents and escalates to a person when needed.
What you pay for is design and maintenance: understanding the process, writing the rules, connecting the systems and fixing things when they break. If someone quotes you a fortune in licences and little in design, ask why.
What are the warning signs?
- Promises perfect accuracy. No AI system has it; the serious conversation is about error margins and what happens when it fails.
- Does not ask about your processes, only proposes tools.
- Never mentions data. Without organised data there is no project.
- Sells “AI” without saying concretely what it does.
- Cannot tell you what this approach will not solve.
- Their portfolio is screenshots, not systems with users.
And if what you need is not AI?
It happens more than you would think, and a good consultant says so.
Many problems arriving labelled “we need AI” are really undefined processes, disorganised data or a missing integration. Automating on top of that produces faster disorder.
The sensible sequence: sort out the process, make the data accessible, and only then decide which part gets automated.
Frequently asked questions
How do I know whether someone is genuinely an AI expert? Ask what they have in production today, who uses it daily, what it costs to run and what has broken. Someone who has implemented answers in detail; someone who has not generalises.
How much does implementing AI in a company cost? It depends on scope, and less than assumed on contained projects. The model cost is usually small; what you pay for is design, integration and maintenance.
Do I need a data scientist? Only if you are going to build custom models on your own data. For agents, automations and assistants, someone who can integrate existing models is enough.
What happens to the confidentiality of my data? Ask explicitly: where it is stored, who accesses it, and whether it is used to train models. The enterprise plans of the major providers do not train on API data, but it should be verified in the contract.
Where do I start if I have no idea? With a repetitive, contained task that is annoying today. A small pilot with a number measured beforehand teaches more than any long diagnostic.
Is it fine to hire someone from another city or country? Yes for the technical part. For understanding the process and supporting the team, proximity helps quite a bit, especially in the first weeks.
What is the difference between an AI speaker and a consultant? The speaker moves the team and explains the landscape. The consultant decides what to do and implements it or directs it. Both are useful, for different things, and it is worth not confusing them when hiring.
