- Global site
- Africa
- Australia
- Austria
- Bulgaria
- Czechia
- Denmark
- Estonia
- Finland
- France
- Germany
- Hungary
- India
- Iceland
- Indonesia
- Kingdom of Saudi Arabia
- Kuwait
- Latvia
- Lithuania
- Malaysia
- Middle East
- Netherlands
- Norway
- Oman
- Philippines
- Poland
- Portugal
- Qatar
- Romania
- Serbia
- Slovenia
- Singapore
- Spain
- Sri Lanka
- Sweden
- Switzerland
- Ukraine
- United Kingdom
- United States
De-risking your AI project
- This guide helps you understand the fundamental elements behind successful AI projects that deliver business value and provides practical tips from experts that have delivered hundreds of AI solutions across a variety of industries.
AI has been touted as the solution to many existing challenges across the private and public sectors, as well as the catalyst for innovation. However, up to 85% of AI projects fail, which means billions of dollars are wasted each year on solutions that never deliver any value.
Exploring AI requires an investment, and therefore it’s important to understand the ways you can push the boundaries to de-risk your project to become one of the 15% that succeed with their project. When defining success of an AI-powered solution, it’s important to highlight that success in this context means delivering measurable business value.
AI contributes to digital transformation of an organization by for example automating manual tasks, but when developing an AI-powered solution, the actual problem must be identified and clearly formulated - not simply take an idea that has proven successful in an isolated scenario or based on a high-level bullet point on a strategy slide.
Another common pitfall is to underestimate the value of high-quality data, appropriate tools and a solid business case that facilitate the seamless collaboration between data scientists and engineers.
If your organization has serious plans to leverage AI solution that drive continuous business value, our recommendation is to build a fit-for-purpose data and ML infrastructure. To ensure that your organisation’s data estate can power AI solutions, it’s important to assess data platform needs and design a solution architecture around the identified use cases, implement a data platform that scales, as well as continuously optimize running costs and performance.
In this guide, Crayon’s data and AI experts share their best tips and practices for how to hit the ground running with AI, based on hundreds of successful customer projects across large variety of industries and organizations with their unique challenges.
The basics:
Understanding the nature of
AI investments vs. traditional technology investments
IT decision makers and senior executives are used to investing in technologies, especially software. AI solutions however differ fundamentally from software, and it’s important to be aware of the differences.
In traditional programming, a programmer defines and writes the logic that the software should execute. The software is programmed to produce an outcome based on a given input. With Machine Learning (ML) however, the logic of the software is determined or “learned” from the provided data as input – this is what we call ML models.
Difficulties of evaluating AI model feasibility early-on. Another significant difference is that in traditional programming, the output is mainly dependent on the quality of the written code. With AI software solutions, while the quality of the code is still important, the output is also highly dependent on the volume, quality, and type of the input data, especially the data used to “learn” the ML model.
The difficulty of knowing if a solution will bring business benefits until a prototype or a PoC is built is another fundamental difference between ML and traditional programming. This highlights the need to understand the importance of iterations and feedback loops when exploring, developing and deploying ML solutions.
How to de-risk AI investments
Define a clear business problem
Perhaps the most typical risk associated with AI projects is about how the project scope is defined. It is crucial to start with a clear business problem that needs be solved – failure to do so can result in a solution that doesn’t necessarily address the actual problem. This leads to limited, if, any, value to the organization.
The best way to start shaping a business problem is by first understanding the end-users’ perspective and their pain points. Being able to articulate what is the issue and its impact, where and why it occurs is key to narrowing down the problem you are trying to solve.
Here are two example business problem statement for two AI projects: one that will most definitely fail and one that might have a chance to succeed:
- A business problem that is poorly defined
“Our sales numbers are not looking great, and we need to do something about it. If we could forecast how many items per product we will sell each month we would be able to make more data-driven decisions and, therefore, positively impact our sales KPIs.”
