Vasyl Grygoriev,
Lizard Soft CEO

In the previous part How to Grow the Power User | Change Management Roadmap, I noted that moving on to the working with people stage is only worth it when the basic homework at the business level has been completed. So today, let’s talk about that.

Now, Business Owners and Managers are being bombarded with a continuous flow of information about AI, agents, autonomous agents, and the conclusion that those who fail to implement it will certainly lose the market. At the same time, Partners come with new proposals, pilots, demos, examples, and their own vision of what's best for you. In such an environment, it’s very easy to fall into the trap of time pressure: to start acting in a hurry, to develop solutions hastily, and to confuse technological novelty with real business impact.

In our analysis, we grouped the main reasons why AI initiatives do not take off or do not give the expected result into four blocks: Business, People, Technology, Processes.

So today, about the Business category.

[Significant impact] Unclear business goal and lack of measurable value

Many initiatives in generative AI start as a response to the We need our own ChatGPT and Let’s make an AI assistant for everyone trend.

It sounds topical, but this formulation usually does not answer the main question: which business process are we changing, and which indicator do we want to improve?

Ultimately, the solution ends up being separated from real operational activities. It quickly turns into either an optional tool that is occasionally used or an R&D activity without a clear link to P&L. The effect is not measured, and support costs are the first to be cut.

And this is quite logical: it’s quite difficult to secure the budget for something whose value to the business has not been proven.

[Significant impact] We are not able / not ready to measure the effect

There are types of economic effects that are relatively easy to calculate: for example, when the Company reduces costs for paper, external services, or the work of Contractors.

But as soon as it comes to saving Employees' time, improving quality, reducing the number of errors, or speeding up decision-making, many say this is almost impossible to measure.

To this, I usually ask a counter question: How do you measure the effectiveness of People and processes today? Do you have these metrics? Does Management trust them? Are management decisions made on their basis?

If there is no clear answer to this, then the issue is not with AI. Just those attempts to assess the AI effectiveness very quickly highlight the weaknesses in management that exist in the Company.

[High Impact] Lack of a Strong Business Sponsor and Implementation Team

If an initiative does not have a defined Business Owner — i.e., a Person who is responsible for the process and for changing the targets — it almost inevitably turns into a technical experiment.

IT can create a tool. But IT itself does not change the operating model. It won't rewrite regulations, revise roles, rebuild checkpoints, change KPIs, or create incentives for Users.

The same applies to the Team. Complex systems are not to be implemented by IT alone. But they're not to be implemented by the Contractor alone either. A cross-functional Team must be formed within the Company that takes responsibility not for the fact of launch itself, but for the business result.

[Medium Impact] Cases are chosen for their spectacle, not for ROI

In many Companies, the first budgets for GenAI go to Sales and Marketing. That’s understandable: such scenarios are easy to show, quickly demonstrate, and beautifully packaged in a presentation. Content generation, commercial materials, support for Sales Managers, personalization of communications — all this looks convincing.

But the most predictable and scalable economic impact very often lies not there. In practice, it often appears in the Back office. It's there that automation is directly converted into cost reduction, reduction of time for performing typical operations, and improvement of quality.

The issue is that such cases are less spectacular. They are more difficult to sell internally at the start. But if the first project is chosen not for economy, but for showiness, and the result isn't impressive, then launching the next initiative turns out to be much more difficult.

[Indirect impact] Pilot trap

One of the most common situations looks like this: The Company successfully conducts a PoC or pilot, confirms the technical feasibility of the solution, everyone is satisfied with the demo. However, the case never reaches regular use.

Why does this happen? Most often, the reasons are quite evident:

• The requirements of the Legal Department, information security, or Trade Unions were not considered.
• In a real environment, the solution filed to handle the volume of data, load, or stability requirements.
• The Business Process Owner, who has the authority to change regulations, roles, KPIs, and control rules, was not involved.
• The pilot tested an ideal scenario but did not reflect the real process — with exceptions, incomplete data, dependencies on other systems, and necessary integrations.

Here, it’s important to remember a simple thing: demonstration of capabilities does not equal guaranteed operational effectiveness. 

[Medium impact] There's no transparent financial model for production

At the pilot stage, the solution almost always looks inexpensive. The workload is small, some of the work is done manually, and a significant part of the costs remains bracketed off.

But as soon as the Company moves on to preparing the solution for production, a completely different economy appears. And it already needs to be considered: payment for the use of LLM models, knowledge base searching infrastructure, auditing and data storage policies, monitoring, etc.

Without a transparent total cost of ownership (TCO) model at the start, the Company does not see the main thing: how much the solution actually costs per unit of use, where its payback limit is, and what budget is needed to maintain quality and reliability at scale. And it's at this stage that the project often stops.

[Medium impact] Simplified choice: in-house development or ready-made solution

Very often, the discussion is reduced to a simplified concept: do it yourself or buy ready-made. But an enterprise GenAI solution is not just a model and a chat interface. It's an entire system that must be deeply embedded in the Company’s daily work: data, access rights, integrations with corporate systems, quality control of responses, etc.

When a Company chooses the do-it-yourself path without a full assessment of the scope of work and the full cost of ownership, the project often enters a mode of endless refinement. First, a demo works. Then security requirements appear. Next, integrations. Then scaling. And at each stage, it becomes obvious that new competencies, new resources, and constant attention are needed.

There's one more underestimated fact: language models and the tools around them change extremely quickly. What was considered top-notch a year ago is already outdated today. Therefore, a GenAI solution cannot simply be launched and left as is. If it's not developed, it does not remain as it is — it loses its competitiveness.

What comes with that

If we simplify all mentioned above to one conclusion, it would be very simple: most unsuccessful AI initiatives do not start with bad technology, but with an incorrectly posed business question: it's not about what bot we need? But what process are we changing, what value do we want to get, and how will we measure it?

When there's an honest answer to those questions, it becomes much easier to:

• Choose the first case.
• Appoint a strong Sponsor.
• Assemble the right Team.
• Do not get stuck at the pilot level.
• Soberly calculate the economy.
• Understand where it's really worth building your own solution, and where it's more reasonable to use a ready-made one.

In the next publication, I will focus separately on the Technology and Processes categories. As even a correctly formulated business goal does not guarantee the result you expect. So then, the no less important part of the work begins.