Successfully adopting AI as a business tool isn't simply a matter of buying a software package and training users. Implementation requires both a business strategy and a planned execution path.

While it’s clear AI has already become a significant tool for improving business efficiency that will have an increasing influence in the years ahead, what’s not so clear is the best path for getting to a place where AI is an established asset for any enterprise. Not only does the best way forward vary with the size and type of business in which an enterprise engages, but there are so many tools to choose from that it’s hard to compare the available options and match them to a particular business case.

Advice on how to proceed abounds, much of it following the same general pattern: define objectives, allocate resources, formulate an implementation strategy, find the "right" tools or partners, launch a pilot project, test and evaluate, and expand and improve the pilot until you reach perfection, In other words, the standard operating procedure for integrating any new software package into business processes. However, this glosses over many lesser steps and some significant differences between using an AI app and traditional apps that focus solely on accounting, customer service, or similar services. A more helpful approach might be to provide a summary of more granular steps for integrating AI into the formulation of a written plan for success.

Starting Small but Thinking Big

One preliminary step that’s easy to overlook is assessing the quality of data available to the enterprise. AI works best on data that's organized, not repetitive, and trusted. If they aren’t already in force, institute such practices as establishing data quality metrics, standardizing data formats, regularly cleaning data to eliminate duplicates and weed out inaccuracies, and setting data validation rules and automated data checks to maintain quality. Consider what enriching additional data sources might be available. Be aware of the difference between data quality and data integrity — the former deals with accuracy, consistency, reliability and security of data throughout its lifecycle, while the latter focuses on data itself to ensure it's fit for its intended purpose. Is the data as granular as it should be? Are the processes for updating it as reliable as they could be?

A second preliminary step is determining if the enterprise has the necessary skills to get started. Does management include people with some knowledge of AI, machine learning, or software engineering — or will it be required to hire new people or consultants to help? Will the enterprise try to develop AI tools in-house or use an off-the-shelf solution?

Because there’s no one best AI tool to start with, actual enterprise business needs, the variety of business tasks to which AI might be adapted, employee fear of a technology that’s already been painted as a job killer, and a host of lesser factors, simply getting a foot in the door is essential. Ideally, the first AI project will be a task or function for which AI can simplify or reduce repetitive actions, thereby winning users over to thinking, “This is helpful.” There are numerous possibilities for the role of an initial AI function. Examples include such ideas as building utilities for handling customer information requests (e.g., reviewing order information, requesting price lists, parsing query types to determine where to refer to more complex inquiries), creating marketing content (e.g., press releases, blog posts, product data summaries, common Q&A responses), creating valid internal documents (e.g., answering frequent general HR questions, providing basic employee training information, structuring basic candidate screenings), product research (e.g., analyzing large data sets, creating visual aids, identifying trends from unstructured information), and providing language translations for enterprises with international markets.

So, one of the early questions to answer about a pilot project is "what’s a basic task that’s widespread in use," or "is strategically pivotal for the enterprise that AI could improve?” Secondly, "what will the success of automating this task look like and how will that success be defined?" Once there is agreement among the powers that be on these two ideas, planning a strategy can proceed.

An AI Governance Committee (or a similarly tasked group) should be established for long-range planning. The committee should comprise members from legal, customer service, human resources, IT, and other relevant departments. This committee will have authority over AI projects, monitor compliance with ethical and legal standards, and review AI project impacts and performance. It should also define acceptable uses of AI in the enterprise, such as data security and privacy, AI biases in training data and algorithms, and articulate AI-related company policies clearly to all users. It must devise an implementation plan that includes measurable performance goals and can potentially guide the entire AI adoption process.

This committee must consider the legal, data security, and privacy ramifications of AI use, as well as any compliance requirements mandated by the type of business the enterprise operates. For example, some more highly regulated sectors, such as healthcare, may have to meet stricter requirements under the U.S. Health Insurance Portability and Accountability Act or the E.U. General Data Protection Act. Obviously, any enterprise should be aware of legal or regulatory mandates that apply to its particular industries and follow them accordingly.

The committee will also eventually need to set objectives for overall AI implementation, such as specific reduced costs, revenue opportunities, improved market analysis, and streamlining tasks. It should also be responsible for researching competitors’ AI strategies and trends in the enterprise’s industry. If sufficient expertise is available, the committee might also need to decide if building AI tools in-house is an option, and if not, keeping an eye out for new AI tools that might work better than those already in use.

