How Much Do AI Projects Cost?

When considering starting with AI, you surely wonder how much money this undertaking will cost you. Is it millions of dollars and years of development? Or a few thousand bucks and a few weeks of work? There is one important thing you have to be aware of here: it depends – and on a number of factors. Estimating an AI project’s cost without knowing any details is pointless, you can just as well guess a random number. Today, we’ll try to have a closer look at the costs related to AI projects – but it’s still not an accurate estimate for a given project. Don’t expect exact figures – the numbers part is very individual and reliable estimates are only available when details on the project are known. 

What we’re looking at in this article is the type of project that utilizes machine learning to solve a given business problem – a custom solution, not an off-the-shelf one. When it comes to off-the-shelf solutions, there are some that are great: take OCR or some instances of face recognition, where you don’t need your own data set to train the model. But if the problem you’re trying to solve with AI is about the processes within your company – like sales, marketing, pricing, recommendations, etc. – your data (specific to the use case: about your company, customers, employees, etc.) is a crucial element of the project.

Because so many factors influence AI projects – like team composition, use case, data, selected model, and more – it’s impossible to provide a fixed price estimate for a large AI project. With smaller projects, like a PoC, that is possible, but remember that there are two things to estimate at the beginning of the project: 1) the price and 2) the delivery time. 

Here are some things you need to remember when trying to estimate your AI project:

1. The fact that company X developed their model in 6 weeks doesn’t have to mean your model will be developed in the same amount of time – so consult experts who build models, not people with limited AI know-how.
2. Team composition is tricky. If you haven’t worked with data scientists before, you may have difficulties assembling your team quickly – and time is money. 
3. The project will require several iterations. You can’t get it all right the first time around, so you’ll have to keep improving the model.
4. The engineers/data scientists working on your project need to understand the business logic behind it. Otherwise, they won’t be able to develop a model that helps your business.
5. You may be lacking data and need time to collect it or obtain it from third parties.
6. The real “AI work” is the last part of the project, after you build the use case, create a data strategy, and preprocess data. All these preparations take time.

Pricing an AI project – a how-to guide

It’s funny, but the pricing of the project doesn’t start with numbers at all. Sure, it’s good to know that your budget is between X and Y, or maybe just X and no more, but in general, you’ll get to the numbers at the very end. The good thing is that the steps you follow when estimating the cost of your project will help you outline some general information for your data strategy. Let’s reuse the work you’ve done, it can’t be wasted!

Step 1: Scope

What does the project consist of? What’s the goal? What are the success and failure criteria? What data are you going to use?

When the use case is identified, you should define the scope of the project. It can get tricky and you may not realize some things should be within the scope – but you can use the help of an AI advisor or a data science partner. Together with an expert, you will also be able to prioritize work and estimate given tasks in hours to be able to better evaluate how much a project will cost. With that, you move on to step number 2.

Step 2: Action items

When you create a data strategy, it should include core activities and action items. Those will be assigned to given team members to make sure that everyone knows what they’re responsible for. At this point, you can also separate the non-AI work from the AI work. If you can do something yourself, like retrieve the data, do it. That’s because a clear division of work will help you move forward faster – and at this stage, estimate the timeframe of your project and the cost of the work that has to be done. Divide the tasks wisely, taking into consideration the skills and capabilities each team member has. Efficiency is a money-saver.

Step 3: Team

Action items won’t be done without a team. Do you want to build an in-house data science team or outsource? Both approaches have their pros and cons – you can read more about that in our article about building your data science team. If you want to build an in-house team, you have to account for recruitment and training costs – which, reportedly, can be about 15 thousand dollars. Then, you have to remember that you have to pay their salary which on average is 120K dollars yearly according to Glassdoor. In this case, you estimate your project counting the months of work needed for the project – since the salaries are the same every month. If you outsource, you will most probably be charged on an hourly basis – so you will pay for the actual time spent on your project. The hourly rates of outsourced data scientists vary and can fall anywhere between 40 and 300 dollars per hour. Or more, depending on the skills, experience, location, etc. The number of hours of your project depends on how much preparation has to be done, how complex the solution is, what output is expected, and so on. 

Step 4: Consider other costs

Is the cost of a data science team the only one you’ll have to pay? You have to consider other related costs: infrastructure (e.g. cloud, data storage), integration costs (e.g. API development, documentation), and maintenance. An AI project doesn’t end with building the model. The costs of production (infrastructure, integration, maintenance) are often overlooked, but they’re there, and they’re not avoidable. Depending on the complexity of the model and technology stack, these costs don’t have to be huge. However, you need to remember that your model needs continuous support, that new data has to be cleaned and annotated, and even if much of the recurring work is automated, the model can’t be left unsupervised.

Read also: How Much Does It Cost to Use GPT Models? GPT-3 Pricing Explained

Bad news now: the costs of AI projects vary from a few thousand dollars to a few hundred thousand. I wish I could estimate your project magically without having to look at it, but it’s simply impossible. For that, it really is necessary to have a closer look at your needs, requirements, and possibilities to assess the complexity of the solution and the work that it requires. But wait, don’t lose hope. You don’t need a million-dollar budget to start with AI.

That seems to be a lot of money… Can I start on a smaller scale?

The good news is that starting on a small scale is actually recommended. That’s because AI has to prove its worth – it needs to be a viable solution that improves something in your organization, solves a real business problem. So you choose one segment to work on – say you want to stop customers from leaving your company, so your model will be churn prediction. End of story. Don’t overcomplicate things, start with one small project to test your assumptions, learn how to work in a more data-driven way, and mitigate the risk of wasting time and money. So you create a piece of your data strategy, a project roadmap, you identify action items and define success and failure criteria, and the work begins. After a few weeks, the first model is ready and you can see how it works. Sounds cool, right? And more good news: a pilot project like that will usually cost between 5 and 20 thousand dollars, and take only a few months to deliver.

Because AI projects rely on good alignment of business and tech, it’s safer to start small. That’s also the reason why we developed our AI Sprint to get our clients on board with AI. The AI Sprint was designed to help companies quickly gain a deeper understanding of the opportunities and limitations of AI, identify the right use case, create a data strategy and an action plan. During a 2-day AI Sprint workshop, we work closely with the client’s team to understand their business, the processes, and the requirements, to share our knowledge on the tech side and provide advice on the next steps. After the workshop, we start developing a PoC of a chosen model that will deliver quick wins – this way, we make sure the delivered project is viable, that it supports the client’s business, and brings tangible value. 

How much AI projects cost is still unknown…

Well, not totally unknown, but difficult to estimate, especially if you’re not sure how your project will be evolving and what the development of it will include. And it’s fine not to know these things if you’re just starting with AI – that’s why we recommend consulting specialists or companies experienced in AI development. They will be able to scope the project and evaluate how much work is required and what other resources are needed: do you need to buy data from third parties? Do you need to pay for cloud? How much will the maintenance cost be? There’s a variety of factors that need to be taken into consideration when pricing an AI project, so it’s not an easy task. Luckily, that doesn’t mean you need an endless budget to keep spending more and more money on the project. Just start making a small first step and validate the idea of AI in your organization before you move forward with company-wide AI adoption. 

Summing up, the costs of an AI project can be drastically different, depending on the size and complexity of the project, the required resources, the quality of data, and the data science team. Before jumping to any conclusions, make sure you have the basics covered: identify the area you want to improve with AI, set up the goals: both long-term and quick wins. Then build and train the models and put them to work. With a small project proving its worth, it’s easier for you to plan for a bigger scale with a bigger budget: you already know what AI development looks like and how to work in data science projects.