It’s 2025, more than 2 years after chatGPT has been released. We have had time to go through several cycles of “this changes everything”, “we are all out of a job”, “this is not changing much”, “OK, how do I use this?”, and possibly a few other directions.
By now, some of the hype settled. People who wanted to “create something, anything, with AI right now” did it, and the ones who needed to “add AI to our product, right now” did.
We already have some idea of what works, what doesn’t, and what makes sense as much as we can in a field where things are moving very fast, where the best tools built a year ago are old and rusty.
*Photo by Gabriella Clare Marino on Unsplash *
If you are reading this, I assume you are considering building something involving AI, how you can use AI in your process, or you want to bring to production some sort of proof of concept idea.
Also, I assume you require some technical help for this, and you are wondering if hiring an AI consultant makes sense.
Enter the AI consultant
*Photo by Clarisse Croset on Unsplash *
I happen to be an AI consultant, so you could expect my answer to be a resounding “Yes, of course, you need an AI consultant for everything”.
On the other hand, I happen to enjoy offering meaningful contributions to clients, I don’t like being part of projects where the most obvious thing is some combination of “you do not need me”, “you do not need AI”, and “I am not sure you know what you need”.
Also, I am an engineer, and the most common reply engineers give to any question is, “it depends”.
So would hiring an AI consultant make sense in your case depends on the consultant, and on the case.
Having stated the obvious, let’s try now to make it a bit more useful: what can an AI consultant do for you, and how to best make use of one if you realize you want to hire one?
What can an AI consultant do for you?
It depends on the consultant: some of us specialize in research, on AI business strategy, on integrating AI in business processes, or in AI and ML development. Some consultants can cover more than one area, but generally, we specialize, and we cannot be a specialists in all of them.
So what are the common types of AI consultants?
ML Engineering Consultant
*Photo by Markus Spiske on Unsplash *
These are the technical specialists who help organizations implement AI solutions. They have strong programming skills and deep technical knowledge of ML frameworks and tools.
They focus on:
- Developing and deploying machine learning models.
- Setting up MLOps pipelines.
- Optimizing model performance.
- Implementing best practices for AI development.
- Building scalable AI infrastructure.
- Training technical teams.
AI Solutions Architect
*Photo by Vishnu Mohanan on Unsplash *
These consultants bridge the gap between business requirements and technical implementation. They have both technical knowledge and system design expertise.
They:
- Design end-to-end AI solutions.
- Create technical specifications.
- Plan system integration.
- Ensure scalability and maintainability.
- Oversee implementation.
- Define technical standards.
AI Strategy Consultant
*Photo by Startaê Team on Unsplash *
These consultants help organizations develop and implement their AI transformation roadmap. They focus on the big picture: how AI can create value for your business, what opportunities exist, and how to prioritize them. They have strong business acumen and an understanding of AI capabilities but may not always dhave eep technical expertise.
They work closely with C-level executives to:
- Assess AI readiness and maturity.
- Develop AI adoption strategies.
- Build business cases for AI initiatives.
- Design governance frameworks.
- Guide change management and digital transformation.
- Evaluate AI vendors and solutions.
Industry-Specific AI Consultant
*Photo by Elin Tabitha on Unsplash *
These consultants specialize in applying AI to specific industries like healthcare, finance, retail, or manufacturing. They combine industry expertise with AI knowledge.
They:
- Identify industry-specific use cases.
- Ensure regulatory compliance.
- Adapt AI solutions to industry needs.
- Optimize industry processes.
- Navigate domain-specific challenges.
- Build industry-specific data pipelines.
AI Research Consultant
*Photo by Dan Dimmock on Unsplash *
These consultants focus on advancing the state of the art and solving novel problems. They typically have advanced degrees and research experience.
They:
- Conduct research into new AI techniques.
- Develop custom algorithms.
- Solve complex technical challenges.
- Write technical papers and patents.
- Evaluate cutting-edge technologies.
- Provide guidance on emerging AI capabilities.
Each type of consultant brings different values to organizations.
For example:
- A healthcare organization might need both an industry specialist for compliance and an ML engineer for implementation.
- A startup might need an AI strategist to plan their AI roadmap before hiring technical consultants.
- A research lab might need research consultants to advance its fundamental AI capabilities.
The key is matching the right type of consultant to your organization’s needs and stage of AI maturity. Many projects benefit from a combination of different consultant profiles working together, each bringing their specialized expertise to different aspects of the initiative.
Looking at the list, you can start getting an idea of who you could need and what they could do for you.
As an example, given my background, I am mainly an ML Engineering Consultant, and I have fulfilled roles as an AI Solutions Architect and as an AI Strategist.
