Every software vendor, every consultant, and every LinkedIn post right now is telling you that AI will transform your business. Some of them are right. A lot of them are selling something.

After twenty years working with data and operations across financial services and SMBs, here's the honest version: AI is a genuinely powerful tool for a specific set of problems. For everything else, it adds cost and complexity without proportionate return.

This is a practical framework for knowing which camp your situation falls into.

When AI is the right investment

AI tends to deliver strong ROI when most of the following are true:

  • You have a repetitive process that runs frequently — daily, weekly, or per transaction
  • The process involves large volumes of data that are too slow to review manually
  • Errors in that process have a real cost — financial, reputational, or in wasted time
  • Your data is reasonably consistent and already in a digital format
  • The decision or output has clear, definable criteria (even if complex)
A good rule of thumb: if you can describe exactly what a good outcome looks like, and a capable analyst could do it given enough time, AI can probably be trained to do it faster and at scale.

Classic examples in small businesses: invoice processing, anomaly detection in financial data, demand forecasting, automated report generation, customer query routing, and data quality checks across large datasets.

When AI is overkill

AI is not the right tool when:

  • The process happens rarely (monthly at best) and takes one person a few hours
  • The underlying data is inconsistent, incomplete, or lives in disconnected spreadsheets
  • The decision requires nuanced human judgement, relationships, or context the system can't see
  • A simpler automation — a script, a formula, a basic workflow tool — would solve 90% of the problem
  • You don't have the data volume for a model to learn meaningfully from

A common mistake is reaching for AI when the real problem is a missing process. If three people are manually reconciling spreadsheets because no one has agreed on a single source of truth, AI won't fix that. Fixing the process will — and it'll be faster and cheaper.

The data readiness question

This is the one businesses consistently underestimate. AI models are only as good as the data they're trained on and the data they operate on. Before any AI investment, it's worth asking:

  • Is our data consistently structured and labelled?
  • Do we have enough historical data for a model to find meaningful patterns?
  • Are there gaps, duplicates, or inconsistencies that would mislead a model?

If the answers are mostly no, the right first investment is in data infrastructure — not AI. We spend a significant portion of most client engagements here, and it's always time well spent. Clean data with basic automation often outperforms messy data with sophisticated AI.

A simple starting question

If you're unsure whether AI makes sense for a specific problem in your business, start here: what would this cost us if we did it manually, and how often does it happen?

Multiply the manual cost by frequency. If the number is meaningful relative to what an AI solution would cost to build and maintain, the conversation is worth having. If it's a few hours a month, fix it with a spreadsheet formula and move on.

AI is most powerful when it's solving a problem that already matters — not when it's creating a solution looking for one.