Corporate AI Adoption Is Stalled: How to Set It Free?

A new report from Slack says employees are distancing themselves from generative AI, with a growing number expressing discomfort about using it at work. Bosses are on the fence too.

A survey by the US Census Bureau found that only 5% of companies use AI to produce goods and services. Others are trapped in pilot mode. Of those firms with projects on the go, only 8% have taken more than half of their generative AI experiments to production.

Messy data is one problem. But so is fear of being left behind if a firm moves too slowly or the reputational damage if it moves too fast and screws things up.

Corporate AI is stuck in a rut. How can businesses get the wheels turning again?

Table of Contents Table of Contents

1. Key takeaways
2. Enterprise AI Adoption: Initial Results vs. Long-Term Reality
3. AI Adoption Challenges
4. Where Do You Grow From Here?
5. Data Quality Is Vital, But Stay Focused
6. The Bottom Line
7. FAQs
8. References

Show Full Guide

Table of Contents

1. Key takeaways
2. Enterprise AI Adoption: Initial Results vs. Long-Term Reality
3. AI Adoption Challenges
Show Full Guide
4. Where Do You Grow From Here?
5. Data Quality Is Vital, But Stay Focused
6. The Bottom Line
7. FAQs
8. References

Enterprise AI Adoption: Initial Results vs. Long-Term Reality

Big tech’s big investments in GenAI are starting to pay off. Q3 earnings for Microsoft (MSFT), Google (GOOGL), Meta (META), and especially Amazon (AMZN) all show surging demand, affirming Amazon CEO Andrew Jassy’s April prediction that AI-related activities would drive “multiple billions of dollars” of AWS revenue this year.

Investors will be relieved, given the tens of billions being burned every quarter on new AI infrastructure.

Inside the organizations they sell to, however, the rate of AI uptake is harder to gauge. According to the latest AI adoption statistics, with around 100 million reported users in the USA alone, individuals look to be leading the charge. Companies, however, are less steady.

There’s no shortage of projects in the pipeline—a key driver behind big tech’s AI revenue spike. But only a few see the light of day.

Deloitte’s Q3 State of Generative AI in the Enterprise report says that 68% of firms have moved 30% or less of their GenAI pilots into production, while only 8% have deployed more than half. That echoes an oft-cited 2023 study that claimed 75% of corporate AI projects fail.

Among techies, a level of AI cynicism has set in.

A survey by Dice.com found that 30% of IT professionals think their companies’ AI initiatives are basically ‘for show,’ designed to give executives just enough cover to tell investors they are ‘doing something’ with the technology.

AI Adoption Challenges

Shy AI

Delve into the adoption metrics for end-users, and the picture gets murkier. November figures from Slack’s Fall Workforce Index show that staff use of generative AI tools has slowed, especially in the US.

Employees appear to be using it but aren’t sure if they should say so. About half of those surveyed told researchers they felt uncomfortable ‘admitting’ AI use as it might be seen as cheating or fuel gossip about laziness or incompetence.

Shy AI is one factor, another is ‘try and goodbye.’ The World Bank thinks there are over 500 million ChatGPT users globally, outpacing even OpenAI’s internal estimates.

That may be true, but ‘users’ can be an ambiguous term. A Reuters Institute study found that most people tinker with a genAI tool once, then abandon it — or maybe use it once every couple of weeks.

So, huge interest, little use, and companies struggling to get AI initiatives out the door. None of that bodes well.

Gap, or Gulf?

Why are companies using AI stuck? Blame the fear factor. Staff are worried about what will happen if they own up to using ChatGPT to write an email.

Executives have reputational concerns of their own. Bide your time on AI, and you might be branded as a laggard. Go the move-fast-and-break-things route, and you might screw up badly—and publicly—as some have already discovered.

Generative AI still hallucinates, flings botshit, churns out underwhelming text and image content, plus it’s opened up new privacy issues and attack vectors. There are concerns about the quality of data that pilot projects feed into large language models (LLMs). And legal and compliance risks loom large.

The EU’s AI Act came into force in August, with penalties for non-compliance levied at between 1 and 7% of a firm’s global revenues.

In the US, at least 40 states have AI bills moving toward enactment. Meanwhile, a bevy of lawsuits related to copyright and privacy breaches are in process.

In industries like finance and healthcare, where regulatory scrutiny is high, executives need more convincing. The fines and reputational damage that an errant AI could inflict if it exposed sensitive medical or financial data give CEOs in those sectors plenty of reason for pause.

