Why Most AI Projects Quietly Fail

Read time: 4 mins

Welcome back everyone 👋

This week’s Automation Playbook covers:

⚙️ Most AI projects start in the wrong place

🎭 Practical AI is deliberately unexciting

💡 A real example of AI that actually works

Let’s get into it 👇

Before we dive in, a quick personal update. Today is my birthday!

I was born on a Saturday, and I love Saturday mornings!

What better way to spend a pre-birthday weekend than standing in the sunshine on the sidelines watching my son play football.

A beautiful away day in the Cotswolds, but hardly the rock n’ roll parties of my twenties!

My son's team is currently in a relegation fight.

I always say these tests are character-building, and how you respond is everything.

After all, it is won over the course of a season, not a single game.

But during this match, the frustration really got to him.

He is a striker, and he was so annoyed about how they were playing as a team that his head completely dropped.

A couple of his teammates even walked over to me on the sidelines to ask if he was okay.

Walking back to the car afterwards with him and some of his friends, he complained that he had been calling for the ball all game.

I had to give him a gentle reality check: "Well, I didn't hear you, and your teammates didn't hear you either."

What I have noticed from watching the teams at the top of this league is that they are incredibly vocal.

They have great communication.

They do the stuff that is not sexy.

They do the hard yards, the constant encouragement, and the relentless communication.

That is why they are the best teams.

This exact same dynamic is what I encounter in on a daily basis.

I regularly run consultancy sessions with teams within organisations.

These AI ‘squads’ are specifically put together to roll out AI adoption and transformation across the business.

Often, they come to session one completely frustrated with how their systems are operating.

They expect me to hand them a shiny new AI tool to magically fix their lack of cohesion and win the season overnight.

What is most interesting is that in session one, we completely avoid talking about specific AI tools.

Instead, we do what we have always done for the last 14 years.

We focus on the unsexy stuff.

The hard yards.

We spend the first session talking deeply about their current systems and processes.

What is not working.

What the frustrations are.

Crucially, what actually does work.

We get them communicating properly about the desired future state.

Only when we can understand exactly what they want a system to do can we prescribe the solution.

That could be a combination of automation, integrations, bespoke apps, or plugging in an existing SaaS product.

We get them to set their own user stories, PRDs and guard rails.

To ensure they do not fall into the trap of wasting hours deliberating over which tool is perfect.

Watching and waiting for the perfect solution with the relentless pace of change right now is not recommended.

When teams start with the tool instead of the problem, the project is designed to demonstrate tech rather than deliver outcomes.

They try to buy a quick solution instead of doing the hard yards as a team.

That is the gap most teams fall into.

AI is easy to showcase.

It is much harder to operationalise.

Most AI projects start in the wrong place

The typical approach looks like this:

1. Pick a tool

2. Run a pilot

3. Show something interesting

4. Hope it becomes useful

The problem is that none of those steps start with the business.

They start with technology.

So you end up with something that works in isolation, but does not connect to real workflows, real constraints, or real value.

I see this time and time again in conversations where the underlying mindset of business owners has shifted from a fear of missing out to a fear of wasting money.

Nugget #1: If AI starts with the tool, it usually ends as a demo.

Practical AI is deliberately unexciting

The AI that actually works rarely looks impressive.

It:

• Automates repetitive decisions

• Reduces manual effort in existing workflows

• Improves speed and consistency

• Integrates into systems people already use

• No dashboards for the sake of it.

• No novelty features.

• No extra steps.

Just quiet improvements that compound over time.

This is where most AI projects fail.

They aim for transformation instead of integration.

They try to replace thinking instead of supporting it.

And in doing so, they create more friction, not less.

Nugget #2: The best AI is the kind your team forgets is there.

A real example of AI that actually works

We recently helped a highly bespoke service business automate their estimation and proposal system.

Because every job they do is completely different, they assumed the process could not be automated.

Salespeople were gathering requirements in meetings and sending them to estimators via a scattergun mix of Teams chats, emails and even WhatsApp.

Estimators were left guessing.

Salespeople were left chasing.

Quotes were taking anywhere from 7 to 14 days to turn around, and every final estimate was being manually typed up into a Word document.

Instead of trying to replace the team with tech, we educated them on the 10-80-10 rule:

human at the start (10%), AI in the middle (80%), human at the end (10%).

We introduced AI into the "unsexy" process steps:

• An AI note-taker on sales calls automatically generates a standardised briefing document.

• Once reviewed by the salesperson, the brief is sent to a secure "AI brain" that the estimator can query for facts.

• The estimator inputs the numbers, and the AI automatically drafts the final Word document for the salesperson to review.

The result?

• Quote turnaround times plummeted from an average of 10 days down to under 24 hours.

• Human admin time dropped from half a day to just ten minutes on either side of the process.

• The human remains firmly in the loop, simply approving, amending, and signing off at key stages.

There was no big launch.

No attempt to replace human judgement.

Just prompt-driven automations working quietly in the background.

Nugget #3 : AI delivers value when it improves a system, not when it replaces one

What you can do today

🔹 Identify one repetitive decision your team makes daily

🔹 Map where time is spent in that process

🔹 Apply AI to one step, not the whole workflow

You do not need a grand AI strategy to get value.

You need to stop building for demonstration and start building for delivery.

AI is not magic, it’s just another tool.