Adopting AI in business requires extensive preparation across multiple areas to succeed.
By John Ghrist
Successful AI adoption in a business setting requires rethinking many aspects of how an enterprise does its basic work, as well as resolving numerous technical issues. Successful adoption can require significant restructuring of many "business as usual" practices to lay a foundation for AI implementation beyond a necessary pilot project, enabling staged adoption later.
Two IBM writers provide some of the best advice for starting. Pick a task or task group as a pilot project that delivers organizational value explicitly and sits across multiple functions. Map the workflow and then redesign how those functions will work using AI instead of current practices. Identify data inputs the AI will need and then optimize the data for AI use as necessary. Pay particular attention to metrics such as the time needed to complete an AI-automated task, the accuracy of the system's answers, the level of human correction needed, the cost per completed task, and the ease of user adoption. Test different AI models to see which performs best for both the pilot and longer-term use. Greater autonomy should count as an improvement only when it meets quality and control requirements. Be aware that deployed models don't necessarily learn continuously from new operational data.
An expert panel set up by the Forbes Technology Council offers strong advice for determining explicit organizational value. Determine how many hours can be reclaimed by doing the pilot's tasks the new way versus the old. Estimate how much decision-making time between insight and action ("decision latency reduction") can be saved using the pilot project’s procedures as a measure of improved strategic agility. Set a value on energy savings made possible by more efficient completion of pilot tasks. Evaluate how well the new procedure can provide model behaviors that reflect enterprise values in conducting business. Estimate how well the pilot delivers similar or superior performance across "varying domains, users, and conditions without retraining." Observe whether the pilot builds trust in the new procedures' efficacy among employees and, potentially, customers. Track how long the AI procedures can function without human correction ("autonomy-to-intervention ratio"), what percentage of end-to-end tasks can be completed by agents without human intervention ("autonomous resolution rate"), and how well AI saves time, effort, and resources "by reducing manual work and enhancing productivity" ("resource efficiency index").
Over time, the Forbes panel notes, the pilot project and later AI-enabled tasks can provide more metrics to justify further automation. Catalog "the number of new products, services or process improvements directly enabled by AI-driven insights." Determine "how much more quickly actionable intelligence is derived from data" ("time to insight reduction"). Assign value to the results of AI-influenced decisions and to how well AI use drives bottom-line improvements. Measure the speed at which AI can learn from new data ("adaptive learning rate"), the percentage of AI output that gets used for operations or decisions ("model utilization rate"), and the percentage of users who continue to use an AI feature after any "initial frustration" ("feature abandonment recovery"). Access management data show how tools affect productivity and point to gaps in training and other aspects of AI use. Finally, "return on disruption" can illustrate the transformative difference AI can make in enterprise procedures, indicating "long-term competitive advantage rather than incremental efficiency."
Build a Strong Measurement Framework
Filling all these slots with numbers will require a comprehensive measurement framework, not only to justify a pilot project's value but also to provide ongoing proof to stakeholders that AI investment is worth the expense and planning effort over the long haul. As part of the pilot project, organizers should set key performance indicators (KPIs) aligned with enterprise business objectives, establish a testing regime that compares AI results with those from previous methods on the same tasks, and use AI performance data to refine procedures over time.
It is also vital to communicate these results to all stakeholders to build confidence that adoption of the new technology is having the expected effects. Openly sharing tangible business benefits (e.g., cost reductions, revenue growth) through charts and other visual aids can encourage acceptance of change as beneficial. Sharing case studies and other data builds confidence and buy-in among hesitant top executives and employees concerned about job security. Designing a pilot that positions AI as a tool to enhance human decision-making encourages collaboration and openness to learning.
Proper Data Preparation is Vital
Optimizing data for AI use is a foundational step — and essentially a project unto itself — that will be useful not only for a pilot project but in improving data reliability and accessibility (whether the enterprise adopts AI or not) and is an important learning step if a pilot project succeeds well enough to greenlight further AI use expansion. Data cleansing includes ensuring data accuracy, removing incorrect values, and purging outdated entries and duplicate records. Standardizing data formats (e.g., machine readability, currencies, dates, metrics) across datasets and internal silos enhances reliability. Properly classifying (i.e., assigning labels or types to data elements based on their characteristics) and categorizing data is also important. However, it's too complex and too specific to each business area to describe here.
Data preparation must also address privacy and security. This aspect of data preparation has no simple answer and depends on the enterprise's overall planned role for AI. The enterprise must collect, store, and process data in compliance with all applicable laws in every jurisdiction where it operates, encrypt sensitive data, and limit access to authorized user roles. Anonymization removes or masks identifiers from data used for AI operations. However, some predictive and generative AIs are becoming adept at pattern matching against third-party data and can sometimes track anonymized records back to actual people. Differential privacy provides a mathematical framework for quantifying privacy loss but can dilute the data's analytic usefulness. Converting real data to synthetic data and using it for analysis is another way to hamper tracing data back to actual people (though it’s still not foolproof). Still, a risk is that synthesizing datasets from real data can undermine the validity of results beyond statistical analysis. The privacy-by-design strategy calls for building dataset protections directly into the AI system throughout its lifecycle, but adding that function increases cost. However, it is a strong method for protecting IBM i customer, employee, and transaction information.
Other general support activities in the early stages include offering AI training to all employees, selecting a pilot program that emphasizes collaboration between AI and people and automates repetitive tasks, and fostering acceptance of the mindset that success with AI will ultimately involve significant changes to how the enterprise work gets done and how the results will be evaluated.