Also, I am a mostly technically focused consultant, and I do my best work when I can put together some code to at least check that my assumptions are right.
How to best use a technically focused AI consultant?
Aligning with my experiences, here are the situations where AI consultants can help the most:
- Triaging and consulting on a business and business ideas before any implementation.
- Creating a quick proof of concept (PoC).
- Transforming a PoC into an MVP.
- Supporting expanding the product itself or a team
*Photo by Kelly Sikkema on Unsplash *
1. Triaging and consulting on a business and business ideas before any implementation
Consultants can help triage internal processes and business ideas and find out what can be easily automated and delegated, where it makes sense to use AI, and where it would probably not make much sense.
Experts have an idea of how reliable AI can be in different situations (getting something 90% reliable tends to be easy, getting it to be 99.9% like you’d want in healthcare or finance, not so much), and of how hard it can be to integrate a solution in a workflow.
Having a perfect technical solution that takes more time to use than doing the task ourselves is usually something no one will end up using.
Experienced consultants can, at times, warn about it or at least foresee the issues to be solved.
For this, you can get a consultant, describe as much of your processes and ideas and business, and get an idea of some low hanging fruits, some tasks that can be easily automated.
Also, experts can warn you of the tasks that are either hard or should not be automated right now unless you are an AI research lab.
I am not allowed to give specific examples from my clients, but in general, it works like this: it is a cooperation between the clients, the target teams, the experts in both what to do and what to deliver, and the consultant.
Photo by Cas Holmes on Unsplash
The parties sit down together (or call or write) and describe the different offerings, business challenges, and actions, where a lot of time is spent.
Often, a client has ideas of what to automate and where to offer AI.
Together, consultants and clients try to find out what is possible, how hard it is, how it would work to integrate it with the current workflow and teams, and if it would actually make sense.
As an example, when checking if something is possible, experts can triage it like this:
- Done now: there are off-the-shelf tools that do this, let’s see if they fit the bill.
- Easy (hours-days): there are very well-known techniques to do this, let’s plug them in.
- Medium (weeks-months): it is doable, it will need lots of tinkering and work to make it happen.
- Hard (months-years): this is a research project. Experts think it’s possible, but not sure.
- Impossible (for now).
The goal is to find the most productive business tasks that can be covered by 1 and 2. If the goal is to offer an AI product (not just using AI), consultants can also consider 3, as long as it will stand out from the competitors.
From this, stakeholders generally have an idea of what to tackle: they have a few low-hanging fruits, a few slightly more effortful ones, and an idea of where to start.
When working with a good consultant, this is also where you find out that many ideas are not worth the effort: spending a month developing a tool that saves the work that a human in the team does for eight hours four times a year will not be worth it unless you can sell the service (I have been in more of these conversations than you would expect 🙄).
To check assumptions and make sure that “this is possible and probably easy” is true, at times consultants will also try to implement something, maybe tiny, to check if the technology actually makes sense.
A design-oriented consultant will also mock the product before developing: stakeholders can understand if they would use a product when they play with a cardboard imitation or with a human taking the role of an AI
This brings us to:
2. Creating a quick PoC
Photo by Phil Hearing on Unsplash
Consultants can help take an idea that seems to make sense, that should be technically viable, that has maybe been briefly tested with an off-the-shelf AI tool, and make it work at 1% to 10% capacity as a PoC.
The assumption is that if it works 1-10% of the time the way experts want, scaling it up is just an engineering problem.
This does not mean it is easy. Building a bridge or a house is just an engineering problem. Both take time, can be very expensive, and should be done by the right people with the right tools and the right materials. But experts know it is doable, and that is an assumption they cannot always take with AI.
This should happen fast. A good consultant will do this in a few hours, a few days at most. At times consultants need to implement something to realize that, well, wait, maybe the hype from the research papers does not apply in this case.
I personally tend to think a lot with my hands, and I do not much trust ideas that I have not at least tested a bit in code.
I would recommend you not to trust them either. As Linus Torvalds, creator of Linux, famously said: “Talk is cheap. Show me the code.” If someone can show you that something actually works, you can trust them more.
This is also the moment when an idea meets the “Where is the data?”, “What examples of the result do we have?”, and “Who will evaluate the results?” issues.
Something that AI consultants seem to have to explain several times is that AI is not magic and that experts still need data and examples of what they want (I personally struggle to implement, or ask an AI to do, something I have no idea how to do myself, at least in the outline), and people that will evaluate the results.
It is good to consider it even before starting, but at times a few hours of implementation will allow these issues to surface.
