According to Deloitte’s 2020 survey, digitally mature enterprises see a 4.3% ROI for their artificial intelligence projects in just 1.2 years after launch. Meanwhile, AI laggards’ ROI seldom exceeds 0.2%, with the median payback period of 1.6 years.
Deloitte also discovered that companies seeing tangible and quick return on artificial intelligence investments set the right foundation for AI initiatives from day one.
PwC echoes the sentiment, claiming that AI leaders take a holistic approach to AI development and implementation and tackle three business outcomes — i.e., business transformation, systems modernization, and enhanced decision making — all at once.
So, how to use AI in your organization and join the cohort of artificial intelligence leaders?
To answer this question, we conducted extensive research, talked to the ITRex experts, and examined the projects from our portfolio. Here’s what we learned.
How to implement AI in business: a 5-step guide for companies undergoing intelligent transformation
Disclaimer: Innovation for its own sake won’t do your company any good.
Sometimes simpler technologies like robotic process automation (RPA) can handle tasks on a par with AI algorithms, and there’s no need to overcomplicate things.
In other instances (think AI-based medical imaging solutions), there might not be enough data for machine learning models to identify malignant tumors in CT scans with great precision.
And occasionally, it takes multi-layer neural networks and months of unattended algorithm training to reduce data center cooling costs by 20%.
Artificial intelligence is not some kind of silver-bullet solution that will magically boost your employees’ productivity and improve your bottom line. Yet, it has solid potential to transform your business.
Without further ado, here’s your guide to implementing AI
Step 1: Familiarize yourself with AI’s capabilities and limitations
On a broader scale, the use of artificial intelligence in business falls to:
Forecasting (as well as “if-else” analysis)
Process enhancement and automation
Resource management and allocation
This list is not exhaustive as artificial intelligence continues to evolve, fueled by considerable advances in hardware design and cloud computing.
Algorithms that facilitate or take over standalone tasks and entire processes differ in their data sourcing, processing, and interpretation power — and that’s what you need to keep in mind when working on your AI adoption strategy.
Let’s take supervised machine learning, for instance. AI engineers could train algorithms to detect cats in Instagram posts by feeding them annotated images of our feline friends. When faced with unfamiliar objects, these algorithms fall badly short.
But if we take labeled data out of the ML model training process, we’ll get unsupervised machine learning algorithms that crunch vast amounts of information — again, let’s use cat picks as an example — down to meaningful insights. Unsupervised ML models still require some initial training, though. For instance, we could tell algorithms that a particular database contains images of cats and dogs only and leave it up to the AI to do the math.
There’s also reinforcement learning — a technique that involves letting algorithms loose in the wild so that they could propose solutions to business problems and learn from their own mistakes. This type of AI can help summarize long texts or predict stock market trends.
Finally, there are deep neural networks that make intelligent predictions by analyzing labeled and unlabeled data against various parameters. Deep learning has found its way into modern natural language processing (NLP) and computer vision (CV) solutions, such as voice assistants and software with facial recognition capabilities.
No matter how accurate the predictions of artificial intelligence solutions are, in certain cases, there must be human specialists overseeing the AI implementation process and stirring algorithms in the right direction.
For instance, AI can save pulmonologists plenty of time by identifying patients with COVID-related pneumonia, but it’s human doctors who end up reviewing the scans to confirm or rule out the diagnosis.
There are several areas where implementing AI makes little sense without efficient monitoring:
Generating creative content, such as opinion articles and conversion-optimized copy
Coding complex software systems (on a side note, tools like GitHub Copilot and Tabnine can indeed predict and suggest lines of code inside your editor, but we don’t recommend using them unless it’s senior software engineers who use them)
Making judgements and ethical decisions independently
Coming up with innovative, out-of-the-box solutions for real-world problems
If your in-house IT team is struggling to navigate the dynamic artificial intelligence landscape on their own, you could enlist the help of an outside company offering technology consulting services.
Step 2: Define your goals for AI implementation
To start using AI in business, pinpoint the problems you’re looking to solve with artificial intelligence, tying your initiatives to tangible outcomes.
For this, you need to conduct meetings with the organization units that could benefit from implementing AI. Your company’s C-Suite should be part and the driving force of these discussions.
Also, audit your processes and data, as well as the external and internal factors affecting your organization. There are plenty of techniques and frameworks to support your decision making. These include the TEMPLES micro and macro-environment analysis, VRIO framework for evaluating your critical assets, and SWOT to summarize your company’s strengths and weaknesses. Another great tool to evaluate the drivers and barriers to AI adoption is the Force Field Analysis by Kurt Lewin. This list is not exhaustive; still, it could be a starting point for your AI implementation journey.
Experts believe you should prioritize AI use cases based on near-term visibility and financial value they could bring to your company. That’s why you need specific objectives and ways to measure them.
Going back to the question of payback on artificial intelligence investments, it’s key to distinguish between hard and soft ROI.
Here’s the hard ROI your company could achieve from implementing artificial intelligence:
Time savings driven by the automation of laborious tasks
Productivity gains stemming from AI-assisted decision making
Labor and operational cost reduction due to increased automation and employee productivity
Revenue increase thanks to the customer base growth and higher value of provided services
The soft ROI artificial intelligence adoption could provide spans:
Personalized client experience, which positively affects customer satisfaction and loyalty
Skills retention, which revolves around constant research and validation of new AI implementation concepts and contributes to the development of in-house artificial intelligence skills
Organizational and digital agility, which empowers your employees to revamp technology systems and entire workflows in a response to new challenges and opportunities
All the objectives for implementing your AI pilot should be specific, measurable, achievable, relevant, and time-bound (SMART). For example, your company might want to reduce insurance claims processing time from 20 seconds to three seconds while achieving a 30% claims administration costs reduction by Q1 2023.
