TL;DR

Most organisations want AI but aren’t ready for it. Real AI adoption means either augmenting employees with copilots or creating autonomous agents—but both require clean data, documented processes, and strong governance. If workflows live in Excel, approvals happen in chats, or data is scattered, AI has nothing reliable to operate on. Once processes are structured and people understand how to work with AI, the organisation can finally unlock decision intelligence, safe automation, and meaningful impact. AI doesn’t fail because the model is bad—it fails because the foundations aren’t there. Build readiness first, value follows.

What companies get wrong about “adding AI”

Every organisation wants to “implement AI”, but few can describe what that actually means.

Is it adding Copilot to meetings?
Automating tasks with Power Automate?
Building agents that take decisions on your behalf?

The reality is that most companies don’t yet know what they want to achieve with AI, and even fewer are ready for it. Not because they lack tools, but because their people, processes, and technology aren’t structured for AI to operate safely, reliably, and at scale.

This post breaks down, in practical terms, what organisations truly need for AI-enabled business processes, the common pitfalls we see again and again, and a clear framework your organisation can actually use to get started.

What “adding AI” really means

When most teams say they want to “add AI”, they usually mean one of two things, and each has very different requirements.

1. Extend the worker (AI-augmented work)

This is where copilots and conversational assistants truly shine: helping employees search company knowledge, summarise decisions, retrieve documents, and take routine actions. But this only works if:

• the AI actually understands your business data,

• the data is structured and governed, and

• the agent is not given decision rights that introduce risk.

The system must understand the company’s knowledge, not just respond to prompts.

2. Create autonomous workflows (AI agents)

This is the more advanced path: agents that make limited decisions, move work between systems, and act without constant human supervision.

But autonomy does not mean freedom. Governance is key. An agent should operate within a clearly defined scope and can only take business-critical decisions when it’s given clear criteria.

This distinction matters because it forces organisations to re-examine how they work. If your processes are unclear, inconsistent, or undocumented, AI will reveal that very quickly.

Before you automate anything, understand the real process

One of the first questions we ask in readiness workshops is deceptively simple:
“How does this process actually work today?”

Almost always, the answer reveals a gap between intention and reality:

• Sales opportunities tracked in Excel

• Approval steps handled informally in Teams chats

• Documents scattered across personal drives

• Edge cases handled by “whoever knows how to do it”

This is where it all breaks down. AI cannot automate a process if even humans cannot describe it. If a process isn’t documented, it's technical debt.

Another red flag is when organisations that want to “keep the process exactly as it is” and simply add AI on top. AI doesn’t work that way. If the process itself is inefficient, undocumented, or built on manual workarounds, no amount of automation will save it.

To get real value, the process must be worth automating in the first place, ideally delivering a 10x improvement when reimagined with AI.

The hidden bottleneck: your data

Every AI workflow, from copilots to autonomous agents, relies on data being structured, governed, consistent, discoverable, and stored in systems designed for long-term work.

If you’re tracking key business processes in Excel, you’re not AI-ready. Excel is brilliant for calculations, bu it is not designed for workflow execution, audit trails, role-based access, entity relationships, or system-to-system integration.

Excel is unstructured data. You cannot build AI on manual data.

The good news is that Microsoft’s systems are AI-ready by design:

• Dynamics 365 for structured sales and service processes

• Dataverse for the unified data backbone

• SharePoint for document lifecycle and governance

• Teams and Loop for shared context and collaboration

If your processes live outside these systems, your AI will operate without context, or worse, without safety.

And if your data sits in old on-premise servers? Connecting them to modern AI systems becomes slow, fragile, and expensive. AI thrives in the cloud because the cloud creates the structure AI needs.

Designing workflows where AI and humans work together, safely

Once processes are structured and data is governed, the next question is:
what should AI do, and what should humans do?

There’s a simple rule of thumb:

• High-impact, high-risk, or ambiguous decisions → human

• High-volume, low-risk, routine steps → AI

This is where human-in-the-loop design becomes essential. A well-designed AI workflow should:

• Define exactly where humans intervene

• Log every AI action for traceability

• Provide confidence scores and explanations

• Avoid overwhelming people with unnecessary alerts

• Keep the final accountability with the human owner

Humans should use judgement, handle exceptions, and ensure ethical and correct outcomes. AI should do the repetitive work, the data consolidation, and the first pass of tasks.

AI readiness is also about people, not just systems

One of the most underestimated aspects of AI readiness is human behaviour. For AI to work as intended, business users must:

• Be curious

• Know how to break their work into steps

• Be willing to adapt workflows

• Understand where data lives

• Ask questions and refine prompts

• Avoid bypassing the process when “it’s easier to do it my way”

Processes fail when people resist the change because they  don’t understand the “why”. And they fail just as quickly when employees work around the automation or keep using personal storage instead of governed systems.

AI introduction is as much a cultural shift as it is a technical programme.

What you can finally ask once AI-readiness is achieved

Once the foundations are in place, people begin asking questions that were previously impossible:

“Which of our suppliers pose the highest risk based on the last 90 days of invoices?”

“What decisions were made in the last project meeting, and who owns them?”

“Show me opportunities stuck for more than 30 days without activity.”

“Draft a customer update using the last three emails, the CRM history, and the contract.”

“Alert me when unusual patterns appear in our service requests.”

These are questions an agent, not a chatbot, can answer. But only if the process is structured and the data is clean.

AI doesn’t fail because the model is bad. It fails because the organisation isn’t ready

Before building agents, copilots, or automations, ask yourself:

• Would AI understand our processes, or would it get lost in exceptions?

• Is our data structured, governed, and accessible?

• Do our people know how to work with AI, not around it?

• Are we prepared to support safe, auditable, and reliable AI operations?

If the answer is “not yet”, you’re not alone. Most organisations are still early in their readiness journey. But once the foundations are there, AI value follows quickly, safely, and at scale.

Want to move from AI curiosity to real, measurable impact? Get in touch for an AI readiness workshop.  

‍