# Common mistakes when implementing artificial intelligence in universities

> Learn the most common mistakes when implementing artificial intelligence in universities and how to avoid them through strategy, governance, training and evaluation.

- Site: Genialoh (https://genialoh.org)
- Language: en
- Category: Governance
- Reading time: 16 min
- HTML version: https://genialoh.org/#/en/blog/common-mistakes-implementing-ai-universities

---
Implementing artificial intelligence in a university is not just about hiring a platform or letting professors and students use generative tools.

Adoption affects learning, evaluation, privacy, teaching work, student experience and institutional identity. When these dimensions are not coordinated, the university can end up with multiple tools, contradictory rules, low participation and difficulties proving value.

Most problems do not appear because AI lacks possibilities. They appear because implementation starts without a clear purpose, without owners, without training or without a process to learn before scaling.

Below are the most common mistakes when implementing AI in universities and the decisions that can help avoid them.

## Mistake 1. Starting with the tool instead of the problem

A university sees a platform, watches an attractive demo and decides to adopt it before determining what need it wants to solve. Then it tries to find activities that justify the purchase. This order can produce an implementation disconnected from academic priorities.

- Professors use different tools without common criteria.
- Students need extra support in certain programs.
- The university lacks academic integrity guidelines.
- Leaders lack information on adoption.
- The institution wants its own alternative to generic platforms.
- Professors need training to integrate AI in their activities.
- Leadership needs to visualize an institutional experience before approving a project.

> **Key question**
> 
> What institutional situation should be different after implementing the platform? The answer guides scope, participants, training and metrics.

## Mistake 2. Trying to implement AI across the entire university on day one

A massive rollout can look more ambitious but adds complexity. The institution would have to train many professors, prepare materials from many programs, handle questions, manage access and set rules for very different scenarios.

- Personalize the experience.
- Work with specific coordinators.
- Review academic materials.
- Train participating professors.
- Compare needs across disciplines.
- Gather feedback.
- Adjust before scaling.

Starting with two programs does not mean lacking an institutional strategy. It means applying that strategy gradually and in a controlled way.

## Mistake 3. Leaving implementation exclusively to the technology team

The technology area is fundamental, but it cannot decide alone how AI should be used in teaching. Implementation affects learning objectives, activities, evaluation, academic integrity, faculty development, privacy, student experience, institutional communication and the commercial model.

| Area | Role |
| --- | --- |
| Leadership / rectorate | Priorities and resources. |
| Academic direction | Validates use cases. |
| Educational innovation | Designs adoption and methodology. |
| Technology or IT | Evaluates access and operations. |
| Data protection or legal | Reviews information and terms. |
| Academic coordinations | Translate to each program's context. |
| Faculty development | Training and support. |
| Professors | Test the experience and provide disciplinary input. |
| Student experience | Supports adoption and help. |

## Mistake 4. Not appointing an operational owner

A broad committee does not replace one person responsible for daily progress. Without coordination, tasks stay scattered: nobody collects documents, meetings are not scheduled, professors receive no information, adjustments go unrecorded, decisions are postponed and the vendor gets contradictory instructions.

The university must appoint an owner able to coordinate objectives, participants, documents, calendar, access, training, feedback, metrics and next actions. Their role is to turn decisions into actions and dates.

## Mistake 5. Not distinguishing individual exploration from institutional adoption

A professor can open an account, try a tool and use it in some activities. That exploration can be valuable but is not yet an institutional implementation.

- Shared objectives.
- Authorized tools.
- Owners.
- Usage rules.
- Information protection.
- Training.
- Support.
- Monitoring.
- Evaluation criteria.

When both levels are confused, the university may assume it is already implementing AI because some professors use it. In reality there may be fragmented adoption. The goal is not to eliminate individual initiative but to integrate it into a common direction.

## Mistake 6. Using a generic tool as the only institutional experience

Generic tools can be useful for personal exploration. However, a university may need something different when it wants to offer an official experience.

| Generic tool | Institutional platform |
| --- | --- |
| Broad, unbranded context | University name, logo and colors. |
| Decontextualized answers | Configured with curricula, programs and methodology. |
| No program tutors | Tutors by program and teacher support. |
| No institutional analytics | Reports and analytics by program. |
| Individual adoption | Training, implementation and institutional support. |

The mistake is assuming that enabling individual accounts equals building an institutional strategy.

## Mistake 7. Not involving professors before launch

Professors should not discover the platform at the same time as students. They need time to explore it, test use cases, identify problematic answers, review alignment with their courses, define rules, redesign activities, prepare communication and request adjustments.

