The stakes of Artificial Intelligence (AI) are rising every day. Early adopters have reaped rich dividends, with the possibilities limited only by imagination. But implementing AI is tough. The McKinsey State of AI 2020 reveals only 22% of enterprises extract quantifiable value from AI. Even when convinced of AI benefits, enterprises often do not commit investments. Here are the key reasons plaguing the adoption of AI in industries and how businesses can roll out AI in cost-effective and viable ways.

1. AI requires sizable upfront investments

Businesses cutting across sectors and sizes are gung-ho about Artificial Intelligence (AI) until they get the bill. The biggest reason enterprises dither on AI projects is the sizable upfront investments required to launch AI. Enterprises sometimes do not take up even sure-shot positive ROI projects, as they do not have funds for capital investments.

The exact investment needed for Artificial Intelligence projects depends on multiple factors, including the scope and extent of the project and the disruption the new project causes. As a ballpark figure, it costs about $45,000-$60,000 on average to replace an existing system with AI.

2. Difficulty in quantifying ROI

There is a strong school of thought that ROI is not the right metric to evaluate the success of AI projects. 

Many Artificial Intelligence projects deliver tangible benefits, such as time savings, cost savings, or productivity boosts. But in many instances, AI only has “soft ROI.” The AI model may improve employee satisfaction, make the enterprise more agile, or enhance brand reputation. But such benefits are indirect and hard to quantify.

For instance, it is easy to quantify the ROI of an AI model that improves click-through conversion by a percentage point. But in another instance, the AI model may give warnings for equipment repair. The model prevents downtime and preempts huge losses, but it is hard to quantify the gains in dollar-value impact.

Also, AI is not efficient from day one. It takes time to train and develop the models. The benefits are realised in the long term, perhaps years ahead. The real AI gains come from future-proofing the business. Evaluating the success of AI on the immediate ROI becomes foolhardy.

3. Artificial Intelligence is fast becoming the cost of doing business

AI is becoming part of the cost of doing business. It is not practical to invest in AI considering only the ROI, just as it is not practical to pay the monthly energy bills looking at the ROI.

Consider using AI to ingest colossal data sets for data analytics. Success depends on embedding the AI model into every process. But quantifying ROI becomes possible only by isolating the model to isolated use cases.

In such cases, the challenge before business leaders is to sync AI models with business realities. Process accuracy is not enough. An ML model being 99% accurate has no benefit if it does not meet business needs.

For example, consider the investment in AI-powered chatbots. AI-driven capabilities reduce the average call handling time and boost customer satisfaction. But implementation without proper application of mind could lead to chatbots going about in circles. The call eventually goes to a human agent, anyway. Here, deploying AI does not lead to any positive ROI benefits, and worse, it may end up annoying the customer.

The C-suite does not care if the model is accurate, robust, or drifting. What matters is if the investment removes gaps, such as improving a process or speeding up a transaction.

4. Problem of scale-up

AI projects are risky. Only a few projects get past the proof-of-concept stage. By then, the enterprise would already have invested sizable amounts. 

Most executives know the importance of data quality for successful AI app development. But 76% of executives face challenges with scaling the implementation of AI projects. 32% of executives took longer than expected to get an AI system into production. 

 The reasons for a low percentage of successful scale-ups are many. The top reasons include:

• Friction between data science teams and the rest of the enterprise. Often, rank-and-file employees do not understand what the data scientists develop or how it will benefit them. Lack of clarity from data scientists aggravates such issues.
• Competing priorities. Many models get rejected despite having the best algorithms and precision-recall. The development teams have priorities, and grappling with a problematic AI rollout may rank low on their priority list.

Building and deploying AI models at scale requires enhancing organisational capabilities for : 

• Software tooling and infrastructure stack. These tools give data scientists the agility to build and deploy models.
• Business processes and systems, enable data scientists to experiment and develop models seamlessly.
• Support for model governance. Data scientists should be able to monitor the performance and impact of AI models. They should also be able to take countermeasures without going through procedural hoops.

Arranging for these infrastructures increases the ROI of AI projects. Enterprises need to take a long-term perspective when making such investments. They could also leverage the cloud to convert capital expenses into easily attributable operational costs.

5. Not leveraging the full potential of Artificial Intelligence

Humans do incredibly complex work but cannot scale beyond a point. The true application of AI is to scale human effort.

Many enterprises over-focus on doing things faster and cheaper. Using AI as a blunt force tool for such projects fritters away the power of AI. They under-optimise the AI infrastructure.

For instance, using AI for pure automation play is a wasted investment, blocks capital, and reduces the ROI of AI investments. Automating simple statistics could do the job just as well.

Some of the generic AI use-cases that deliver value and qualitative gains for the enterprise include:

• Protecting personally identifiable information (PII) data. PII data falling into the wrong hands can result in costly regulatory fines, damage lawsuits, and loss of reputation. AI enables rolling out projects to classify and protect PII at a speed and scale impossible until not too long ago.
•  Easing knowledge management. AI systems automate access to relevant information for each employee at scale.
• UI improvements. AI-powered algorithms proactively analyse how humans interact with software and improve user experience.

6. AI-powered intelligence hardware shows promise

Advancements in technology have embedded situational awareness and cognitive intelligence in edge devices. AI enables these devices to read and analyse sensor data and solve problems. These AI-powered ready-to-use devices unlock several real-world use cases. Examples include:

• Managing critically ill bedridden patients’ vitals
• Industrial process controls in remote locations.

Such ready-to-install intelligent hardware spares enterprises from costly and resource-sucking AI projects. It will speed up AI adoption in a big way and ensure better ROI for AI investments. 

Businesses can no longer afford to regard Artificial Intelligence as a series of science experiments. Competitive pressures mount by the day, and the C-suite demands maximum returns for each dollar spent. Enterprises have to become more accountable for their AI investments than ever before.