Artificial intelligence as a business tool

The rapid advance of AI is bringing this technology into the heart of business operations which means it must be treated as a tool

It’s been three years since usable AI models became available and individuals and businesses started to experiment with them.

We were early into AI. We tried agentic AI (that ‘thinks’ for itself), and NLP LLMs that understand and generate language. We tried retrieval-augmented generation (RAG) to ground our models in internal documentation and knowledge, which could assist our researchers with trials and experiments.

Some of the first agents we used were Copilot in the Microsoft environment and also RAG agents that we subscribed to ourselves and kept in an isolated box, only using Harrison Manufacturing data and selected inputs.

We started with the back-office and logistics side of the business, a successful AI efficiency implementation considering we use 132 procured inputs to manufacture some of our grease products.

Now

Just three years into the AI journey and we’re learning as the field expands at great speed.

One of the big changes has been our acceptance of enterprise offerings for AI agents and models. We initially thought that our specific R&D work – especially through Harrison SPARC – had to be secured. However, the speed of improvement in AI, and the quality of the firewalls and cyber measures, means we have switched into the Microsoft enterprise system. The power, depth and quality of the AI agents is improving month by month and the place to be is in an enterprise subscription environment.

The switch to enterprise AI occurred for other reasons. Mainly that with the consumer rollout of AI agents – Open, GPT, Gemini, Grok, Claude – our staff were using them on their phones and laptops to solve fast queries and run quick scenarios. This meant we were creating the security leaks of our IP that we’d been trying to stop through a sealed-off AI RAG system.

Now, Harrison Manufacturing staff are required to only use Microsoft Enterprise Copilot which incorporates Claude and GPT. It’s safe, it’s fast – especially the web version – and it doesn’t disperse our IP into environments where the copyright and usage rules are hazy.

Operations

Production management is a very specific discipline and the main preoccupations – efficiency, quality and productivity – have already attracted specialised applications that include AI elements.

These AI elements are at the site of the production – cooking, pressurizing, cutting, extruding etc. – and the production management architecture includes embedded AI which is valuable in operations.

Productivity

However, efficiency, quality and productivity is the bread-and-butter of manufacturing. It’s what we do. And as I have mentioned in other columns, most Australian manufacturing companies were using AI (machine learning) in production operations before AI was being rolled out to consumers.

So, it might be that productivity is not going to be the main game for AI. I was interested in McKinsey’s April 2026 paper – Where AI will create value – and where it won’t. In this essay the consulting firm names three overlapping waves of change expected from AI: productivity gains, differentiation, and reduction of transaction costs. And they conclude:

“Productivity gains will matter, but they are unlikely to define the winners. Executives need to look beyond efficiency to product and service innovation, and ultimately to how AI will reshape industry structures and redistribute profits.”

That’s an interesting perspective and given the number of sectors that McKinsey works across, I suspect they are on to something. In blunt terms, productivity is what manufacturers do. And productivity/quality/efficiency gains are passed on to consumers rather than necessarily accumulating as profits inside the business.

But beyond productivity, the power of AI is quite obvious in product differentiation and in lowering transaction costs.

We are heavily reliant on AI for the R&D side of our business – which drives new and differentiated products – and we started our AI journey in the back office because getting transaction costs down was an achievable (and so-far successful) goal.

Unproductivity

Now that everyone is using AI in some form, it’s okay to be honest about its shortfalls. Perhaps the obvious one is honesty: why are people not admitting to their colleagues that a piece of written work is generated by AI? If we’re all okay with it, why aren’t these AI declarations being made? Do they think their colleagues expect them to be the author?

A bigger problem is the sheer amount of time people are spending endlessly returning to AI to get it to re-edit what has already been edited many times. People are using agents to produce long, detailed documents where, if it was coming from their own brain, they’d only write a page. The longer the AI-generated article, the more proof reading and editing that has to be done. And people go back to the prompts rather than doing the edits themselves. It’s a time-sink not a time-saver.

I’m also noticing that if you ask for the main points on a subject, AI will give you four but only the first two are worth anything.

Which means our own brains are still the executive – human judgement rules.

AI is a revelation and it’s a very powerful addition to a business. But it’s not a crutch, it’s a tool. And productivity may not be its real legacy.

Julie Harrison is CEO of Harrison Manufacturing Company