- A business problem that is well defined
“Our sales numbers are not looking great. After a closer investigation of our online ordering process based on customer interviews and using the tracking and order data, we determined that many customers spend a lot of time looking for a specific product and many of them either don’t add anything to the cart or don’t complete their order. Moreover, this is happening despite the new campaign we launched that doubled the web shop visits. To address this, we want to help our customers easily find the product that they are interested in and do that faster and more accurately than any of our competitors.”
Set clear objectives for an AI solution
The success of an AI-powered solution should never be defined and measured purely in technical terms, e.g., the accuracy of the ML model should reach at least 85%, but also, and even more importantly, the success should be measured in (expected) business value. The ROI (return on investment) for an AI solution is significantly easier to determine and calculate when the business problem has been clearly identified. For example, if the problem that we are addressing is the high costs and impact of human error that is caused by an activity that is currently done manually (e.g., inspecting product quality), the ROI can be calculated by understanding to what extent that manual activity can be automated and its accuracy improved, and what that means in terms of saved costs.
However, not all cases are quite as simple to assess. This is especially true when you’re building a new innovative service, like for example a recommender engine for an existing online shop. In this case, it is difficult to establish and measure the difference and impact of your new service, so the ROI calculation would have to be based on a set of assumptions formulated in measurable KPIs and objectives. These assumptions then can be validated when the first iteration of the solution is ready. For the example of the recommender engine, one might define the following KPI and objective: increase purchase conversion rate by at least 15%.
Get your data in order
In many cases, an AI journey starts with a modern data and ML infrastructure. There is a general consensus that high-quality data is the key asset for build well-performing AI models.
Poor data quality is one of the top reasons why AI solutions fail to deliver business value: Gartner has identified it as one of the top 3 barriers to AI adoption. Addressing data quality early on will significantly increase your chances of delivering measurable value with your AI-powered solution.
When data scientists explore data to build models and validate hypotheses for a given business problem, they typically experience one or more of the following challenges:
(a) data is often hard to access because it is buried in complex systems; | |
| |
(c) storing large and diverse volumes of data using a technology not fit to support your analytics and AI use cases, which is typically costly to revert or change; and | |
|
Leverage existing cloud technologies
To put it briefly, how you set up your tech stack has a high impact on how fast you can develop AI-powered solutions. In some cases, it makes sense to build tools in-house that address your organisation’s unique needs, especially when there are sufficient resources for maintaining and further developing these tools. However, leveraging existing, proven cloud technologies and services often help speed up development of AI-powered solutions and is therefore a good strategy for many organizations.
While there are many cloud services available, awareness about their existence and capabilities is in some cases lacking, or the reason behind preferring in-house built solutions can be related to reluctancy to end up in a vendor lock-in situation. The benefits from leveraging cloud technologies however often outweigh the possible risks, and therefore it’s worth considering them to accelerate time to market with AI-powered solutions.
Get your house in order
Perhaps the least discussed aspect of succeeding with AI is the organizational perspectives. Setting your organization up for success with AI requires that you not only consider data, technology, business and strategy, but the people as well.
One of the top reasons enterprises state as hindering AI and machine learning adoption is the skills of staff – according to Gartner, 56% of enterprises state this is limiting their AI maturity. With this number in mind, it’s not surprising that an increasing number of organizations are accelerating their AI journey and de-risking their AI projects with the help of external experts that have built a track record of successful AI solution deliveries.
There are other organizational aspects to consider as well. Diverse teams, executive sponsorship and, as Gartner perhaps surprisingly points out, fewer proofs of concepts are shared characteristic of organizations that have had success with AI. Especially mixed-role teams for AI initiatives have a strong correlation with making AI an integral part of the business. Another factor that contributes to a higher chance of success is C-level sponsorship and a dedicated budget for AI.
To set up your organization for success, you will need to involve the right stakeholders, foster internal champions, have subject matter specialists on the team, as well as the ability to identify competence gaps which can be addressed with external advisers.
Crayon’s 4-step framework for AI powered solutions
Crayon has a successful track record of delivering AI-powered solutions to organizations across a myriad of industries. To ensure AI-powered solutions deliver measurable business value, Crayon has developed a 4-step framework that covers the entire journey of an AI-powered solution from an initial opportunity assessment to continuous AI model management. When delivering custom AI-powered solutions, individual customer needs and strategy are always at the core of the process.
