Hands-On AI Training Without Internal Company Data
Learn how to prepare effective hands-on AI training using public, synthetic, and approved company data while establishing clear learning objectives, evaluation criteria, and data security guidelines.
π‘ Key Takeaways: A 1-Minute Summary
When internal company data cannot be used for AI training, start by defining the business tasks employees need to practice and the skills they should demonstrate, not by searching for replacement datasets.
Public data, synthetic training data, and approved company data each serve different purposes and come with distinct requirements and limitations. Simply removing a few names from an internal document does not make it synthetic data.
When preparing synthetic training materials, include expected answers, intentional errors, and explanations of the required corrections. Before delivering the training, test the exercises in an approved environment and confirm the conditions for entering, storing, and sharing data.
When employees cannot bring internal company data into AI training sessions, organizations can still design meaningful hands-on learning experiences.
The key is to define the tasks employees need to practice and the capabilities they should demonstrate, then select appropriate materials from public data, synthetic training data, or company-approved data.
Creating training materials that merely resemble internal documents is not enough. Organizations must also determine which decisions participants should practice making and what they need to review or verify.
At Codepresso, we recommend that organizations refrain from entering internal information into AI tools until the necessary data-sharing and tool-use permissions have been confirmed.
Instead, training teams should first agree on which learning activities can be conducted safely using approved educational materials.
The following recommendations provide a practical framework for designing AI training. Organizations should adapt them to their own internal policies and security requirements.
For guidance on choosing between theory, demonstrations, and practical exercises, see our related article: AI Theory vs. Hands-On Training: Which Should Your Organization Start With?
This article focuses on the next step: how to prepare training materials once your organization has decided to incorporate hands-on AI exercises.
1. Start with the Training Task, Not the Dataset
Consider a hypothetical AI training program for a sales team preparing weekly performance reports.
Suppose the learning objective is to help employees verify totals and reporting periods against source data and correct interpretations that lack supporting evidence.
In this case, the organization does not necessarily need to use actual customer sales figures.
Instead, the training team can create a fictional dataset containing sample customers and sales figures that allows participants to practice the same review process.
For example, the dataset could intentionally include:
- Incorrectly calculated totals
- Rows with inconsistent reporting periods
- Interpretations that are not supported by the underlying figures
Instructors should prepare an answer key that identifies these errors and explains why each correction is necessary.
However, successfully identifying errors in synthetic training data should not automatically be interpreted as proof that an employee can perform the same task effectively in a real business environment.
Codepresso's published Day 1 training record for Korea Pizza Hut's AX Grow program also explicitly identifies the demonstration data as synthetic values created for educational purposes.
This highlights an important starting point: Even when real business data is not used, participants should clearly understand what type of data they are working with.
2. Understanding Three Types of Data for Hands-On AI Training
Organizations can generally consider three categories of training materials: publicly available data, synthetic training data, and approved company data.
Each category supports different types of exercises and requires different precautions.
Comparing Data Types for AI Training
| Data Type | Suitable Training Activities | What to Verify Before Use | Limitations |
|---|---|---|---|
| Public Data | Source verification, summarization, and comparisons of published figures | Usage terms, original sources, reference dates, and whether the data may be entered into the selected AI tool | Public availability does not automatically grant unrestricted reuse rights |
| Synthetic Training Data | Error identification, refining task instructions, and practicing corrections | Clear synthetic-data labeling, expected answers, and confirmation that no real confidential information is included | May not fully reflect the complexity of actual business operations |
| Approved Company Data | Reviewing outputs against internal document formats and business requirements | Authorized approvers and permitted tools, inputs, storage methods, and sharing conditions | Approval for one use does not authorize other tools, activities, or distribution methods |
Public Data: Check Usage Rights and Sources
Publicly available information may be useful for exercises involving summarization, fact-checking, and data comparison.
However, information being publicly accessible does not necessarily mean it can be reused without restrictions.
Before including public data in AI training, confirm its usage terms, source, publication or reference date, and whether it may be entered into the AI tools being used.
Synthetic Training Data: Create It for the Learning Objective
Synthetic data offers a practical alternative when internal information cannot be used.
However, it should be created specifically for the learning activity rather than produced by simply modifying existing company documents.
For example, removing employee names or changing a few customer identifiers in an internal report does not automatically make that document synthetic.
