An All-Hands has been called. The CEO steps in front of hundreds of adoring local employees, while thousands are participating remotely. Thumbs-up and Heart emojis are flying across the screen; the applause is ringing through the room. Then, silence. The room darkens; the spotlight is on the CEO.
"We are going to run full speed into adopting AI into our business," he says, with thunderous applause in the room, and more heart emojis flying across the screen. The CEO then goes into an in-depth pitch on how AI will streamline processes and workflows, increase productivity, and take the company to a whole new level of success. The crowd eats it up, both locally and on Zoom.
Then, the All-Hands ends, and folks go back to their daily tasks, often asking themselves and each other: How? How will adopting AI help? What will it look like? Does this mean layoffs, new hires, hiring freezes? Will my customers be taken care of? In Learning and Development, the first question would be "Are we going to be using AI-generated content?" and the next is "Will we use AI avatars in place of live instructors?"
The lost step here, which should be obvious, is implementation discussions. Before an announcement was made, before the CEO broadcasts it to the company, the Board, and to investors, the question that needs to be answered is "how will AI be implemented?" More often than not, that answer is "give everyone ChatGPT or Claude." That's not adopting AI for an ROI; that's adding a chatbot tangential to the business; an extra step that now has to be taken so that the company can feel like they are "adopting AI" while nothing really changes. It's an expense, and an added complication for the employee.
The Chatbot Trap
There's a reason why chatbots are so popular for "AI adoption" in the enterprise: It's really easy to implement. It's just a software subscription that is easy for Procurement to sign off on, and for IT to deploy. The thing is, it doesn't really solve a problem, address workflow or process questions, or provide any guidance on uploading and sharing files that may or may not have Personally Identifiable Information (PII). The chatbot alone doesn't have company- or task-specific information; rather, it has general information that may or may not be accurate. So it's a less-helpful tool that can carry a conversation. But it's easy to deploy, so it's easy to sign off on.
The thing is, the chatbot could be helpful if it had the right harness and connections, and was empowered by the right people at the right time, and secured properly. But that takes time, effort, and planning that companies don't feel they have: They need to prove they are AI-ready now, not in a couple of weeks or months post-implementation. The Street is demanding AI relevance; the Board has seen YouTube and LinkedIn where individuals have created "teams of employees" from AI Agents, and want that at scale, and the CEO is pressured with a sense of urgency to make it happen. Fast.
Then, reality hits. Billions of tokens and thousands of dollars later, the results are in: AI didn't help much; it just created more work. The outrage, the frustration, the CFO coming down on the CIO, CISO, CTO, and CPO for not making this work, the CEO worrying about what to tell the Board; all this comes to a head. The plan didn't work, but that's because it wasn't a plan: It was a vision. A vision that wasn't given time or direction to implement.
The AI Productivity Sweet-Spot
The illusion is over, but that doesn't make AI a bad investment. New technologies have a history of being misused. I can't count the number of "my cupholder is broken" calls I received while in tech support, when CD-ROMs were still considered new (as were computers). And yes, I realize I just aged myself. If properly implemented, AI has a lot of potential and can really increase the productivity of company employees while providing a level of convenience that is unprecedented, and perhaps a little intimidating to many leaders. Let's break it down through each layer of leadership and see how this disaster could have been avoided.
The C-Suite Vision
This part of the story fits well. The CEO had a vision: Implement AI to increase productivity and profitability. The C-suite signed off, and they announced it to the world. C-suite executives are supposed to come up with visions for their teams as they represent the vision of the company. Some, if they are part of smaller organizations, will plan out the VP's approach as well, but we will get to that. The part that broke was everything downstream.
The VP Plan
The Vice President is responsible for creating a plan that matches the C-suite's vision. They know the business, they know their part of the business (or should), and would come up with the plan. This is usually where things start to go wrong.
The instinct of most leaders is to solve the problem, and to solve it in the most efficient way possible. For the VP, the plan is pretty simple: Have everyone use AI. The easiest way to do that? Use a chatbot to ask questions. Boom. Done. Easy.
But business isn't that easy. Business is complicated by thousands of workflows, hundreds of processes, and dozens of policies that complicate any new technology rollout. CIOs and CISOs know this, and would likely try to advise teams on the needs and implementations.
What should happen here is the VP should call in their Directors, and share with them the Plan: Implement AI to benefit employees and increase revenue and productivity across the board. Then, put it to the Directors to come up with their department-specific Direction. The VP wouldn't know enough about each department to do more than that, while the Directors have a more narrow, more targeted perspective to understand how they will want to start with AI in their departments, and how they would measure success.
The Director's Direction
Now, we come closer to the tactical, and further away from the visionary. We are in the strategic: The view where a clear direction that would benefit the specific department can be defined with enough knowledge of how AI could benefit the team.
Unless you don't know how, and at this point of AI implementations within companies, that's a very real possibility. Teams haven't deployed AI globally yet; they don't know what they don't know. As the Director, you need to sponsor the discovery phase, review the results, and set a direction for the best possible AI implementation within the department.
The disconnect at this level is generally dictating from on high how the Director thinks AI should be deployed, rather than how it could be deployed to the best benefit of the team. What needs to happen is:
- The Director outlines the vision and plan as created by the C-suite and the VP.
- Then, go to their Managers and ask how those plans could be implemented, based on feedback from their teams.
- The Managers will then go to their respective Teams and talk through the workflows and processes, looking for opportunities to use AI to increase productivity. They come up with a general plan, then assign one person from the team, ideally someone who is both good at the job but often questioning how things are done, to be the test-runner.