To set realistic targets, you could leverage several techniques, including market research, benchmarking against competitors, and consultations with external data science and machine learning experts.
Step 3: Evaluate your AI readiness
The artificial intelligence readiness term refers to an organization’s capability to implement AI and leverage the technology for business outcomes (see Step 2).
Once you’ve identified the aspects of your business that could benefit from AI, it’s time to appraise the tools you need to execute your AI implementation plan.
According to Vitali Likhadzed, ITRex CEO and Co-Founder, your AI implementation strategy will rely on five key building blocks:
AI development talent. Do you have in-house IT specialists and subject matter experts (SMEs) knowing how to implement AI — both on the tech and business side — within a timeframe specified in the previous step? If not, do you have a budget to outsource AI development to a third-party or purchase and deploy a SaaS solution? With the latter option, though, you’ll still have to hire AI developers to configure and customize the software.
Software development, procurement, and maintenance costs. Depending on your business objectives, you could opt for a SaaS-based artificial intelligence tool or take the custom software engineering route. Both approaches have their advantages and downsides, such as the trade-off between longer AI implementation cycles and limited customization options. The total cost of ownership (TCO) for AI systems, either bespoke or SaaS-based, will also include vendor and maintenance fees, as well as the price of setting up and operating a cloud infrastructure (more on that later). The cost of SaaS-based data analytics platforms, for instance, could range between $10,000 and $25,000 per year, with licensing costs comprising a small fraction of the final estimate.
Data. AI algorithms are only as good as the data you feed to them. Images, videos, audio files, PDF documents, sensor readings, and other data that are hard to interpret and modify (i.e., unstructured data) comprises up to 90% of all information stored across your company’s IT infrastructure. Locating, aggregating, and preparing it for algorithm training is an essential step towards creating accurate, high-performing AI solutions.
Computing and storage resources. Microsoft Azure, Amazon Web Services, Google Cloud, and other prominent cloud computing vendors provide the resources to train, deploy, and run machine learning models in the cloud. Your data will live in the cloud, too — in a neatly organized data warehouse, in data lakes, or hybrid data storage solutions known as data lakehouses. Tapping into cloud computing services is thus key for the implementation of AI. And you should configure your cloud infrastructure properly — otherwise, the cost of implementing AI may exceed your potential revenue.
Employee training. Even if you partner with experienced AI developers, you’ll still have to educate employees on the new technology so that they can perform their jobs effectively — both now and in the future, when you get close to enterprise-wide AI adoption.
According to Intel’s classification, companies with all the five AI building blocks in place have reached foundational and operational artificial intelligence readiness. These enterprises can carry on with the AI implementation plan — and they are more likely to succeed if they have strong data governance and cybersecurity strategies and follow DevOps and Agile delivery best practices.
If your organization does not meet these criteria, you could partner with a digital transformation services company to upgrade your IT infrastructure and consider AI implementation options.
Step 4: Start integrating AI into select processes and while planning for scale
At ITRex, we live by the rule of “start small, deploy fast, and learn from your mistakes.” And we suggest our customers follow the same mantra — especially when implementing artificial intelligence in business.
Gartner reports that only 53% of AI projects make it from prototypes to production. One reason for this may be companies’ failure to replicate the results they’ve achieved with their POCs in sterile test environments in real life, with AI algorithms consuming data from multiple sources and enhancing different processes.
A pragmatic approach to adopt AI is to have a bigger picture in the back of your mind instead of focusing on isolated proof of concepts (POC) for the selected use cases, even though the latter might look like a low-hanging fruit compared to ambitious moonshot initiatives.
By creating a blueprint for your company-wide AI adoption strategy early on, you’ll also avoid the fate of 75% of AI pioneers who could go out of business by 2025, not knowing how to implement AI at scale.
Also, a reasonable timeline for an artificial intelligence POC should not exceed three months. If you don’t achieve the expected results within this frame, it might make sense to bring it to a halt and move on to other use scenarios.
Step 5: Achieve AI excellence
After launching the pilot, monitoring algorithm performance, and gathering initial feedback, you could leverage your knowledge to integrate AI, layer by layer, across your company’s processes and IT infrastructure.
For this, you need to set up:
A robust data governance framework ensuring secure and efficient data management across your entire company
An integrated data ecosystem for collecting, storing, and organizing information for algorithm training
An AI excellence center where your in-house team will work hand in hand with third-party experts, acquire new skills, continuously improve AI performance, and test new concepts
A foundation facilitating Agile decision making and continuous business process redesign: as AI will enhance or automate more processes within your organization, you’ll need to validate that humans and machines augment and complement each other’s work
The incremental approach to implementing AI could help you achieve ROI faster, get the C-Suite’s buy-in, and encourage other departments to try out the novel technology.
Understanding artificial intelligence is the first step towards leveraging this technology for your company’s growth and prosperity.
To help you get started, we’ve written Business guide to artificial intelligence — an eBook covering all the questions you might have about the technology, from its types and applications to practical tips for enterprise-wide AI adoption.
Just enter your email address in the form below and grab a free copy of your AI implementation plan!