The document may still contain confidential business information or personal data.
When possible, consider creating entirely new fictional materials based on the skills participants need to practice, rather than copying and editing actual internal records.
Approved Company Data: Stay Within the Authorized Scope
Company data may be appropriate for certain training activities when its use has been explicitly authorized.
However, the approval should define the conditions under which the data may be used.
These conditions may include the approved AI tools, types of information that can be entered, storage locations, and permitted sharing methods.
Authorization to use company data in one training environment does not automatically extend to other tools, platforms, or distribution channels.
3. Training Materials Also Need Clear Evaluation Criteria
Even when an exercise uses synthetic data, participants still need clear instructions and evaluation standards.
Each training file should include three essential elements:
- The task: What participants are expected to accomplish
- The data description: Whether the materials are public, synthetic, or approved company data
- The completion criteria: What a successful result should demonstrate
For example, if participants are asked to prepare a sales report draft, the evaluation criteria might include whether they:
- Verify that calculated totals match the source data
- Confirm that reporting periods and measurement units are consistent
- Flag unsupported interpretations as requiring further verification
Participants should also document which errors they identify and how they correct them.
This creates a record that instructors can use to provide specific, evidence-based feedback.
Collecting only the final output may overlook important aspects of the review and correction process.
Training files should also remain clearly labeled as educational or synthetic materials so they are not confused with documents intended for actual business use.
4. A Five-Step Process for Preparing AI Training Materials
Organizations can follow a structured approach to prepare hands-on AI exercises without relying on confidential company data.
Step 1. Define the Expected Deliverable and Review Criteria
Start by identifying what participants should be able to produce after the training.
Specify the expected output and the elements employees must review or validate.
Step 2. Separate Tasks That Require Internal Data from Those That Do Not
Determine which parts of the training genuinely require company-specific information.
Identify which skills can be practiced using public or synthetic examples instead.
Step 3. Verify Data Usage Terms and Internal Approvals
If using public data, review the applicable usage conditions.
If using internal company data, confirm the scope of authorization before including it in the training environment.
Step 4. Prepare Synthetic Data with Expected Answers and Explanations
When using synthetic materials, develop an answer key that includes the expected results, intentionally introduced errors, and explanations of why corrections are required.
This helps instructors deliver consistent feedback and evaluate participants against clear standards.
Step 5. Test the Exercises in an Approved Training Environment
Run through the exercises before the training session.
Verify that participants can complete the assigned tasks using the approved tools and confirm the rules for entering, storing, and sharing information.
If an exercise cannot be completed under the approved conditions, adjust the training scope.
For example, you may begin with an instructor-led demonstration or an activity focused on reviewing AI-generated outputs, then introduce individual hands-on exercises once the necessary preparations are complete.
Training should not be designed around workarounds that bypass account restrictions or data usage policies.
For more guidance on separating training accounts from production environments, read How to Set Up a Safe AI Training Environment.
5. Reevaluate Data and Access Requirements Before Applying AI Skills at Work
Successfully completing an exercise with synthetic data does not automatically mean the same workflow is ready for use in everyday business operations.
Before applying the training to real work, organizations should verify the relevant tools, data, and access permissions again.
For example, if employees have practiced reviewing fictional sales reports, the organization should confirm with the appropriate stakeholders whether the same evaluation criteria can be incorporated into actual reporting templates.
This helps bridge the gap between educational exercises and real workplace implementation.
Plan Hands-On AI Training with Codepresso SkillCamp
When considering Codepresso's instructor-led training program, SkillCamp, we recommend sharing any restrictions on internal data usage along with the intended learning outcomes.
This helps the training team understand the conditions under which the program must operate.
The availability and permitted use of specific datasets or AI tools should be confirmed through consultation before the training begins.
Start with the AI Training Readiness Checklist
Codepresso's AI Training Readiness Checklist can help organizations identify unresolved planning requirements, particularly those related to training objectives and scope.
Use the checklist to document which decisions are still pending and who is responsible for confirming them.
The checklist is designed to evaluate organizational readiness for AI training planning. It does not assess employees' actual AI competency or measure training effectiveness.
If you need assistance reviewing training requirements, data restrictions, or delivery conditions, our team can help you explore an appropriate approach.
Check Your AI Training Readiness