---
title: "Most AI Projects Fail Before the Technology Is Even Selected"
id: "20036"
type: "post"
slug: "ai-readiness-assessment"
published_at: "2026-06-17T12:07:49+00:00"
modified_at: "2026-07-06T11:12:18+00:00"
url: "https://volt-technologies.com/post/ai-readiness-assessment/"
markdown_url: "https://volt-technologies.com/post/ai-readiness-assessment.md"
excerpt: "Modernize manufacturing AR with Microsoft Copilot. Improve cash visibility reduce manual work and apply payments faster."
taxonomy_category:
  - "Business Central"
  - "Copilot"
  - "Post"
---

# Most AI Projects Fail Before the Technology Is Even Selected

## Introduction

We have run enough AI readiness assessments to recognize the pattern. A mid-market company has decided to adopt an AI. They evaluate platforms, pick one, kick off implementation, and six months later, they are stuck. Not because technology failed. Because the organization was never ready to receive it.

McKinsey research shows that 70% of change management programs, specifically system implementations, fail to go live, and the cause is people’s adoption, not the technology. MIT research shows that 95% of AI projects fail to meet customer expectations. Neither of those numbers is a technology problem. Both are a readiness problem.

In a recent video, I walked through the exact framework we use at Volt Technologies when we assess AI readiness at a company. It evaluates three layers in a specific order: people, then process, then technology. Get the order right and implementation moves fast. Skip a layer and you will spend the back half of the project fixing what should have been addressed before day one.

Table of Contents

**See the complete AI readiness framework in action, including how we score each layer with real clients.**

## What an AI Readiness Assessment Actually Measures

An AI readiness assessment is a structured evaluation of whether an organization has the people, processes, and technology infrastructure in place to successfully adopt and scale artificial intelligence. In my experience it produces a scored view of gaps across three dimensions and a sequenced roadmap for closing them before implementation begins.

To me, it is not a software shortlist or a vendor comparison. It is an honest internal audit of three questions: Are your people prepared for this change? Are your operations documented well enough for AI to execute against them? Is your technology stack accessible to modern AI tools?

The companies that answer all three honestly before they buy anything are the ones that go live on time and get the outcomes they were expecting. The companies that skip straight to the platform decision are the ones producing the failure statistics above.

## **Why You Have to Assess in This Order**

The sequence matters because each layer is a prerequisite for the next. You cannot document a process that lives entirely in one person’s head. You cannot connect an AI tool to a system with no API access. You cannot expect clean AI outputs from data that have never been audited.

When we score a company across all three layers, the lowest score sets the ceiling on what AI can deliver. A business with enterprise-grade cloud infrastructure, but no documented processes will not get reliable AI performance, because there is nothing reliable for the AI to execute against. The infrastructure does not compensate for the missing layer below it.

## **Layer One: Are Your People Ready?**

### **The Real Question Is Whether They Know Their Own Jobs**

When we start a people assessment, the first thing we look at is not AI literacy. It is a role clarity. Do your employees have clearly defined job descriptions? Does your org chart reflect what people actually do, or what their titles used to say three years ago? Are responsibilities documented anywhere other than in someone’s memory?

In most mid-market businesses, the honest answer is no. We ask for a documented process and we get a name. “Talk to Karen. She handles that.” That is not a process. That is a single point of failure. Before any AI system can assist with that workflow, someone must write down what Karen actually does, in enough detail that a defined trigger-action logic can replicate it.

### **Find Your Champions Before You Launch**

The second thing we look for is champions: people inside the organization who are genuinely excited about AI, credible with their colleagues, and willing to get hands-on with the technology before the broader rollout begins.

A top-down mandate from a CIO does not drive adoption. A peer who has already used the tool, seen it work, and can speak to it in practical terms does. Getting those people identified early, running them through proof-of-concept sessions, and building their confidence before go-live creates a multiplier effect that no training program can replicate. This is consistently one of the strongest predictors of whether an implementation gains real traction.

### **Map Skill Gaps by Department, Not Company-Wide**

Finance, warehouse operations, purchasing, and customer service each come into an AI rollout with different baselines, different resistance points, and different use cases. A single training program cannot address all of them. The right approach is to map gaps department by department before implementation begins: understanding what each team needs to do differently and what support they will need to bridge that gap. This work takes time upfront. It saves significantly more time on the back end.

