Cost and ROI

How much does AI implementation cost

The honest answer is that it depends, but not in the way most vendors hope. Here is what the research says, where the money actually goes, and how to stop your budget becoming fiction.

By Zakir Hoosen, Director, Fuzzelogic Solutions. Board-level guidance in plain English.

The question every board asks first is how much. The question every board should ask second is how much twice, because most AI budgets are wrong the first time. Not slightly wrong. Wrong in a way that makes the CFO reach for the antacids.

I have seen boards approve budgets based on a vendor slide and a hopeful smile. Twelve months later, the project costs three times the original number, delivers half the promised value, and someone is explaining to the audit committee why the contingency was spent before Easter.

This article gives you what the research says about AI costs, where the money actually goes, and how to set a budget that survives contact with reality.

What the research says about cost overruns

The pattern is consistent across industries and geographies. First-year AI budgets do not hold.

Two thirds of programmes cost more than planned. That is not a failure of technology. It is a failure of estimation, and it is predictable. The costs that blow a budget are rarely the ones on the vendor proposal. They are the ones nobody listed: data preparation, integration with legacy systems, training the people who have to use it, and the rework nobody anticipated because the first version did not fit the business.

Read that again. Most companies are spending real money and getting little back. The problem is not that they chose the wrong tool. It is that they did not understand what they were buying.

Where the money actually goes

Most boards think the cost of AI is the software licence. That is the visible part. The rest is hidden, and it is usually larger.

The 10-20-70 rule is worth knowing here. It comes from BCG's research on what separates companies that get value from AI and those that do not.

1. 10% goes to algorithms and models.
2. 20% goes to technology and data infrastructure.
3. 70% goes to process change, people, and ways of working.

Seventy percent. That is the cost most budgets forget. It is the cost of redesigning a workflow, retraining a team, changing how decisions get made, and keeping the whole thing running when the first version breaks in ways nobody predicted.

A licence that costs fifty thousand pounds a year can easily need three hundred thousand in process change to actually work. If your budget only covers the licence, you have not budgeted for AI. You have budgeted for a subscription you will not use.

The cost of doing nothing

There is a cost to standing still as well. That is the trap. Boards look at AI costs and think the safe option is to wait. Sometimes it is. But sometimes the cost of waiting is higher than the cost of starting, and you only find out after a competitor has moved.

The Genpact and HFS research put a number on this.

That number is almost too large to be useful. What it means for your board is simpler. There is value locked inside your existing processes that you cannot reach without changing how things work. AI is one way to unlock it, but only if you know which processes are worth the effort and which are not.

How to set a budget that survives

The boards that get this right tend to do four things.

1. Start with the process, not the tool. What are you trying to change, and what is it worth if it works?
2. Budget for the seventy percent. If the technology costs X, budget at least three X for everything else.
3. Plan for the overrun. If 67% of programmes exceed budget, build in a contingency that reflects that.
4. Cap the pilot. Set a fixed price and a fixed time for the first phase. If it does not work, stop and reassess rather than throw good money after bad.

The pilot is where most money gets wasted. A team runs a small experiment, it shows promise, and everyone declares success while no one scales it. Months pass, costs mount, and the same pilot appears in next year's budget with new branding.

Fuzzelogic runs pilots differently. Twelve to sixteen weeks to a working MVP. That is what it takes to prove whether something works in your environment, with your data, under your constraints. If it does not work, you know quickly and cheaply. If it does, you have something real to take to the next stage.

The honest version

AI implementation costs vary because businesses vary. A simple automation in a mid-sized firm might cost tens of thousands. A complex transformation across a large enterprise can run into millions. The number itself is not the risk. The risk is not understanding what you are buying.

What I tell boards is this. If the budget only covers technology, it is wrong. If it does not account for process change, it is wrong. If it does not include a contingency for overruns, it is wrong. And if it does not have a clear off switch when the pilot does not work, it is not a budget. It is a blank cheque.

That sounds encouraging until you remember that 60% report minimal or no value overall. The 74% figure is about the most advanced initiative only. The rest, the bits that did not make it to "most advanced," are where the money went.

If the honest answer is that AI should not touch a process, we put it in writing rather than build it anyway. That honesty is worth more than any budget forecast.

Start with the assessment. Two to four weeks, fixed price, and you own the verdict and the roadmap whether or not we build any of it. When you are ready to talk AI, call Fuzzelogic Solutions and ask for Zak. www.FuzzelogicSolutions.com | info@FuzzelogicSolutions.com | +44 (0)1624 618950

For more on measuring return before you spend, read AI ROI: how to measure value before you spend next, then The 10-20-70 rule. The full library is on our index. Our site explains how Fuzzelogic approaches AI for business. You can reach Zak directly via our contact page.