Many Power Platform AI projects often fail several months after the demo. Common issues include misclassified invoices entering enterprise resource planning (ERP) systems without review, or unexpected increases in licensing costs due to inadequate modelling of actual usage.
Copilot Studio and AI Builder are the AI layer on top of Power Platform. Copilot Studio has improved over the past few years, and AI Builder exceeds expectations for document extraction. The problem is the gap between what a Tuesday afternoon proof-of-concept suggests is possible and what a production system demands.
In this expert opinion, Petro Zabenko examines where Power Platform AI excels, where it struggles, and the best practices that successful teams follow before development begins.
1. What is the current state of AI adoption in Power Platform?
Many projects still run into trouble when moving from demo to production because of weak architecture, missing governance, unexpected costs, and overlooked edge cases. For IT leaders and architects, the main challenge is using artificial intelligence in a way that avoids solutions that are too hard to maintain.
2. What AI capabilities does Microsoft offer within Power Platform?
Microsoft has built AI into the platform in two ways: Copilot Studio for conversational AI agents and AI Builder for prebuilt models.
- Copilot Studio is a low-code platform for building AI-powered conversational agents. It supports natural language understanding, knowledge base grounding via SharePoint or external data sources, and orchestration across other Power Platform services. It's a tool for building internal assistants, customer-facing bots, and agentic workflows.
- AI Builder provides prebuilt AI models for document processing, sentiment analysis, and prompt-based actions accessible directly from Power Apps and Power Automate flows.
Both work well, but both have limits that only surface in production.
3. What Power Platform AI use cases work well at the MVP stage?
The first is internal helpdesk and HR bots. Keep the scope narrow. Password resets, leave policy questions, and ticket routing work because the topics are limited and the questions repeat. Internal users are also patient when the bot doesn't know the answer. The key is connecting it to a maintained knowledge source. A bot reading from a curated SharePoint library gives consistent answers.
The second is document processing with AI Builder. It works well on consistent documents such as invoices, purchase orders, and onboarding forms. Set a confidence threshold and route anything below it to a human reviewer. That one rule keeps the risk low.
The third is email classification in Power Automate. Build a flow that reads incoming emails and routes them to the right team. The logic is simple, and the risk is low. A person still reviews the output before anything happens.
4. What are the most common ways Power Platform AI projects fail in practice?
- The scope is too wide. Copilot Studio was not built to be a general-purpose reasoning engine. That is why, when a single bot is expected to cover HR policy, IT support, project management, and finance simultaneously, results degrade. Responses become inconsistent with risks of hallucinations, and the maintenance burden becomes unsustainable. The platform requires deliberate architecture: separate agents with distinct purposes, clear topic boundaries, and explicit fallback handling.
- Variable or unstructured document inputs. AI Builder struggles when documents vary in layout or formatting. Without confidence thresholds and exception handling, extraction errors slip into finance systems or compliance databases. Nobody notices until the damage is done.
- Underestimated Microsoft Power Platform licensing costs. AI Builder runs on a credit-based model that adds up fast at enterprise volumes. Copilot Studio is billed per message — a high-traffic bot can exceed licensing expectations by an order of magnitude. Model realistic production volumes before committing to a rollout.
- Treating AI output as trusted data. If a support ticket is misclassified, an invoice amount is misread, or a product code is made up and sent to an ERP or CRM without checks, it can cause serious problems later. Every production process that uses AI output should treat it as untrusted. It means checking the structure, reviewing confidence scores if possible, and setting up ways to handle exceptions.
- Launching without a Power Platform governance framework. When Power Platform is widely accessible and oversight is thin, shadow automation spreads fast. People build flows that pass sensitive data through AI models, connect AI agents to external APIs without a security review, or create automations that IT never sees. For organisations subject to GDPR or data residency requirements, Microsoft's data handling settings must be configured before adoption scales. Not after.
5. What does a practical decision framework for Power Platform AI deployments look like?
Three questions should be answered before any AI feature is deployed to production. They provide a simple transition from idea to implementation.
- Is the scope bound? Describe in one sentence what this AI feature does and what it doesn't. If the answer needs "and also" more than once, narrow the scope before building starts.
- Is a human reviewing output before it reaches a system of record? AI output flowing directly into a business system without human review or programmatic validation is a risk that needs to be explicitly modelled and accepted, not discovered after the fact.
- Has licensing been modelled at production volume? Use real production volume estimates, with a 3x buffer for adoption growth. If the numbers haven't been run, they should be before the project moves forward.
Power Platform AI readiness framework
| Readiness | Recommended use cases |
|---|---|
| Green |
|
| Amber |
|
| Red |
|
6. What does it take to get real value from Power Platform AI?
Getting value from Power Platform AI is less about the technology and more about the decisions made before anyone opens Copilot Studio.
The projects that work all look the same. Small scope. Clear boundaries. Human review from the start. Costs checked against real numbers. Governance is set up before it becomes a problem.
The ones that struggle usually made the same mistake — they moved fast, kept adding scope, and treated artificial intelligence output as reliable before they had any evidence it was. By the time problems surfaced, they were already embedded in production systems.
The tools are good. Use them carefully.
Related insights
The breadth of knowledge and understanding that ELEKS has within its walls allows us to leverage that expertise to make superior deliverables for our customers. When you work with ELEKS, you are working with the top 1% of the aptitude and engineering excellence of the whole country.
Right from the start, we really liked ELEKS’ commitment and engagement. They came to us with their best people to try to understand our context, our business idea, and developed the first prototype with us. They were very professional and very customer oriented. I think, without ELEKS it probably would not have been possible to have such a successful product in such a short period of time.
ELEKS has been involved in the development of a number of our consumer-facing websites and mobile applications that allow our customers to easily track their shipments, get the information they need as well as stay in touch with us. We’ve appreciated the level of ELEKS’ expertise, responsiveness and attention to details.
