The AI is new. The foundations aren’t.

The AI part of an AI project is usually the smallest part.

That surprises people. The model gets all the attention, but in most of the work we do, the call to the model is a few lines. Everything around it is the real job. Where the data lives. How it connects to Xero or the CRM. Who is allowed to see what. What happens when something breaks at four on a Friday. What it costs when you’re three times busier than you are today.

None of that is new. It’s the work IT people have been doing for decades, and it’s where Logic8 comes from. Sam and Mithun founded Logic8 after more than twenty years each in IT, building and running technology for other people’s businesses. AI is the newest tool in the kit. It isn’t the kit.

Why AI projects stumble, according to the research

The RAND Corporation looked into this and found that more than 80 per cent of AI projects fail, which is twice the rate of IT projects that don’t involve AI. Then they asked experienced engineers why. The answers were:

• The problem to be solved was misunderstood, or poorly explained
• The data needed to do the job wasn’t there
• The project chased the newest technology instead of the real problem
• The infrastructure to manage the data and run the finished thing was missing
• The problem was simply too hard for AI

Gartner came at it from another angle and predicted that, through 2026, organisations would abandon 60 per cent of AI projects that weren’t supported by data that was ready for them.

Read that list again and notice what isn’t on it. Nobody said they picked the wrong model. It’s a list of foundations.

We’ve watched a few waves come through

More than twenty years in IT means you see a few waves arrive. Businesses moving onto the web. Servers leaving the cupboard for the cloud. Smartphones. Software you rent by the month instead of buying in a box.

Each wave came with the same excitement, and each time the businesses that did well out of it were the ones that got the plain things right. Clean data. Systems that talked to each other. Backups that had actually been tested. A bill somebody understood.

AI is a bigger wave than most, and we’re as excited about it as anyone. But the pattern hasn’t changed, and knowing the pattern is useful. It’s why we’ll ask about your data before we talk about models, and why we’ll ask about your week before we talk about either.

What the foundations look like in practice

Here is what sits underneath anything we build, and where each piece shows up in our work.

• The right job. Before anything gets built, we work out which jobs are worth changing and which are fine as they are. That’s what AI consulting is for.
• Data you can trust. An assistant is only as good as what it reads. If the numbers live in Xero, a job system and a spreadsheet someone updates on Fridays, we bring them together first.
• Systems that talk to each other. Most of the saving sits between your systems, where a person is copying, pasting and checking. Closing those gaps is the heart of AI integration and workflow automation.
• Security and privacy. Who can see what, where customer information is stored, what gets sent to a model and what never should. Decided at the start, not patched on afterwards.
• Something that stays up. Hosting, backups, monitoring, and a plan for the day a supplier changes their prices or retires a model. When nothing on the shelf fits the way you work, we build it on the same footing.

Will it still add up when you get busier?

A tool that looks cheap in a trial can look quite different a year on. More jobs going through it. More files to store. More people on the team using it.

So we do the sums early. What it costs at 100 runs a day, and what it costs at 10,000. A smaller model where it does the job well, a bigger one where it’s worth paying for. Sensible storage like R2 or S3 once the files pile up. A light database when a heavy one would be overkill.

A little planning here means growth is good news for the bill, not a nasty surprise.

Your team’s time belongs in the numbers

The AI bill is only part of the cost.

Say you’re looking at a tool that drafts customer quotes. The monthly price looks reasonable. What you really want to know is how much work is left once the draft arrives. Someone still confirms the details, checks the price and makes sure it’s ready to send. That checking is part of doing a good job, and it belongs in the numbers.

We work it out with a small sample of the work your team really does, timed from start to finish. Sometimes paying a little more for a model means less rework. Sometimes a simpler, cheaper option handles it just as well. It depends on your work, which is why we test on your work.

Five questions worth asking before you build anything

These are useful whoever you end up working with, including us.

1. Where will our data be stored, and who can see it?
2. What will this cost each month at the volume we have now, and at five times that?
3. What happens when it gets an answer wrong, and who finds out?
4. How does it connect to the systems we already use?
5. Who looks after it in a year, when the model it runs on has been replaced?

Good answers are plain and specific. You shouldn’t need a glossary to follow them.

The short version

AI is moving quickly, and it’s fair to find that a lot. The parts that decide whether it pays off are the steady ones: the right job, good data, connected systems, sound security and honest sums. We’ve been doing those for more than twenty years. AI is the newest and most exciting thing we get to build on top of them.

If you’re weighing up an AI tool, or there’s a job you’d like some help with, bring us that one job. We can work through the numbers together, no pressure.