Early participation does not require unanimity. It requires that those who will support learning understand the project and contribute disciplinary knowledge.

## Mistake 8. Limiting training to teaching prompts

Learning to write instructions can be useful but is not complete faculty training.

- Capabilities and limits of generative AI.
- Information verification.
- Possibility of false or incomplete answers.
- Privacy and academic integrity.
- Activity design and evaluation.
- Responsible use of materials.
- Communicating rules.
- Interpreting analytics.
- Process to report issues.

Training should include practice by discipline and have follow-up. An isolated lecture rarely resolves the questions that appear when a professor applies the tool in a real course.

## Mistake 9. Not setting rules before students use AI

Without guidelines, each professor may decide differently. This inconsistency creates confusion and can cause evaluation conflicts.

| Level | Condition |
| --- | --- |
| AI not allowed | The activity must be done without generative tools. |
| AI allowed with restrictions | The tool can be used only in certain phases. |
| AI allowed with disclosure | The student may use it but must explain purpose, how, and what was verified. |
| AI required | The activity deliberately incorporates the analysis or use of the tool. |

Rules must be communicated before the activity and connected to the learning evidence the professor needs to observe.

## Mistake 10. Sharing personal or sensitive information

An educational implementation does not require using student records, names or individual grades to prove value.

> **Do not share unnecessarily**
> 
> - Student names, IDs or records.
> - Individual grades.
> - Medical, financial or family data.
> - Personal identifiers or passwords.
> - Confidential documents without authorization.

When work or examples are used, they must be anonymized. To prepare an institutional demo you can use curricula, program maps, syllabi, methodology, values and authorized representative materials.

## Mistake 11. Requesting too many documents without explaining their use

Academic personalization needs context, but the request must be clear and proportional. Indiscriminate access to full repositories can create concern and delay implementation.

- Which documents are needed and for what.
- Which formats are accepted.
- Which information must be removed.
- Who will have access and for how long.
- How deletion can be requested.
- Which materials should not be sent.

For a first demo, logo, colors, values, methodology, curricula, syllabi from two programs, representative materials, use cases and objectives may be enough.

## Mistake 12. Requiring an NDA for any conversation

Confidentiality is important, but turning an NDA into a requirement for a first meeting adds unnecessary friction. The initial conversation can happen without sharing strategic materials.

A mutual confidentiality agreement can be offered when the institution requests it, the documents are proprietary, the material is not public, the information has strategic value, or legal owners consider it necessary.

## Mistake 13. Building a demo with public information without authorization

It may look faster to search programs, colors and content on the university website. However, that information may be outdated, incomplete or fail to reflect the scope the institution wants to evaluate.

An institutional demo must be built with documents provided or authorized by the institution. This allows using correct versions, selecting relevant programs, respecting internal guidelines, avoiding assumptions, keeping traceability and aligning the experience with the agreed goal.

## Mistake 14. Confusing a personalized demo with a full rollout

A functional demo lets you visualize and evaluate the experience, but does not mean the platform is deployed to the entire university community.

| Functional demo | Full rollout |
| --- | --- |
| Visual identity and two programs | Final scope and number of users. |
| Representative tutors and examples | Programs and area owners. |
| Limited access for evaluators | Institutional schedule and training. |
| Functional walkthrough | Commercial model and implementation terms. |

## Mistake 15. Delaying the project for months before showing anything functional

The opposite extreme is also common: the university tries to solve the entire strategy, policy and rollout before visualizing the experience. This can generate long meetings without enough practical information.

> **With Genialoh**
> 
> - A first 30-minute meeting.
> - Selection of two programs.
> - Delivery of authorized documents.
> - Functional demo in less than 14 days from complete documentation.
> - Personalized 30–60 minute presentation.
> - 30-day access for up to five authorized people.

## Mistake 16. Measuring only users or conversations

A high usage number may look positive, but does not explain what value is being generated. It also does not tell whether users understood the rules, found useful answers or need training.

| Dimension | Sample metrics |
| --- | --- |
| Adoption | Invited users, sign-ins, frequency, participation by program. |
| Usefulness | Faculty and student perception, adapted activities, needs met. |
| Quality | Clarity, relevance, observed errors, needed interventions. |
| Implementation | Preparation time, documentation, training, support, adjustments. |
| Decision | Interest in continuing, additional programs, proposal, next owner and timeline. |

## Mistake 17. Presenting correlation as causation

A university may observe that students who use the platform most also achieve better results. This relationship can be interesting but does not automatically prove that AI caused the improvement.

- More motivated students use more resources.
- Students with better habits participate more.
- Certain professors promote both usage and best practices.
- Programs have different conditions.
- The sample may be too small.