## **Layer Two: Are Your Processes Documented?**

### **If the Process Lives in Someone’s Head, AI Cannot Use It**

This is where I see most companies hit the first real obstacle. The honest picture I find in a lot of mid-market organizations is that critical business processes exist as institutional memory rather than written procedure. How orders move through the system, how invoices get approved, how inventory gets reconciled, and these workflows run in people’s heads, not in written SOPs.

AI operates on defined logic: when this condition is met, take this action. It cannot interview your senior buyer about what she usually does. It cannot infer the right response from fifteen years of experience on the job. It needs a trigger and an action, written down, tested, and covering the full range of scenarios your business encounters. Standard operating procedures are not bureaucratic overhead. They are the raw material AI needs to function.

### **Triggers and Actions: How to Document for AI**

Every business process, stripped to its core, is a trigger and an action. A purchase order arrives, and inventory levels are checked. An invoice clears approval and a payment is scheduled. A customer service ticket is submitted, and a priority tier is assigned. When your team can articulate operations in those terms across sales, purchasing, finance, and fulfillment, you can configure AI to execute them reliably.

For organizations on Microsoft [Dynamics 365 Business Central](https://volt-technologies.com/dynamics-365-business-central/)
, this maps directly to what is available today. The Sales Order Agent, Payables Agent, and Expense Agent are each built to execute specific trigger-action patterns inside the platform. The more precisely those patterns are defined in your SOPs, the faster those agents can be deployed, and the more accurately they will perform.

### **The Cost of Skipping This Step**

Projects that walk into implementation without documented processes run two to three times longer than projects that come in with SOPs already in place. Every hour spent in discovery, figuring out what a team does so AI can be configured to assist, is an hour that could have been spent on configuration, testing, and adoption. The MIT figure that 95% of AI projects fail to meet expectations is downstream of this gap more than almost anything else.

## **Layer Three: Is Your Technology Stack Ready?**

### **Bad Data Is Not a Small Problem**

IBM estimates that bad data costs businesses $3.1 trillion (about $9,500 per person in the US) per year globally. The reason that number is so large is that AI scales and accelerates whatever it is given. Feed it clean, consistent, well-structured data, and it makes reliable decisions fast. Feed it the kind of data that most mid-market companies have when they have never run a formal data audit, duplicate records, inconsistent field values, siloed systems that have never been integrated, and it produces unreliable outputs at speed, across your operations, with no human in the loop to catch the errors before they compound.

A data audit before implementation is not optional work you can defer. It is the step that determines whether your AI investment produces value or produces problems at a scale.

### **Your Systems Have to Be AI-Accessible**

AI tools operate by hooking into your transactional systems: reading data, performing actions, surfacing results. That requires either cloud-native infrastructure or modern API access. A legacy on-premises system with no API layer is functionally invisible to most AI applications. You can build workarounds, but workarounds add cost, latency, and failure points.

Microsoft Dynamics 365 Business Central is built for exactly this. It is cloud-native, integrates directly with Microsoft Copilot, and exposes an open API architecture that allows AI agents to act inside the system rather than just report on it. The Autonomous Agents available in Business Central, the Sales Order Agent, Payables Agent, Expense Agent, and Agent Designer, operate natively within the platform, requiring no third-party middleware.

### **End-of-Life Platforms Cut You Off from AI Investment**

Dynamics NAV reached mainstream support end in 2023. Microsoft’s AI investment is concentrated in the platforms it is actively building: Business Central, Copilot, the Business Central MCP Server, and the Autonomous Agent suite. None of those capabilities are being backported to NAV. Organizations still running NAV are on a platform that current AI tooling cannot effectively reach. Migrating to Business Central is the move that puts you inside the environment where these capabilities exist and where new ones are being released.

### **The Scorecard: What Ready Actually Looks Like**

The output of my framework is a scorecard across all three layers, each scored independently. The goal is to max out every category before implementation begins.

**People:**Champions identified by name and department. Skill gap plans built per team. Job descriptions accurate. Org chart reflects real responsibilities.

**Process:**SOPs documented across sales, purchasing, finance, fulfillment, and customer service. Trigger-action logic written for all key workflows. No critical processes that exist only in someone’s head.

**Technology:**Data audited and centralized. Systems cloud-enabled or migrated. No critical operations running on end-of-life platforms.

When the scorecard shows gaps, those gaps have to be closed before implementation begins, not during it, not after it. The lowest-scoring layer is always the constraint. Investing in the other two will not compensate for a foundational gap in the weakest one.

## **Key Takeaways**

AI readiness is a sequencing problem, not a technology problem. The specific takeaways from the framework:

- **People first.**AI implementation is a change management project. The organizations that treat it that way, identifying champions, mapping skill gaps, and establishing role clarity before go-live, are the ones that see adoption. The ones that skip this layer encounter the consequences after the budget has moved.

