Cutting Through the Noise

The coverage of artificial intelligence in business has an unfortunate quality: it tends to cluster around extremes. On one end, you have enthusiastic predictions about AI replacing entire departments and transforming industries overnight. On the other, you have sceptical takes arguing that current AI tools are little more than expensive toys dressed up in credible-sounding marketing.

Neither extreme is especially useful if you're trying to make a sensible decision about whether and how to bring AI tools into your business. What's harder to find — and what this article attempts to offer — is a more honest look at the specific situations where AI tools have demonstrably improved how work gets done, and the situations where the hype hasn't matched the reality.

The perspective here comes from working with small and medium-sized Canadian businesses across a range of industries. The patterns that follow aren't theoretical — they're drawn from real engagements where AI tools either delivered genuine value or failed to justify the investment.

Where AI Tools Actually Deliver Value

1. High-Volume, Repetitive Text Tasks

One of the clearest and most consistent use cases for current AI tools is the handling of text tasks that are repetitive in nature but require enough variation that simple templates don't work. This includes things like drafting first versions of routine customer emails, summarising lengthy documents, extracting key data points from unstructured text, and generating first drafts of content that a human then reviews and edits.

The key phrase here is "first version" or "first draft." AI-generated text is rarely ready to go without review — it misses nuance, sometimes gets facts wrong, and tends toward a certain smoothness that doesn't always match a specific brand's voice. But if you have someone spending three hours a day writing variations of the same email, having a good first draft generated in seconds shifts their job meaningfully. The quality goes up, the tedium goes down, and you get more consistency.

A useful mental model: AI handles the structural draft, humans handle the judgement. The division of labour matters.

2. Data Extraction and Formatting

Many businesses collect data in messy, unstructured formats — PDFs, scanned documents, emails, mixed-format spreadsheets. Getting that information into a usable form has traditionally required either significant manual effort or expensive custom development.

AI tools have become genuinely capable at extracting structured data from unstructured sources. A tool that can take a pile of invoices and pull out vendor names, dates, amounts, and account codes — and do it accurately for 90–95% of the documents — changes the economics of a back-office process significantly. A person still needs to review and correct the exceptions, but the bulk of the work shifts.

This is one area where real-world results have often exceeded expectations. The catch is that accuracy varies considerably depending on document quality and consistency. Printed, high-quality PDFs typically yield much better results than scanned handwritten documents or heavily formatted spreadsheets.

3. Internal Search and Knowledge Retrieval

For businesses with large amounts of internal documentation — policies, procedures, product specs, historical contracts, training materials — one surprisingly effective AI application is building search tools that understand questions rather than just matching keywords.

The practical example: a new employee asks "what's our return process for wholesale orders?" Instead of navigating multiple document folders and hoping the right file turns up, a well-built internal AI search tool surfaces the relevant section of the operations manual. It's not magic — it only knows what's been fed into it — but for businesses with documentation spread across a dozen different places, the time savings are real.

This application tends to work well when the underlying documentation is actually maintained. If the internal knowledge base is out of date or disorganised, the AI search tool will dutifully surface outdated, disorganised answers.

4. Customer-Facing Response Tools

AI-powered chat and response tools have matured considerably. For businesses that receive a high volume of similar customer enquiries — frequently asked questions about hours, pricing, policies, or order status — a well-configured AI assistant can handle a significant portion of initial contacts, escalating to a human when the question is outside its scope or the customer requests it.

The important word is "well-configured." Out-of-the-box AI chat tools installed without careful setup tend to either frustrate customers with irrelevant answers or fail at the first genuinely specific question. The configuration work matters as much as the technology. When it's done properly, the outcome is generally faster initial responses, consistent answers, and staff freed up for more complex interactions.

5. Scheduling and Coordination Logic

For service businesses, field service organisations, and teams managing complex scheduling, AI-assisted scheduling tools can reduce the manual effort involved in matching availability, sequencing tasks, and managing last-minute changes. The value here is in handling the optimisation problem — figuring out the most efficient route, flagging conflicts, or redistributing work when something changes — faster than a person can do it manually.

This is a narrower use case, but for the businesses where scheduling complexity is a genuine operational burden, the impact can be substantial.

Where Current AI Tools Fall Short

Complex Reasoning and Judgment

Current AI tools are pattern-matching systems trained on large amounts of text and data. They are not reasoning systems in the sense that a skilled professional reasons. When a task requires genuine judgment — weighing ambiguous information, making a call based on incomplete context, or applying experience to a novel situation — AI tools tend to produce plausible-sounding outputs that may or may not reflect sound thinking.

This is why using AI to make consequential decisions without human oversight is risky. Draft a response: fine. Decide whether to extend credit to a customer: not a job for AI alone.

Niche Industry Knowledge

General-purpose AI tools are trained on broadly available data. For industries with specific terminology, regulations, or operational norms — construction contracts, agricultural supply chains, specialised manufacturing — the gap between what the AI "knows" and what a domain expert knows is often significant. This doesn't make AI useless in these contexts, but it does mean the output requires more careful review and the tools require more careful configuration.

Consistency Over Time

AI tools from major providers are updated regularly. A tool that behaves a certain way this month may behave differently after an update. For business processes where consistency is important, this introduces a management overhead that many organisations underestimate. You need a process for validating that updates haven't changed behaviour in ways that affect your workflow.

Accountability and Explainability

When an AI tool produces an output, it typically can't explain why it produced that output in a way that a person could audit. For decisions where accountability matters — anything customer-facing, anything compliance-related, anything that affects someone's livelihood — the inability to trace the reasoning is a real constraint.

Reasonable Starting Points for Most Businesses

If you're early in exploring AI tools and want to start somewhere sensible, a few areas tend to have reasonable return on time invested without requiring significant infrastructure:

  • Writing assistance for your team: Even basic use of AI writing tools for drafting emails, summarising documents, or generating first drafts of reports can recover meaningful time if it's adopted consistently.
  • Document Q&A for internal knowledge: If you have a specific document set that people refer to frequently, a simple AI search interface over that document set is often a useful entry point.
  • Structured data extraction from a specific document type: Pick one document type that comes in at volume and test whether AI extraction reduces manual processing time.

The common thread here is starting with a specific, bounded task rather than a broad aspiration. "We want to use AI" isn't a useful project brief. "We want to reduce the time our admin team spends extracting data from incoming purchase orders" is a project with a defined scope and measurable outcome.

The Realistic Summary

AI tools in their current state are genuinely useful for a specific category of business tasks: those that are high-volume, text-heavy, pattern-driven, and where accuracy of around 85–95% is acceptable with human review handling the remainder. That description covers more of most businesses' day-to-day operations than people often realise, which is why the category of legitimate use cases is real and worth paying attention to.

At the same time, treating AI as a catch-all solution — as something that will transform operations, replace strategic thinking, or make complex decisions — leads to disappointed expectations and wasted investment. The businesses that tend to get the most value from AI are the ones that approach it as a productivity tool in specific contexts rather than a general transformation programme.

If you'd like to think through where AI tools might fit your specific situation, we're happy to have that conversation. No obligation to proceed — just an honest discussion of where the opportunities and limitations lie for your context.