Artificial intelligence (AI) is front-and-center in popular conferences, podcasts, books, websites, and CIO agendas. It seems everyone is talking about AI, but other than large tech companies that have invested heavily in it, few organizations have seen much progress or success with it yet.<\/p>\n
AI offers huge potential to solve problems in areas like demand forecasting<\/a>, production planning, predictive maintenance, customer analytics<\/a>, medical diagnostics, and so much more. Yet too often, teams see their AI hopes and dreams dashed.<\/p>\n Why? Because deploying AI applications is a challenging process that can be derailed by many factors. Several issues companies face during AI implementations are technical in nature: insufficient volumes of data, skills shortages, scalability, and interoperability, to name a few. However, organizational or process-related challenges can often be more difficult to overcome. These are the challenges we\u2019ll explore in this article.<\/p>\n For example, let\u2019s say you\u2019re deploying an advanced forecasting application in conjunction with a production scheduling algorithm to better match production of your company\u2019s products to the uncertain demand in your industry.<\/p>\n Simulations indicate you could reduce finished goods inventory by nearly 10% without affecting the company\u2019s impeccable service levels. Field trials support this, but the project fails to lift off.<\/p>\n Below are five scenarios that could lead to this failure to launch and that are common roadblocks to AI progress. As you prepare to develop and deploy your own AI applications, these challenges must be considered.\u00a0<\/span><\/p>\n When deploying AI, stakeholders often agree on the business need or use case, but they don\u2019t always agree on the specific goals.<\/p>\n Using the demand forecasting example above, let\u2019s say the organization overall has a goal to reduce inventory, but some parts of the organization have strong incentives that work against that goal.<\/p>\n For example, purchasing departments are often rewarded for getting the lowest prices on raw materials. Things like minimum order quantities and large batch sizes \u2013 intended to provide price breaks and efficiencies \u2013 often result in excess inventory and clogged flow. The department may achieve its cost targets, but it fails to meet inventory targets in the process.<\/p>\n When looking at an AI business case, it\u2019s critical to look across all departments that feed into it and understand how their goals and KPIs align. If there is a chance for conflict, it must be resolved before the implementation begins.<\/p>\n If you implement new AI models but certain parts of the organization won\u2019t use them or won\u2019t use them consistently, then you have an adoption problem.<\/p>\n In our demand forecasting example, this could be represented by a scheduling team that is hampered by an overly complex legacy system that they use to understand inventory levels and plan and schedule production. They may not like this system, but they don\u2019t have the budget or resources to replace it with the new AI application. They follow recommendations of the new forecasting models only some of the time, which erodes the results.<\/p>\n There are countless reasons that AI adoption is a challenge for organizations. It\u2019s critical to understand how new AI models will affect various functions within the organization, then plan and budget for a rollout that includes them, their processes, and their budgets.<\/p>\n As new AI models are deployed, they must be integrated into the organization\u2019s process workflows, such as the scheduling process mentioned above. Often, the original workflows are so dependent on a legacy process or system that it can be daunting to integrate with new models.<\/p>\n Some organizations still have highly complex calculations that were built into spreadsheets. In those cases, it\u2019s necessary to reverse engineer the spreadsheets to get requirements. The model results have to then be manually entered into the workflow, and the complex calculations must be completed as part of the production process. The cost and resources required for this type of effort can be more than the budget allows and can hold the AI integration back from being completed.<\/p>\n Too often, reporting and measurement are left out of the plan or are under-represented, as teams opt instead for ad hoc reporting and measurement. However, it is critical to ensure you have agreement in advance regarding how you will measure success. If this is not in place before you begin field trials (or move to production following successful field trials), it won\u2019t be possible to validate the business case or understand the actual lift the AI application provides to the enterprise.<\/p>\n A clear production performance measurement plan with standardized reporting can help an organization better understand what is happening in an application field trial.\u202fThe following questions need to be answered:<\/p>\n AI success depends heavily on management commitment. Management support is needed to secure sufficient budget and resources, to encourage buy-in from stakeholders throughout the organization, and to maintain alignment with business goals. Without management support, you risk failure at several levels.<\/p>\n The best approach to getting management commitment is often presenting a clear business case that identifies the insight, the action, and the model\u2019s performance that maps the business impact of the activities and KPIs to the financial results the business can expect.\u202fIn addition, the business case must be tied to an investment plan that identifies clear checkpoints where investment and progress are measured and a risk\/benefit decision is tied to a go\/no-go decision.<\/p>\n Finally, the same business case must be tied to the production performance measurement plan.<\/p>\n Though this article has focused on the example of an AI model for demand forecasting, the challenges listed are not uncommon for any use case. While there are many technical challenges involved in deploying AI applications, it is typically organizational complexities like these that present the most critical challenges.<\/p>\n Even with management support and an eager field, pushback from specific people or groups, poor planning, and insufficient budgeting can put your project at risk. To be successful, make sure you understand the effort involved in rolling out new models, integrating them with existing workflows, and measuring the results.<\/p>\n Ready to start a successful AI implementation? Get in touch<\/a> with our team today.<\/p>\n","protected":false},"excerpt":{"rendered":"1. Inconsistent goals<\/h2>\n
2. Poor user adoption<\/h2>\n
3. Lack of integration<\/h2>\n
4. Lack of reporting and measurement<\/h2>\n
\n
5. Management commitment<\/h2>\n
A successful path for AI implementations<\/h2>\n