Is it AI, or is it automation wearing a better costume?

Written by May Wong - Practice Manager

Every conversation I have with clients lately ends up in the same place. Everywhere I turn, LinkedIn, business forums, industry events, it's AI, AI, AI. Yes, the algorithm's probably doing some of that work for me. Everyone wants to be "on the journey", chasing the same promise: innovate with it, cut costs, free up humans for the work that supposedly matters more.

Funny thing is, we say we are tired of hearing about it and in that same breath we're reading every story that comes through. We are hooked either way. We know most of it is hype, but we keep reading anyway, hunting for that nugget we can actually use.

Here is another great story I heard this week. A small New Zealand company is making a name for itself using AI to scan fruit for quality and cut down on waste before it ever leaves the packhouse. That's the real thing, a judgement call, done at a speed and scale no person could match. I mention it because it's worth being clear. I believe AI can help and has genuine use cases. But plenty of what gets called AI these days isn't that, and what it’s built on often cannot support what you’re asking it to do.

Underneath the hype, there's real pressure on organisations, from wanting to use AI and having to be seen implementing it. Boards and C-suites want it on the table, whether as an innovation story to sell or an efficiency gain. Competitors are making noise about what they've rolled out. Nobody wants to be the organisation still thinking about AI while everyone else appears to be doing it.

Not everything needs to be AI

AI isn't free and it doesn't get cheaper the more you use it. Every call costs tokens and tokens add up fast once a "smart" feature is running on every record in your database rather than the handful that actually needed judgement.

I'd manage AI the way we eventually learned to manage cloud spend. Nobody blinked at cloud costs early on either, until the invoice showed up and someone asked why a dev environment had been running at full size since March. AI spend needs the same discipline. Know what you're using it for, check whether it's earning its cost. Some of what we have seen being ‘AI’ed’ could be done just as well by a decision/rules engine paired with automation.

The test is simple. If the task has one right answer and no real ambiguity, you probably don't need AI for it. You need a rule, and where you do need something smarter than a fixed rule, it's worth asking whether a narrower, purpose-built model would do the job for a fraction of what a general AI tool costs to run.

Not all AI needs the same foundation

One useful business lens for thinking about AI is in terms of four buckets: Retrieval, Drafting, Recommending and Acting. All of them are underpinned by data. The data

quality bar rises with each move from Retrieval to Acting. Retrieval can survive a few duplicated records. Acting will be less forgiving and there will be no one to catch it until it’s too late.

Here’s the problem, most organisations want the Recommending and Acting because that’s where the efficiency gains actually show up. But this is exactly where the data, content and information the AI depends on needs to be reliable and consistent. Most organisations don’t have solid data. There’s plenty of manual patching: “We know that field’s unreliable. Just ignore it.” That’s fine when a person’s making a call (well not really but that’s a discussion for another day). It’s a real problem when AI has to act on rubbish data.

So the question is this: If organisations are aiming for the two buckets that need the most from their data and if most don’t have data that’s ready for that, what outcome should we expect?

Before your next AI investment

Before pushing for AI, ask yourself two things: 1. Does this need judgement, or would an automation and decision engine do the same job? 2. If it does, which bucket does it sit in and does your data hold up?

Going back to where I started, organisations are under real pressure to move fast on AI, from competitors, from fear of being left behind or just cost efficiency. There’s a lot sitting between that pressure and a good outcome. We need to ask questions like whether it should be AI, which bucket it falls into and if the foundations can support it. We haven’t even touched on security yet which is its own conversation entirely.

If this sounds familiar, it’s what we will be digging into at The next Knowledge Space event - AI and architecture: the pressure, the promise, and the people caught in between on Wednesday 21st October. Register here

And look, ironically, I used AI to help write this piece, poke holes in my logic, argue back at me, and organise my rambling into something with actual structure. Something that would've taken me days of muttering to myself now took an afternoon. Make of that what you will.