- The Test Runner will then start building out with AI: working through workflows, creating automations, identifying clear opportunities for real impact to increase productivity, accuracy, and quality. NOTE: This can take a while; there's a lot of trial and error here, sandboxing will be necessary, and working closely with the CISO's team for security and the CIO's team for resources will be not just necessary, but expected. Plan on a quarter for discovery.
- The Test Runner then reports back to their Manager with the results, findings, opportunities, and suggestions on how AI could be implemented.
- The Manager then collects the recommendations with benefits and costs properly measured, and reports back to the Director with their findings for their team.
- The Director then takes the accumulated information and sets the Direction for their department's AI implementation, utilizing the best pros with a plan to mitigate the risks involved.
The process is long, longer than most C-suite teams would want, yet necessary for a real, solid gain in using AI per team. Once the direction has been set, it's time to implement.
The Manager's Orchestration
Now, we get to the tactical level: Implementing. Managers will work with the CISO and CIO to deploy AI Agents to their team members, grant them access to their necessary tools, models, applicable skills, etc., to make the move to AI that much easier. Because the automations and orchestrations were already test-driven by the Test Runner, general team implementations should be pretty straightforward.
Then comes the Change Management training: Yes, you need to train your people on the new processes and procedures they will be using with AI Agents in their work stream. Ideally, they will be using the AI Agent as their primary work interface, so they need to know how it works. Sure, AI chatbots make it look good, but phrasing and variations in adaptive work make skills incomplete, so employees need to know how to help the Agent update their skills to better work with their employee.
Then, give it time. It takes time for employees to trust new technology (I know some departments that refused to let go of the Fax machine or DOS apps), and that will be a barrier. Proper change management implementations will facilitate the migration.
All that's left is to monitor based on the KPIs that had been discovered during the initial Test Runner phase, and identify how to best massage processes and procedures with changes in model efficiency and other new technologies.
Avoiding the Traps
Now that you have a general outline of how to implement AI in an enterprise, let's talk about common traps that come up:
- Impatience: The Board wants profitability with AI NOW; CFOs want a quick ROI on all token money spent. Gains won't be shown over a week, or a month, or even a quarter. They start to improve once all AI orchestration is running, connecting, and streamlining processes and procedures. Until AI is at all levels, there will always be bottlenecks; they just get moved. Be patient, manage the bottlenecks you can control, and for those you cannot, like not having enough customers to consume the volume you are producing, invest in getting more customers.
- Greed: There's been a lot of talk about companies firing employees for the sake of AI automation, only to turn out they were clearing out headcount because of over-hiring. Yet it's tempting, when you are increasing productivity with a few employees, to drop the number of employees and match your current financial numbers with less payroll. This is short-sighted for the company: If you can't see your company growing beyond its current capacity when you are given employees who can increase productivity by several multiples, then you might need to step down and let someone with more vision run the company. Headcount should never be reduced because of AI: AI can only take on a portion of any job, not an entire job. Don't sell your company short because you are looking for a fast buck.
- Dictating from on High: I see this a lot: Leadership decides how everyone will use AI, and forces that decision on the employees at work. They disregard customer feedback, employee feedback, and even middle-management feedback: They know best. They've been to Harvard/Yale/Brown/etc. I saw this all the time while working at ServiceNow, and leadership was generally shocked that the ServiceNow PreSales teams would sit down with employees and figure out what they were doing, then automate the painful bits. Employees loved it; leadership had no clue those were pain points. Don't let your pride get in the way of progress. Listen to the people who are on the ground, and get them involved.
- Unclear Parameters: Throwing AI at someone and telling them to "figure it out" is not an R&D strategy; it's punting. You don't want to punt; you want to win. Sit down, know what budgets you have, what security restrictions are in place, and where your hard lines are. Walk through with your Test Runners and let them know that, while they have some latitude, they need to be conscious of these hard lines.
- Bad Data/Processes/Procedures: Don't use AI to implement your current convoluted, unclear, and often shadow workflows. Sure, they were created because your existing processes aren't working, but you don't want broken processes automated: they don't get fixed, and the automation runs "broken" but faster than you can monitor. Sit down and, with your Test Runner, work through the current processes, see how they can be changed and improved using AI in the workflow, and where automations can be made to streamline reaction times. Clean up your data so it's consistent, and use AI to maintain consistency (likely the best possible use for the Sales team!). Migrate as much as you can to the AI Agent prompt, so employees have one place to work: Their Agent prompt.
Conclusion
Implementing AI within a company isn't a bad thing; it just needs the right approach and strategy for a proper and lasting benefit.
- Work with your teams to understand how AI can be implemented within their workflows effectively instead of dictating how they will use it.
- Enable the process with full collaboration with the CIO and the CISO, so AI remains a valuable asset and not a potential regulatory violation.
- Don't focus on one particular type of implementation as the only potential option. Every team will use AI differently depending on their needs.
- Be open to the freedom that working from the AI Agent can give your employees. Offices, particularly open offices, become distractions, not valuable workspaces. Expect employees to see the value of working from home, or anywhere, and make that a possibility.
Successful AI implementations mean reevaluating your entire workflow structure for gains; no one app will solve the problem. It may mean more investment in different, better tools that fit within the new structure, much like complete overhauls were made when migrating to Cloud services, or before that, centralized computing, and before that, computers in general. This isn't a flippant software solution, but a complete corporate overhaul. Go into it with your eyes wide open, and you will see some amazing gains.