Meeting Governance and Ethics Concerns
AI ethics is a framework that provides rules and limits on what AI systems should do. AI governance is the process of ensuring an AI system obeys those rules and limits.
A committee of humans typically sets AI ethics by formulating rules, translating them into enforceable policies, and overseeing system monitoring to ensure compliance. The foundation of ethical rules begins with following governmental legal requirements but also includes maintaining approved data access, holding data owners accountable, facilitating records auditing, providing escalation procedures, and requiring human approval for consequential actions.
Formulating ethical principles starts with answering what values the organization considers important to follow. Determine this by consensus among executive decision-makers, technical teams responsible for implementation, internal users of the system, and even the customer base. Next, decide what mechanisms are needed to ensure these values are followed in practice, who is responsible for ensuring that, and what the consequences of failure will be. These form the foundation for what to look for in a governance system the enterprise will purchase or build. Key aspects to consider include human standards such as fairness, respect for human dignity, transparency, and privacy protection. A good place to turn for guidance is the IEEE Standards Association, which offers seminars and general advice for organizations trying to formulate their own ethics standards.
Overcoming Insufficient Data
One difficulty some enterprises may encounter in starting an AI migration is lack of sufficient proprietary data to adapt AI models for testing. While generic data may show how different models work in principle, using actual enterprise data will be more representative. Internal databases such as transaction records and user logs are obviously most useful, although formatting and organizing the data may be complicated.
If the enterprise (or a particular line of business) is too young to have extensive data, other sources can help by providing data from similar businesses. These methods can work by establishing data pipelines that harvest data from public datasets (e.g., government portals, regulatory websites), by compiling statistics from manufacturing equipment sensors and other Internet of Things devices, and via customer reviews, social media posts, partner and vendor data, and IT support tickets detailing solutions to past challenges. Other potential information sources include recommendation engines, crowdsourced or distributed data pools, computer-vision systems, and consulting companies that specialize in gathering specific types of data.
Enterprises can combine these outside data sources with ELT and ETL tools that transform data into useful formats. Alternatively, organizations can use Retrieval-Augmented Generation (RAG) to improve a large language model (LLM) output by letting it use an external knowledge base as context. A RAG system retrieves selected data, formats it, and stores it for efficient search. Although it doesn't inherently retrain the model, a RAG pipeline lets an LLM use a query, stored data, and context to generate an appropriate response.
However, an IBM i organization can begin with an existing model and controlled access to approved business information, so a shortage of proprietary information doesn't automatically prevent a useful pilot. Of course, an integration project must preserve authorization restrictions.
Overcoming Lack of In-House Expertise
Adopting AI successfully will be transformative. The enterprise will do business in new ways, and existing staff will have to learn new tasks. This reality, along with employees' increasingly ingrained expectation that AI will cost them their jobs and the high failure rate of AI pilot projects, can create a perfect storm of an unfavorable environment for change.
The best way to address these concerns is to build a culture of AI literacy across the enterprise workforce from the start. In doing so, the organization shows it's serious about its AI commitment, wants to retain existing staff as much as possible, provides a growth ladder for those willing to climb it, and fosters a more positive attitude toward the technology among employees and stakeholders alike. It helps make the workforce feel included in the new paradigm.
Providing organization-wide AI education (e.g., online courses, in-person workshops, a knowledge hub for self-directed learning) can help the workforce understand AI concepts and practice with specific tools before using them to produce a specific work product. Designate and recognize internal champions who can influence others to be more open to change. Involve these champions in mentorship programs and cross-functional teams. Use those teams to evaluate pilot projects. Include AI literacy in executive development programs and offer AI discussion groups. If needed, bring in AI consultants to support internal champions or lead workshops. Send selected workers to industry conferences. Enable champions to participate in strategic planning. If using commercial software tools, partner with software vendors to educate employees. Over time, build a talent development network that includes outside educational institutions, implement internal career progression paths, and offer rotational assignments that give employees experience with AI in use outside their original environments or departments.
Gaining Customer Acceptance
AI's impact on customer service will strongly influence customer acceptance. Features such as efficient order processing, self-service options, and rapid complaint and query handling will go a long way towards promoting a positive response from the customer base.
Although it is offered as a promotion for a software product, this blog from German vendor OMQ provides a step-by-step procedure for building a strong customer service interface. In summary, the recommended steps include analyzing current customer queries, defining goals and PKIs for an AI-based automation system, building a customer-specific knowledge base, selecting the right AI bot for the enterprise, and running a pilot project for the customer service function. Then, as with any AI project, establish continuous monitoring and optimization, and avoid the enumerated pitfalls.
What we have covered here is an outline of some of the major issues involved in implementing AI in a business, but this summary should provide a useful framework. Long term, successfully executing an AI transformation plan will require researching these issues in greater detail and recognizing that transformation is not simply adopting a new software system, but a fundamental shift in how the enterprise does business going forward.
Business users want new applications now. Market and regulatory pressures require faster application updates and delivery into production. Your IBM i developers may be approaching retirement, and you see no sure way to fill their positions with experienced developers. In addition, you may be caught between maintaining your existing applications and the uncertainty of moving to something new.
IT managers hoping to find new IBM i talent are discovering that the pool of experienced RPG programmers and operators or administrators with intimate knowledge of the operating system and the applications that run on it is small. This begs the question: How will you manage the platform that supports such a big part of your business? This guide offers strategies and software suggestions to help you plan IT staffing and resources and smooth the transition after your AS/400 talent retires. Read on to learn:
LATEST COMMENTS
MC Press Online