The bar for new project approvals is high, which presents another challenge: How do you measure a GenAI pilot’s value?

Deloitte’s State of Generative AI report says that 41% of organizations have struggled to define and measure the exact impacts of their GenAI efforts.

It underpins the uncomfortable truth in Big Tech’s quarterly AI earnings: revenues haven’t really taken off. Even if the multi-billions predicted by Amazon’s chief materialize this year, it will be a drop in the bucket compared to the $110 billion AWS expects to make in 2024.

For corporate AI adoption to be sustainable, firms need help to break the logjam.

Where Do You Grow From Here?

Picking up on themes in Deloitte’s Q3 AI report, Varvn Aryacetas, AI Strategy & Innovation practice leader for the consultancy’s Monitor UK arm, told Techopedia that it’s easy for companies to get caught up in the excitement of generative AI.

Scoping pilot projects for a successful outcome, however, requires a rigorous approach.

“GenAI can be applied to such a broad set of objectives and problems, sometimes it’s hard to bring focus and ground it in business value,” he says.

“Many stakeholders who sponsor pilots can get excited by a use case that suddenly seems more feasible than before because of generative AI. But it’s important to bring a level of skepticism and pragmatism to piloting GenAI.”

The following questions might help you assess the project’s potential right.

1. Is the Use Case Suitable?

“Is generative AI the best fit for solving your problem? Is it a responsible use of this technology, and is it compliant from a regulatory standpoint? Sometimes it’s not,” says Aryacetas.

“For example, if you wanted to automate the creation of standardized employment offer letters, GenAI may not be appropriate because these are well-thought-out legal templates that don’t require advanced personalization. There are simpler and cheaper tools for this.”

2. Is the Use Case Desirable From a Business Standpoint?

Will it increase productivity, boost morale, or provide differentiation? Will it improve the experience for your customers, employees, and supply chain partners?

3. Is the Use Case Feasible?

Do you have the right data, or is the required data too sensitive? Is it technically feasible in the context of your technology estate?

“It’s important to look at the systems that will need to integrate with the solution and the data structures and types they use. There could be concerns around data privacy, data bias, or data sovereignty. All these things come into it from a feasibility standpoint,” says Aryacetas.

4. Is the Use Case Viable?

“This is based on how much the pilot might cost and being able to calculate what the ROI will be if you scale it,” he added.

Understanding the expected maturity of the technology is another consideration. Some generative AI models and implementation patterns have become considerably better over the last 24 months.

Which other advances and breakthroughs might take place over the next 24 months?

The inference costs of LLMs have also come down 100x in the last 24 months from about $50 to $0.50 per 1 million tokens—and that may continue. How will that impact ROI? There may be a compound effect of the advancing technologies.

Data Quality Is Vital, But Stay Focused

Addressing data quality and governance are vital considerations, but depending on the use case, data management needn’t always be a major barrier.

“Algorithms are only as good as what you teach them and how you teach them. This means that the quality and quantity of data you feed them are crucial, but so are the training techniques such as fine-tuning, chunking, indexing, and choices in terms of foundation models and architectural patterns,” Aryacetas says.

“That’s true for generative AI solutions just as it is for other machine learning solutions broadly. Whether it’s for a pilot or a proof of concept, you must do the non-glamorous work of choosing the right data and cleaning up your data. But if you’re just starting out with generative AI, there is some low-hanging fruit you can take advantage of.”

He notes the example of a project to create a customer service tool that augments the work of a human agent:

“The dataset it needs to function may start with something as small as  50 FAQ documents and manuals which typically already exist. Addressing data quality issues and then harnessing this dataset as a knowledge base for a GenAI proof of concept should be fairly straightforward.”

Aryacetas says it’s essential to engage front-line staff from the start when using GenAI solutions.

“You don’t want to ‘do it to’ a team. You want to ‘do it with’ them. AI transformation is about collaboration, not dictation. Make sure that everyone from the executive sponsor to project leaders and the people on the ground selected for the team all help shape the use cases from day one.”

The Bottom Line

Rising AI skills shortages underline how important it is to have the right people in place.

According to figures released this week by Indeed’s Hiring Lab research unit, US job postings mentioning GenAI or related terms leaped by 350% over the past 12 months. The study also found that mentions of GenAI were most commonly attached to roles where such tools could eventually replace human skills.

What’s really driving corporate GenAI adoption, its transformative potential or the promise of reducing headcount? The tension between executive ambitions and front-line realities looks set to influence AI rollouts for the foreseeable future.

FAQs