Opportunity assessment
Perhaps the most crucial stage for ensuring the success for your AI-powered solution: opportunity assessment. This is the first step on your journey towards impactful AI solutions, as it requires you to articulate which problem you want to solve. Challenges that arise later when developing an AI-powered solution, often stem from lack of clarity when articulating the pain point or business problem the solution should solve.
As many AI-powered solutions are perhaps understandably approached from a technical perspective, a common pitfall in this crucial stage is to begin defining the solution instead of focusing on the problem. This can lead to significant investments being made without a clear understanding of how the business case for the AI-powered solution can be built.
At this stage, also existing data is assessed to ensure there is data that can be used to address the problem. The Crayon team will also support you with assessing and designing the data infrastructure required for the development of the desired AI capabilities.
(Callout) The output of the first stage of Crayon’s AI framework is an assessment report, which summarizes the key findings as well as provides actionable insights which can be used for internal alignment around the AI project and it’s goals.
AI solution validation
AI is a powerful way automate tasks, increase the efficiency of processes and explore new business opportunities, but before investing in the development of the solution, it’s key to validate the business value and technical feasibility of the solution. In Crayon’s AI framework, this is done by Exploratory Data Analysis (EDA) on the relevant data identified in the opportunity assessment stage, and developing a prototype of the AI-powered solution.
A prototype is a powerful way to validate the initial hypothesis of the solution addressing the business problem, as well as the performance expectations, for example when the goal is to automate manual tasks.
(Callout) The output of the second stage of Crayon’s AI framework is a solution prototype, which can be used to validated the business case for the solution.
AI solution in production
Following a successful prototype and any required adjustments to the data model, it’s time to extend the prototype to a production-ready AI solution and deploy it into production. As the hypothesis and performance of the solution have already been validated in phase 2, the solution will begin delivering business value after it has been deployed.
While an increasing number of organizations have in-house data scientists and engineers, they often lack the needed skills and tools to bring scalable solutions into production. A key element for succeeding with AI-powered solution is tools that facilitate the collaboration between data scientists and engineers. MLOps, or Machine Learning Model Operationalization Management, is a set of practices that supports in achieving the goal of successfully deploying and maintaining machine learning models on production. It supports the collaboration between data scientists and engineers, and enables teams with the right solutions that bring AI solutions into production and manage the solutions.
(Callout box) The extent of challenges related to lack of tools is visible in Gartner’s research: they found that only 53% of projects make it from prototype to production, with lack of tools for creating and managing production-grade AI pipeline as the key reason behind this.
(Callout) The output of the third stage of Crayon’s AI framework is an AI-powered solution in production. By leveraging the expertise and experiences of Crayon’s data & AI team, you increase your chances of succeeding with your AI-powered solution and accelerate time to business value.
Managing your data model for always up-to-date solution
An AI solution in production will only deliver value as long as it remains relevant, any maintenance needs are addressed without delay and the continuous running costs of the data platform are optimized. This can be achieved by automating the model monitoring and maintenance.
Downtime can be minimized by automating model monitoring and management towards timely identification and notification of incorrect results, as well as automatic handling of these issues. This also ensures that the end-users of the AI-powered solutions are continuously provided with relevant results, and value creation to the organization.
(Callout) The output of the fourth stage of Crayon’s AI framework is a monitoring solution that ensures your AI-powered solution continues to deliver measurable business value by remaining relevant and reducing downtime.
How Crayon can help you with AI-powered solutions
Understanding our customers’ business is at the heart of our customer-first strategy and way of working. We leverage existing cloud technologies to speed up the development of AI-powered solutions. Our close partnerships with leading public cloud vendors means we can support you regardless of your cloud strategy; whether you are currently migrating to the cloud, have deployed a hybrid, single-cloud strategy or a multi-cloud strategy. We build custom AI solutions and data platforms and leverage existing technologies to speed up development. Learn more about some of the successful data & AI projects we have delivered over the years.