Internal evidence must be described as such and not presented as independent scientific research.

## Mistake 18. Not defining what happens after the demo

A demo can spark interest but lose momentum if nobody defines the next action. After the presentation, it should be recorded whether the institution wants to move forward, needs follow-up, or does not want to continue for now.

- Meeting with academic direction.
- Technical evaluation.
- Faculty feedback.
- Presentation to leadership.
- Commercial model review.
- Proposal request.
- Implementation decision.

## Mistake 19. Framing the first meeting as a filter

An institution should not feel it needs to meet a minimum size, have approved budget or pass an evaluation to learn about the proposal. The first meeting must be available to any interested institution.

Its purpose is to present Genialoh, explain institutional AI, show capabilities, learn about the university, understand its needs, resolve questions and determine if it wants to move forward with a demo. The institution decides after seeing the solution.

## Mistake 20. Over-promising

AI can add value, but responsible communication must avoid claims like eliminating all errors, guaranteeing better grades, replacing professors, perfectly personalizing learning, solving any academic problem or automatically complying with any regulation.

It is preferable to talk about support, contextualization, guidance, observed patterns, analytics, training, implementation, optimization and institutional evaluation. Credibility grows when the university clearly understands what it will receive, what it must provide and what decisions it will need to make.

## How to avoid these mistakes: recommended path

| Step | Action |
| --- | --- |
| 1 | Book a first meeting: the institution learns about the platform and explains its needs. |
| 2 | Define a goal: pick the problem and use cases to visualize. |
| 3 | Choose two programs: they can be two degrees or areas. |
| 4 | Deliver authorized documentation without sensitive data. |
| 5 | Review a functional demo prepared in less than 14 days. |
| 6 | Involve academic, technical and commercial owners. |
| 7 | Evaluate for 30 days with up to five authorized people. |
| 8 | Record findings, questions, risks and use cases. |
| 9 | Define the next action: proceed, adjust, keep evaluating or postpone. |
| 10 | Implement with training, support and adoption monitoring. |

## Checklist before implementing AI in a university

> **The institution should be able to answer yes**
> 
> - Is there a defined problem?
> - Is there an executive sponsor?
> - Was an operational owner appointed?
> - Is academic direction involved?
> - Were two initial programs selected?
> - Are use cases documented?
> - Are provisional guidelines in place?
> - Will professors receive training?
> - Can the platform be contextualized?
> - Does the experience reflect institutional identity?
> - Is it known which documents are needed?
> - Have personal sensitive data been excluded?
> - Is there an access and deletion process?
> - Were metrics defined?
> - Is there an evaluation date?
> - Does each opportunity have a next action?

## Frequently asked questions

### What is the most common mistake when implementing AI?

Starting from a tool without defining the problem, owners and evaluation criteria. This makes it hard to prove value and sustain adoption.

### Should the university start with its entire community?

No. Starting with two programs allows personalization, training and learning before expanding the scope.

### Which areas should participate?

Leadership, academic direction, technology, data protection, innovation, coordinations and professors, according to the scope.

### Is student data needed to prepare a demo?

No. Use authorized academic documents without personal or sensitive information.

### Is an NDA required?

Not for the first conversation. A mutual NDA can be used when the institution requests it or when documents are strategic, proprietary or non-public.

### How long does Genialoh take to prepare a demo?

Genialoh can prepare a personalized functional demo in less than 14 days from receiving the complete documentation. The timeline can be adjusted based on the agreed scope.

### What does the demo include?

Institutional identity, logo, colors, academic context, two programs, representative tutors, interaction examples and a functional walkthrough.

### How long can it be evaluated?

After the presentation, up to five authorized people can explore the demo for 30 calendar days.

### Are implementation and training included?

Implementation, teacher training, support and adoption monitoring are included in the confirmed commercial models.

### How should success be measured?

Through a combination of adoption, usefulness, quality, risks, implementation and institutional decision, not only by the number of sign-ins.

## Avoid months of uncertainty before seeing a real experience

A university does not need to choose between improvised adoption and a project that takes months to produce a visible result. It can start with a controlled process.

> **With Genialoh you can**
> 
> - A 30-minute meeting.
> - Two selected programs.
> - Authorized institutional documents.
> - A functional demo prepared in less than 14 days.
> - A personalized presentation.
> - 30-day access for up to five evaluators.
> - An informed decision about implementation.

Genialoh lets you create a platform with the university's name, logo and colors, contextualized with its programs, methodology and values, and accompanied by analytics, training, implementation and support. The less-than-14-day timeline starts when we receive the complete documentation and can be adjusted based on the agreed scope.