The Three Killers of AI Initiatives
Australian mid-market companies are spending millions on AI. Most are getting nothing back. Not because the technology doesn't work—because they're approaching it backwards.
After working with dozens of firms across manufacturing, logistics, professional services, and healthcare, we've identified three patterns that consistently predict AI failure. Fix these, and your odds of meaningful ROI jump dramatically.
1. Wrong Sequencing
The most common mistake is jumping straight to "sexy" use cases—generative AI for customer service, predictive analytics for sales forecasting, LLM-powered document processing—without building the data infrastructure that makes any of it possible.
Here's what actually happens: a CEO reads about ChatGPT and demands a "customer-facing AI assistant." The IT team spends six months trying to connect it to fragmented CRM data, inconsistent naming conventions, and siloed customer records. The project either dies in pilot purgatory or launches with such degraded performance that internal stakeholders refuse to use it.
Good AI strategy starts with ugly problems. Clean data. Process standardisation. Integration layers. These aren't glamorous, but they're where real value compounds.
2. No ROI Logic
Most businesses approach AI the way they approached digital transformation in 2015—spending first, figuring out value later. This is financial recklessness dressed up as innovation.
Before any AI investment, you need three numbers: current cost of the problem, expected cost reduction, and implementation cost including change management. If you can't articulate those numbers in a business case, you shouldn't be spending a dollar.
We see companies routinely spend $200K+ on AI proof-of-concepts with no clear success criteria. Then they wonder why the board treats AI as a budget line to cut when times get tough.
3. Vendor-Led Decisions
AI vendors are optimised for one thing: selling you their product. They're not paid to understand your operational context or tell you when AI isn't the answer.
The typical vendor pitch goes like this: "Industry X is using our platform to achieve Y results." What they don't tell you: the three similar companies that tried and failed, the six months of data preparation their case study glosses over, or the fact that Y only matters if you have the same data quality they assume you have.
Good AI strategy is vendor-agnostic. It starts with your problems, not their solutions.
What Good AI Strategy Actually Looks Like
The firms winning with AI share a common approach:
- Problem-first thinking: They identify specific, measurable operational inefficiencies before considering any technology.
- Sequenced delivery: They build data foundations first, then layer on intelligence incrementally.
- Hard ROI targets: Every project has a clear business case with defined success metrics before a single vendor is engaged.
- Internal capability building: They're not outsourcing intelligence—they're building it internally to compound over time.
The gap between firms doing AI well and firms doing AI badly isn't technology. It's discipline. Most companies treat AI as a magical solution to skip the hard work of operational excellence. It doesn't work that way. AI is a multiplier—it makes good operations better and bad operations more expensive.
Get the fundamentals right first. Then, and only then, start spending.