This article, the first of a four-part series, goes through the initial steps needed when getting started with AI. Specifically, we look at how to pick the right project for your company, understand the effort needed, and do a quick back-of-the-napkin calculation for how much it would cost.
With the incredible speed with which Artificial Intelligence (AI) is developing nowadays, its introduction into established business processes will remain a major trend in the global technology market for years to come.
According to the International Data Corporation report, companies' spending on artificial intelligence will double to $110 billion by 2024. According to McKinsey, 61% of companies that implemented an AI project increased revenues. But most organizations still struggle with setting budgets for and understanding the potential value of AI projects.
Unfortunately, pricing an AI project is challenging because you have to work in a state of constant uncertainty. Basically, you are working against a black box. You know the metric you want to reach and the business logic you will build on top of it but, for example, cannot find out the amount of data or the time needed to build an optimal model until you start working on a project. So, you have an anticipated result but lack any information of what you will need to achieve it. Therefore, understanding how to price AI will massively help you.
In this four-part series, we'll go through how to prioritize AI projects (part 1) and then go into how what costs exist budget (part 2), followed by practical examples of the cost for proofs-of-concept and MVPs (part 3). We will then finish up with a summary and give you some tips on how to get the most for your money (part 4).
Before we start, I should mention that we at Hasty.ai are offering an end-to-end solution for vision AI. These posts should be generalizable to whatever software solution(s) you end up going with, but if you're interested in knowing more about how we can help, check us out.
Today, you can say that there are two different flavors of AI projects. On the one hand, there are research projects that are looking to prove if something is possible, and push the boundaries of the current state-of-the-art. On the other hand, you have applied AI projects. These are projects that are attempting to apply current research to solve a real-world, specific problem, and then deliver that solution to a customer or user.
For this post, we'll focus on applied AI projects, as those are how organizations extract value from AI.
Before estimating the cost of an AI project, the crucial thing you must think of is its value for a company. It might seem quite obvious, but to estimate a reasonable budget, you must first know the upside - how your company will benefit if it builds a proposed AI solution. Also, if you understand the project's value, you will be able to prioritize and even filter them value-wise according to the company's goals.
For applied AI projects, the value often comes in one of two different flavors: increasing revenue or decreasing operational costs.
To make this a bit more tangible: The new AI department for a tractor manufacturer wants to make some improvements and has two different possible AI solutions to choose from. The first one proposes adding AI-based quality control to three problematic production lines. The second requires adding AI into the tractor to control the fertilizer usage automatically, based on the type of crop, the health of crops, and visible pests - reducing the fertilizer used in agriculture. The question is simple; which AI solution should the tractor manufacturer implement?
From the company's standpoint, adding AI-based quality control will reduce costs. However, the production quality in modern factories is already relatively high (90–99+%), so the cost savings might not be as high as the company would like. Depending on how much you can improve quality, let's say you'll save €100k-€2m in operations yearly.
For our AI-controlled fertilizer spreader, integrating AI in the product will solve one of the major existing customers' problems. Thus, a company will get a competitive advantage and a feature that could be sold to the users, resulting in a significant profit expenditure. For the sake of argument, let's say initial market research says that this can be a €20m opportunity in terms of increasing profits annually.
It might seem like adding AI to the tractor has much more significant potential. However, you also need to take into account how much work will go into making it a workable solution. Here, the quality control project is a much easier task because it will be built in a controlled environment without too much variance. Therefore, the manufacturer will get a working prototype quite fast.
In contrast, adding AI to a tractor will require more time and labor because the use case is much wider in scope. You'll have more variance in your data, and you will have to build a model that is more complex. Let's say you'll need to support at least the ten most popular crops to make the AI solution something you can sell. And you will get data from hundreds of different farmers, taken on different phones in different weather and light conditions. That is objectively more complex.
However, if the company is working in an agile fashion, developing MVPs and building functionality using quicker iterations, it can scope its AI product down to one crop or even one variant of the crop. If so, the fertilizer automation project can have the first deliverable in the same time frame as our quality control use case, for a similar budget. To get it all the way to production will, of course, be a larger investment. But to test if the solution will work and if farmers find it useful can be done in one MVP - giving you the knowledge you need to decide if you want to continue.
Any decision for any AI project will come down to risk appetite. Whoever makes a decision will have to figure out whether it is better to work on the project with a higher upside or faster return. To do that, understanding the potential value of an AI project will help you decide what to prioritize and whether something is worth building at all.
So what project do you pick given our example above? You save €500k with your QA project and it's a small win. €500k may sound like a lot at first, but if you're a large company €500k will hardly impact the bottom line.
On the other hand, if you can create a new value-adding service that can bring in €20m+, you will really make a difference. In addition, today, most corporates are expanding their service offering to expand the USP they have to compete with lower-price competitors, so the solution might actually help the company survive in an ever more competitive landscape. You are also directly impacting the strategy of the company going forward, going from being a manufacturer of goods, to offering a combination of manufactured goods and AI-powered services.
For 99% of manufacturers of advanced goods like tractors, it makes more sense to invest in what's strategically relevant, than it is to save costs. Your decision then should be an easy one to make.
Although this scenario is made-up, it's not something we made up out of thin air. With technology as new as AI, any large organization will have hundreds (if not thousands) of possible use cases where AI can help. Often, different departments will either compete for research budget to test different AI solutions or pitch a centralized AI team why their use cases need solving first.
So let's say we have 50 different ideas to pick from instead of these two. How do we prioritize?
In most cases, focusing on using your AI capabilities to deliver value to your customer is the right way to go. Going back to the McKinsey report, adding AI to products and services yields the highest potential. 19% of companies that adopted AI reported Earning Before Taxes (EBITDA) of 10%+ for the business units in question.
Now that we covered where to invest your AI capabilities, the question becomes how to actually calculate an initial budget.
We'll go more in-depth in the following articles, but for the sake of giving you a concrete takeaway, let's do a quick back-of-the-napkin calculation.
Take the estimated return to the company, then divide that with some reasonable number (5–10), and then reduce the budget slightly if it's a high-risk project. This will give you a budget that will fly by most budget review processes.
For budget controllers, approving a budget when there's a concrete outcome of value to the company will be much easier than throwing money into a black hole of AI spending. If you can prove that it will cost less to do so than what you will bring in, it checks a box for every stakeholder in the organization.
In our next article, we'll go through the costs of a vision AI project and give you some estimates for how much different stages of a project can cost (proof-of-concept, MVP, production). Stay tuned if you're interested in that one.
Recently, we also created a "Blueprint for vision AI" where we go more in-depth in terms of modern development practices. You can find that one here.
Only 13% of vision AI projects make it to production. With Hasty, we boost that number to 100%.
Our comprehensive vision AI platform is the only one you need to go from raw data to a production-ready model. We can help you with:
All the data and models you create always belong to you and can be exported and used outside of Hasty at any given time entirely for free.
You can try Hasty by signing up for free here. If you are looking for additional services like help with ML engineering, we also offer that. Check out our service offerings here to learn more about how we can help.
This post is a high-level exploration of the most common ways of implementing image-based deep learning (often referred …
Hasty is a unified agile ML platform for your entire Vision AI pipeline — with minimal integration effort for you.