When they first adopt AI, many organizations push for higher AI use and encourage experimentation. But is this the right approach?
Through AI, marketing may test content generation; project managers may summarize meetings; developers may draft code. These efforts can save time at first, but subtle errors and misinterpretations build up if the organization lacks a clear workflow and guidelines.
For CEOs, the challenge goes beyond deciding where employees may use AI. The organization needs to identify worthwhile applications, while preserving human judgment. The Siemens case study How Siemens Democratized Artificial Intelligence provides an example.
What Siemens Learned About Scaling AI
Siemens did not democratize AI by simply giving every employee a tool and leaving them to experiment. Rather, the company developed its approach through three stages (pages 6-7).
First, Siemens pursued tactical pilots led by domain experts and small business teams. These addressed specific needs, but independent development made knowledge transfer and reuse difficult.
Siemens then created a more collaborative environment. Its AI Lab brought domain experts and data scientists together to identify use cases. Siemen’s AI Academy, workshops, showcases, and self-service tools helped employees with different levels of AI literacy explore what AI could accomplish.
Finally, Siemens invested in shared infrastructure, so successful models and practices could move across products, services, and facilities.
Not every organization will succeed with this exact formula. However, three lessons apply more broadly:
Leaders must communicate how AI creates value. Domain experts must help find and develop worthwhile applications. Finally, organizations need training, knowledge transfer, and scalable infrastructure—not simply impressive pilots.
In our next article, we will explore this case study and leadership’s role in more detail. However, like Siemens, we need to agree on proper AI use within each team before we can promote shared infrastructure among departments. Each organization will vary in its structure, but most will have at least some of the following categories or roles:
- Marketing, Sales, and Writing
- Administration, Human Resources, and Project Management
- Development and Data Analytics
- Accounting and Other Regulated Work
- Consulting, Legal Work, and Research
For each one, we must consider four questions:
- What should employees contribute before using AI?
- What can AI reasonably assist with?
- What must the employee verify?
- What should never enter an unapproved system?
Marketing, Sales, and Writing
What should employees contribute? Define the audience, objective, product facts, brand voice and desired response. Writers should begin with original ideas, while sales and marketing employees should contribute customer knowledge. Like Siemens’ domain experts, employees should help identify valuable use cases and define what the AI-supported task should accomplish. [Table 3, page 12; Table 4, page 13; Table 5, page 17– Recommendation 2.1].
What can AI assist with? AI can brainstorm campaigns, draft copy, create early visual concepts, and summarize engagement data. AI may also analyze customer feedback and identify patterns among successful opportunities.
What must employees verify? Confirm claims using reliable sources, check product information, and review the final work for tone, intellectual-property concerns, and reputational risk.
What should remain protected? Do not enter customer lists, proprietary research, unreleased product information, or personally identifiable data into an unapproved system.
Administration, Human Resources, and Project Management
What should employees contribute? Provide priorities, stakeholder expectations, constraints, and the context behind decisions. The people closest to the work must distinguish meaningful patterns from misleading correlations.
What can AI assist with? AI can summarize meetings, organize requirements, review project plans, and evaluate RAID logs. AI can also organize employee feedback, draft job descriptions, and identify training needs. Weekly notes may reveal repeated delays or communication gaps. Similarly, Siemens used workshops and showcases to share lessons across teams (page 7).
What must employees verify? Confirm task ownership, deadlines, dependencies, and summary accuracy. Because employment decisions carry legal and ethical risks, AI should support human review rather than independently screening, ranking, or rejecting candidates.
What should remain protected? Do not upload private employee discussions, customer information, contracts, or sensitive project records.
Templates can improve routine communication, but messages requiring tone, context, or judgment should remain human-led.
Development and Data Analytics
What should employees contribute? Developers should define requirements, architecture, security, and maintainability standards. Analysts should understand the data’s source, quality, and meaning.
What can AI assist with? AI can draft code, recommend tests, suggest charts, and explore methods. Shared platforms can help teams reuse approved tools and practices. Siemens similarly combined self-service tools, knowledge exchanges, and shared infrastructure to scale successful AI applications across its manufacturing network [Table 3, page 12; Table 4, page 13; Table 5, page 19 — Recommendation 3.2].
What must employees verify? Evaluate test-generated code for security and maintainability. Validate the data, recalculate important figures, and examine the methods, assumptions, and conclusions.
What should remain protected? Do not enter source code, credentials, vulnerability information, proprietary algorithms, or restricted datasets into an unapproved system.
Accounting and Other Regulated Work
What should employees contribute? Provide the applicable rules, reporting purpose, materiality considerations, and transaction context.
What can AI assist with? Teams may use AI to categorize transactions, identify unusual spending, forecast cash flow, and draft preliminary variance explanations. Also, AI can flag possible inconsistencies.
What must employees verify? Recalculate important figures and reconcile them with source records. Additionally, review regulations through authoritative sources and require qualified professionals to approve high-impact work. AI cannot serve as an independent accounting quality check because it may overlook errors or invent explanations.
What should remain protected? Financial statements, payroll records, tax information, bank details, audit materials, and personally identifiable information require approved systems.
Consulting, Legal Work, and Research
What should employees contribute? Begin with a professional assessment of the problem, relevant evidence, possible approaches, and ethical obligations.
What can AI assist with? Through a back-and-forth with AI, consultants can challenge initial thinking, researchers can explore next steps, and lawyers can organize approved documents or develop arguments. As Siemens found, AI creates more value when combined with domain expertise.
What must employees verify? Consultants must determine whether recommendations fit the client’s circumstances. Researchers must validate their methods and findings. Finally, lawyers must verify every case, statute, and quotation in an official database. Apparent confidence is not evidence of accuracy.
What should remain protected? Do not enter client communications, privileged information, trade secrets, unpublished research, or sensitive medical data into an unapproved system.
Management Must Connect the Pieces
Democratizing AI does not mean removing controls. Rather, leaders need an organization-wide policy defining approved systems, protected data, verification standards, and ownership. They also need role-appropriate training and infrastructure for safe scaling.
At the end of the day, Siemens’ experience shows why AI adoption cannot remain solely an IT initiative. For instance, domain experts identify worthwhile problems, technical specialists develop solutions, and IT professionals ensure reliability and scalability.
Now that we have explored how such roles can specifically approach AI, we can dive into how to connect these contributions. When done properly, leaders create a sustainable AI capability.
Stay tuned for an article on this topic later this week! In the meantime, we recommend that you explore some of our related articles:
(Series Summary) Why Original Thought Will Determine Organizational Success
Think, Assist, Verify — And How It Can Be Your New AI Framework
Will AI End the Billable Hour Pyramid? (Summary)
Part 1 — Professional Services: The End of the Billable Hour Pyramid
Part 2 — The AI Industrial Shift: The White Collar Divide Is Spreading
Leave A Comment