---
title: "Five practical obstacles to useful AI | SamuylovAI"
description: "Five recurring obstacles beyond model capability: data foundations, scope, staff readiness, security and interface."
url: https://www.samuylov.ai/stories/why-does-genai-not-work
language: en
---

Article · October 25, 2023

# Five practical obstacles to useful AI

When an AI project disappoints, model capability is one question to examine. Five other obstacles recur in our work: data foundations, scope, staff readiness, security and interface.

These areas can be reviewed and improved alongside the model itself.

## Unreliable or unclear data

A useful data foundation includes reliable sources, clear field definitions, ownership and access. Without that context, an AI system may produce a plausible answer that does not fit the business question.

## A use case that misses the constraint

Choose a use case by the workflow’s constraint and business value. If a sales team has more leads than it can handle, generating additional leads may not help. Look at the step that limits the result and whether AI is suited to improving it.

## Nobody was taught to use it

Training should use real work: which tasks are suitable, how to check an output and when to seek a human decision. Prompt guidance can help, but staff also need practice with the workflow and its limits.

## Security arrived last

Set data-use rules before rollout: where the system runs, what it retains, who may access logs and which information is excluded. Staff need those rules before using the system with company or customer data.

## An interface that adds unnecessary work

An extra login or a difficult route to the tool can discourage use even when the model performs well. Integrate the system into a suitable workflow and evaluate whether people can complete the task with it.

## Review the system around the model

Data, scope, training, security and interface design are practical parts of the delivery plan. Review them with the people who will use and operate the system.

Our three service lines support different parts of that work:

1. [**Advisory & Implementation**](https://www.samuylov.ai/consulting) – the data foundation and the security practices GenAI depends on, deciding where it earns its keep, and taking it into production. Workshops bring the team along.
2. [**Agents-as-a-Service**](https://www.samuylov.ai/ai-as-a-service) – once a solution works, we run it as a managed subscription: hosted, monitored and improved.
3. [**Design & Experience**](https://www.samuylov.ai/design) – the interface and experience that help people understand and use the system.

Review these questions before committing to a build, and keep them part of delivery. See [advisory & implementation](https://www.samuylov.ai/consulting), or [Start a conversation](https://www.samuylov.ai/contact).

[Download as PDF](https://www.samuylov.ai/stories/why-does-genai-not-work/pdf)

## Have a project in mind?

Tell us what you are building or improving. We'll help you choose a practical next step.

[Book an intro call](https://www.samuylov.ai/contact?book=1&topic=other)

## More stories

[All stories](https://www.samuylov.ai/stories)

[A Slack knowledge assistant for Nezasa, built in one week](https://www.samuylov.ai/stories/internal-knowledge-assistant-for-a-travel-technology-sme) – Case study · Nezasa AG · April 2024 · Questions answered without involving a product expert: 55% · Capacity benefit estimated by the client: ≈2 FTE

[An AI workshop for AXA France](https://www.samuylov.ai/stories/samuylovai-and-recursion-labs-launch-first-workshop-for-axa-france) – Article · July 2023