The pilot project definition is next. Starting small first and expanding later as the value of a pilot project becomes apparent. The committee will either guide the pilot project directly or designate a person or team to do so. Recruitment of workers participating in the project should prioritize individuals who approach it with a positive attitude and are willing to focus on necessary training. Data summarizing and analysis will likely be central to the pilot task. The committee must set up testing and audits to ensure AIs used in the pilot project (and future systems should the pilot succeed) are functioning correctly. What safeguards, for example, will be required to guarantee the accuracy of the results the AI produces? How can "hallucinations" be weeded out or avoided? Will actual company data or synthetic data be used for training, and what are the risks of using the former? Are established ethical guidelines being followed? Are applicable data-protection laws being complied with? In the long run, the committee will become the "AI experts" in a strategic sense, such as determining what steps may be necessary to make new AI apps compatible with existing legacy systems, unless this task is given to a separate group.

Some thought will need to be given to methods for calculating the return on investment for the project. Determining key performance indicators (KPIs) will vary situationally and the actual methods for determining them are beyond the scope of this article. In general, KPIs attempt to measure such aspects as evidence of progress toward a desired outcome, team and individual performance, cost reductions, time saved, and work quality. Training the Users

There are two significant prongs to the question of training users. Those involved in the pilot project will need to learn a specific new tool that is being tested. Suppose the enterprise has determined that adopting AI apps is the best future path. In that case, attention must be given to more generalized training of employees in not simply AI-related skills. Still, a broader campaign about the functioning of generative AI to make it more familiar, neutralize fears about the technology, and acquaint employees with the concepts of using AI ethically and effectively.

For example, all employees should understand data-governance policies, particularly when compliance with regulations is necessary, and learn enough to understand the need to collect, cleanse, and organize pertinent data to support training of AI algorithms. The inherent value of data should be something everyone who works with it is aware of. Employees must be helped to see the benefits of AI not only for themselves but also for the organization. For example, create opportunities for employees to see how using automated text creation supports order-substitution and personalized marketing, how Virtual Assistants can personalize training in employee onboarding and mastering AI tools, and how it’s possible to provide conversational interfaces for a wide variety of tasks. It’s worth noting that AI familiarity could become a key entry requirement for jobs in the enterprise industry within a few years. Upgrading this familiarity with technology will be valuable to employees, regardless of their future career paths.

Set a tone that encourages experimentation. Trying different things out can lead to all kinds of valuable discoveries that weren’t necessarily part of the original plan. Encouraging AI use in general can help reduce employee fears of being "left behind" career-wise and encourage them to consider automating other repetitive tasks in the future. Insights into using new tools can lead to ideas for redesigning workflow processes and finding new value or new markets for the enterprise itself. Perhaps there are ways in which generative capabilities can inspire enhancement of the company's products and services once employees feel empowered by new AI-related knowledge rather than being threatened by it. Awareness of more complex variations in data could spur new ideas for existing products and services or help workers more readily spot problems and data outliers that further improve the accuracy of data analysis.

Gather Data that Illustrates AI Value

Prepare for the ongoing task of continuously monitoring and improving the AI processes that the organization inaugurates. The enterprise should observe, record, and internally publicize pilot project successes. What tedious or repetitive tasks have been overcome? What task steps are being usefully consolidated? What insights have been gained from data analysis or from successful rethinking of how tasks are accomplished thanks to AI? These contributions to overall value in business processes or operations build confidence and trust that can encourage support from management and employees alike. Open communication of this information within the organization can go a long way towards encouraging acceptance of any technological change.

As an AI system develops, incorporate a change management plan that continually adapts to new developments in the fast-changing AI industry environment. Latest insights, products, and methods seem to pop up on almost a weekly basis in the AI market segment. Flexibility, experimentation, and continuous learning can all lubricate an AI implementation and contribute to success. Another critical process to incorporate into any AI system is the idea of using a Human-in-the-loop (HITL) approach. Essentially, it requires human involvement in decision-making when building a machine learning model. This ensures that there is a check that the new system is producing fair and accurate results.

Combining government, open-source, private, and organizational data can enhance business forecasting, improve supply chains, support customer service, enable valuable product innovations, and provide other benefits. Improved data discovery and search will make information more accessible, leading to better analysis and operational decisions that can boost enterprise profitability. Introducing AI into procedures and processes can provide significant rewards to large and small enterprises alike provided the implementation follows an organized and thoughtful plan.

John Ghrist has been a journalist, programmer, and systems manager in the computer industry since 1982. He has covered the market for IBM i servers and their predecessor platforms for more than a quarter century and has attended more than 25 COMMON conferences. A former editor-in-chief with Defense Computing and a senior editor with SystemiNEWS, John has written and edited hundreds of articles and blogs for more than a dozen print and electronic publications. You can reach him at This email address is being protected from spambots. You need JavaScript enabled to view it..