- **Process before platform.**AI executes against defined logic. If your processes live in people’s heads, you are not ready for AI. Write the SOPs and trigger-action documentation first. Buy the tool second.

- **Technology is the third layer, not the first.**Clean data and cloud-accessible systems are prerequisites. Business Central gives you the AI-native infrastructure that makes this layer straightforward. A legacy on-premises ERP makes it extremely difficult.

- **Score honestly.**The scorecard only works if the assessment is accurate. A gap that is rationalized away at the readiness stage becomes a project failure at the implementation stage.

The framework above comes directly from a video I recorded walking through our complete AI readiness assessment process, including how we score each layer and what the remediation roadmap looks like when a company is not yet ready.

If you want to run this assessment against your own organization, or if you are evaluating whether Microsoft Dynamics 365 Business Central is the right foundation for your AI roadmap, reach out to the Volt Technologies team. We will walk through it with you.

## **How Volt Technologies Approaches This**

Volt Technologies is a 10x Microsoft Inner Circle partner, top 1% of Microsoft Business Applications partners worldwide, with 30 years of Microsoft ERP implementation experience across [apparel](https://volt-technologies.com/apparel/)
, [distribution](https://volt-technologies.com/distribution/)
, [manufacturing](https://volt-technologies.com/manufacturing/)
, and professional services. Clients include Rag and Bone, Marc Jacobs, Vera Bradley, Groove Life, and 5.11 Tactical.

What that experience means when we run an AI readiness assessment:

- **We assess against what we have seen fail.**People issues that derail go-lives, process gaps that double project timelines, data problems that surface three weeks after launch. We know where to look because we have seen all of it.

- **We work hands-on throughout.**We do not hand off a framework document. We work through the assessment with your team, identify specific gaps, and build a sequenced roadmap to close them.

- **For Business Central clients,**the roadmap connects to real, available tools: Copilot, the Business Central MCP Server, the Sales Order Agent, Payables Agent, Expense Agent, and Agent Designer.

- **For companies still on Dynamics NAV or evaluating Business Central,**the assessment starts the same way regardless of where you are in the stack.

## Frequently Asked Questions

What does an AI readiness assessment include?

It evaluates people (role clarity, skill gaps, champion identification), process (SOP coverage, trigger-action documentation, workflow completeness), and technology (data quality, system accessibility, end-of-life risk). The output is a scored view of each layer and a sequenced plan for closing gaps before implementation.

Why do most AI projects fail to meet expectations?

The root cause is almost never the technology. McKinsey data shows 70% of system implementations fail to go live due to people adoption gaps. MIT research shows 95% of AI projects miss expectations, primarily because the processes AI is meant to execute have not been defined. Both are addressable before a platform is selected.

How long does becoming AI-ready take?

Companies with clean data, documented processes, and cloud-native systems can move to implementation in weeks. Companies with legacy platforms, undocumented workflows, and no change management foundation need three to six months of preparation work before AI tooling delivers reliable value.

Does Business Central support AI natively?

Yes. Business Central includes native Copilot features, supports the Business Central MCP Server, and hosts the Sales Order Agent, Payables Agent, Expense Agent, and Agent Designer, all running within the platform without third-party middleware.

What if our ERP is not cloud-enabled?

If your system has no cloud access or API connectivity, AI tools cannot reliably connect to it. Migrating to a cloud-native platform like Business Central resolves this and brings your stack inside Microsoft's active AI development cycle.

How is Volt different from other Microsoft partners?

Thirty years of Microsoft ERP implementation experience, 10 Inner Circle designations, and vertical depth in apparel, distribution, and manufacturing. Our engagement model is hands-on across the full lifecycle. We do not deliver a readiness framework and disengage.

## **Conclusion**

Most companies that struggle with AI did not pick the wrong tool. They picked the right tool at the wrong time: before their people were prepared, before their processes were documented, before their data and systems were ready to support it.

The framework in the video above exists to prevent that. Assess your people, document your processes, audit your technology, in that order, scored honestly. When all three layers are solid, AI implementation is fast and the results are predictable. When any one of them is missing, you will find the gap at the worst possible moment.

We have run this assessment with companies across apparel, distribution, manufacturing, and professional services. The ones that invested in readiness first got to value faster than the ones that tried to fix it mid-implementation. If you want to know where your organization stands, reach out. That conversation is where it starts.

#### Mason Whitaker