What challenges does the AI project bring? | AI in business #65

How to effectively integrate AI project into your business strategy?

Gartner research says that by 2030, 80% of project management tasks will be handled by AI. What the percentage of projects using AI to complete tasks will look like – remains to be seen. However, it is already worth considering how to integrate AI into a company’s operations strategy.

The first step is to understand this technology’s potential and limitations. Artificial intelligence does well at analyzing trends and patterns but fails at things like multistep reasoning and moral decision-making. It creates sensational visuals, but getting it to consistently generate materials that match a brand’s image requires considerable skill. That’s why when we start working on an AI project, we can’t assume with a probability comparable to other projects whether it will produce concrete, measurable results.

A good starting point, therefore, is to analyze the pros and cons:

• What is the total cost of the various stages of AI project implementation?
• What KPIs should be defined to assess the business impact of an AI project?

To achieve a credible answer to these questions, it’s best to select simple AI projects that bring considerable value are easily measurable, and will fit into the company’s strategy. A startup offering courier services can serve as an example. Its goal is to improve customer service and increase supply chain flexibility. A simple but valuable AI project, for example, is the implementation of a chatbot that handles customer inquiries. Such a virtual assistant will handle more requests than a traditional call center, increasing customer satisfaction through quick responses to inquiries and consistent communication quality. In contrast, an advanced system that optimizes courier routes fits the goal of improving delivery flexibility but is complex and has much higher risks.

Once the initial AI projects have been determined, the startup should assess their feasibility, for example, in terms of the budget within which the AI project should fit.

AI project budgeting. Key challenges

Implementing an off-the-shelf SaaS or AI as a Service (AIaaS) solution, or so-called “off-the-shelf AI,” has many advantages. One is the predictable cost of using the tool and the relatively easy-to-estimated cost of implementing an AI project. You can choose from solutions such as:

• chatbot for customer service – such as Intercom Fin, LiveChat from Chatbot.com, Drift or FreshChat,
• Social media analytics to increase the reach of marketing messages – with Cortex, Buffer or Lately, or
• business data analysis with Microsoft Power BI, Tableau, or for less complex tasks – Google Bard, which integrates with Google documents.

For larger-scale AI projects, their costs can often be underestimated. Especially when it comes to the resources and time needed for data collection and preparation. For example, according to Arvind Krishna of IBM, the data preparation stage for AI learning can account for as much as 80% of a project’s duration.

Source: DALL-E 3, prompt: Marta M. Kania (https://www.linkedin.com/in/martamatyldakania/)

Moreover, the more we require personalized AI models for a project, the more qualitative data we need to collect. For example, deep neural networks for learning require hundreds of thousands of examples. This brings up the cost of acquiring and cleaning such huge data sets. Fortunately, the rapid development of artificial intelligence means that more and more AI projects can be implemented without the need for costly learning of a custom model.

Nevertheless, a company planning an AI project should consider not only the solution development stage, but also the preparation of data and the continued operation of the system, including the cost of maintenance, updating, or collecting new data. Only then you can assess the real return on investment in AI.

Data management issues in AI projects. What you should know

A key challenge in AI projects is data – its availability, quantity and quality. So what to do? Before starting an AI project, you need to:

• carefully examine what data the company has – in what form it is stored and where it comes from,
• take care of the infrastructure and develop internal data acquisition processes,
• Consider purchasing external datasets or crowdsourcing if they are in short supply.

A common problem is that data is scattered across multiple systems and formats. It can be challenging to merge them, clean them and prepare them for AI learning. A good practice is for the AI team to work closely with the IT department or data analysts. Together, they should ensure that the right infrastructure and data acquisition processes are in place.

Technical and security challenges in AI projects

AI is not just machine learning algorithms. To make them work in practice, an entire IT infrastructure is needed. Meanwhile, integrating new AI systems with a company’s existing ones can be challenging. It often requires adapting older business systems, which for many companies means considerable upgrade costs.

Besides, AI projects require expertise in data science and data engineering. Meanwhile, the world is seeing a shortage of specialists in this field. According to McKinsey’s “Technology Trends Outlook 2023” report, the ratio of job advertisements to available specialists is 7 to 100, and demand is constantly growing.

The issue of data security is also not insignificant. AI systems process huge amounts of sensitive information, which must be properly secured against leakage. Meanwhile, data breaches have increased significantly in recent years. This is therefore another important risk to keep in mind when implementing AI projects.

Key competencies in AI for entrepreneurs. What difficulties might you encounter?

A common barrier to implementing an AI project can be poor knowledge of artificial intelligence among managers and business decision-makers. Without an in-depth understanding of the technology’s capabilities, it is difficult to assess the viability of specific projects and make sound decisions. That is why it is essential to invest in improving the knowledge of managers in the area of new technologies.

Retraining current employees can also help. There is increasing talk of so-called “citizen data analysts” (“Citizen data scientists”). These specialists exploit cutting-edge technologies to solve specific business problems they face daily They are highly knowledgeable about the industry in which they work. By being part of the team working on an AI project, they enable AI specialists to focus on implementation problems by answering industry-specific questions.

In addition to technical skills, such as evaluating AI recommendations and making decisions, soft skills are also important, including leadership and strategic thinking. This is another way to address the shortage of AI skills in companies.

Analyzing AI project success. How to avoid mistakes when measuring ROI?

There is an unsubstantiated (and probably untrue) rumor circulating on the Internet that up to 87% of AI projects never reach the production phase. While we have not been able to access reliable studies of successful projects, an early definition of ways to measure success is key to assessing the real impact of AI implementation.

A good practice here is a small-scale experiment. It involves testing AI performance, for example, on a random sample of users and comparing the results with a control group using a standard solution. Such an A/B test helps you to verify whether the new AI system can bring the expected results like an increase in conversions or customer satisfaction.

Source: DALL-E 3, prompt: Marta M. Kania (https://www.linkedin.com/in/martamatyldakania/)

A/B testing is worth repeating periodically even after AI implementation, as models can lose accuracy and relevance in solving problems. This will enable you to quickly identify emerging anomalies and the need to recalibrate the system so that it continues to deliver the expected business results.

Summary

While AI offers tremendous opportunities, projects in this field carry significant challenges. To succeed, you have to assess feasibly the costs and benefits of AI, take care of data acquisition and quality, develop in-house competencies and bet on the gradual implementation of new technologies. It is also crucial to measure the tangible business impact of implementations and react quickly to emerging problems. Only then the AI will become an enhancement rather than a threat to the company.

If you like our content, join our busy bees community on Facebook, Twitter, LinkedIn, Instagram, YouTube, Pinterest, TikTok.