If the PoC makes sense, if the technology is there, if experts have the data they need to start, and so forth, consultants can then help
3. Transforming a PoC into an MVP
*Photo by Daniel Öberg on Unsplash *
In the world of LLMs (large language models, think chatGPT or Claude) and current AI and ML in general, getting a PoC can take a skilled practitioner a few hours or days and can be done by anyone with an online tool in a few days or weeks.
Going from PoC to some sort of production (at least a minimum viable product) is way harder, the tools are still being developed, and most of it is good old software engineering that is faster with AI tools, but not 100 times faster.
See the comment above about just engineering: this is the phase where it is the most valid.
A good AI consultant who has experience as a software engineer can help here, either alone or supporting an internal IT team.
This will involve improving performances, checking edge cases, developing all the bits that are needed when real people use products, and generally will take longer than initially expected.
It is also less shiny, less involving “bleeding edge AI tools,” and more “databases, web servers, UIs, logging”. One will hear about cloud services and provisioning and “requests per minutes”.
Being involved at this stage can be less fun on the “shiny tools” side but satisfying because experts know this product makes sense and they are creating something real.
To make sure consultants can help, it’s important to get someone who knows how to work on shipping barebone (or polished) products.
4. Supporting expanding the product itself
If this is your case, you very probably do not need me to tell you how to make use of an AI consultant: you have a working product, shipped or almost shipped, and there is some technological area where you need support.
AI consultants can help if that is one of their strengths: it could be MLOps, it could be squeezing the models used on smaller devices, or it could be something to do with security, reliability, and speed.
If you have a technical need to be filled right now, maybe the right AI consultant can help. Again, it depends: in this case, a lot. The more ahead in the pipeline, the more specialized skills you will require.
Sure, in many cases, they can be acquired: if your company uses a different deep learning framework than I am used to, it will take me just a little while to catch up, and I will not be the person to ask for deep enhancement for the first month or so.
If you hire an AI business consultant and want them to re-train a GPT4 level LLM from scratch, that will take much longer (if it will happen at all).
What an AI consultant cannot do
*Photo by 愚木混株 cdd20 on Unsplash *
I really want to stress that AI is not magic, even if the most common icon for “use AI in this product” is a magic wand. It feels a bit like magic, we have machines that we can talk to and have them do things for us without having to spell it out in code: but they still need data, and they still have capabilities and ways in which they go wrong.
Experts cannot work without data. At least the data of “from this input, we want this kind of output”. Consultants cannot make people adopt a product.
Some consultants are more business oriented than others, some have marketing skills, but in general, the client will be the one to decide what can be useful, or not. Experts can make it clear how much effort (read: time and money) that could cost. Like any other consultant and contractor, they will probably be off by a factor of 2 or more, so budget and multiply it by 2 at least. But consultants can offer an order of magnitude: if they think something will take a month, it will rarely take days, and probably, hopefully, not years.
Experts cannot implement what is not clear at all. They can participate in the exploration and help clarify: but before starting to build, they need at least some sort of direction.
How to tell if someone is a good AI consultant
*Photo by Museums Victoria on Unsplash *
Good question. Generally, the best idea is to have someone technically skilled to evaluate them. If you know someone you trust and who proved to be reliable and competent, they can help you find and evaluate others (I am expecting you cannot hire them directly. If you know someone you trust who is competent and reliable and you can hire, do it).
If you have to do it yourself, explore some of these questions:
- have them talk about what they built, what the challenges were, what they would do differently now
- ask them to describe where AI does not make sense. Give them an overview of your situation, and ask them where you should NOT use AI.If they think you should always use AI, or always use tech, they are not aware of the trade-offs, and they are either inexperienced or not very ethical
- bring some ideas, and ask them to rank them from good to bad, from easy to implement to hard
- if everyone is OK with that, hire them for a few hours to build a quick demo or PoC. I tend to tell people to “date before marriage”. It is very, very hard to evaluate all the technical competence needed before starting a project (even huge technical companies struggle, even techies struggle), and it is even harder to evaluate how it would be to work together.
For example, freelancers are easy to test: you can hire them for a few hours. If they pass the items above, this would be the best next step.
What to do before hiring one?
Try things out: ChatGPT, Claude, etc, are powered by the best technologies available. Go and test ideas. See what they can do. If you can get them to do what you need, like 80%, you are good to go.
They are also fun tools to use. Experiment and play a bit.
Now what?
Photo by Xavi Cabrera on Unsplash
Get ideas, play with ideas.
See what you could do better if you had a tool that is almost as intelligent as a human (at times; at times more intelligent, many times much less), see what would allow you to scale yourself or your business.
If something stands out, consider hiring an AI consultant. And if you need assistance, let’s talk, we are happy to help 👋
Need help with finding the right talent? CodeControl can support you either with remote hiring or with finding top tech freelancers.