Every business is being told they need AI. Most of the advice is terrible.
Here is an honest breakdown of what we have seen work — and what has wasted client budgets — across dozens of AI/ML implementations.
What Actually Delivers ROI
1. Document Processing & Extraction
If your team spends hours manually reading PDFs, invoices, contracts, or forms, this is the highest-ROI AI use case available today. Modern LLM-based extraction pipelines can replace most of that manual work with 95%+ accuracy, and they improve over time.
Real example: A logistics client reduced invoice processing time from 4 hours per day to 12 minutes. The tool cost less than a part-time employee.
2. Customer Support Triage
Not replacing support agents — routing and triaging. An AI layer that reads incoming support tickets and either resolves the simple ones automatically or routes complex ones to the right human dramatically reduces response times and agent frustration.
3. Personalised Content Recommendations
For e-commerce clients with catalogue sizes above a few hundred products, recommendation engines driven by collaborative filtering consistently lift average order value by 8–20%. The implementation is not trivial, but the payoff is measurable.
What Sounds Good But Rarely Works
Custom-Trained LLMs for Small Datasets
You do not need to train your own model. You need to prompt engineer an existing one. Fine-tuning makes sense at significant scale with significant data. For most businesses, it is expensive and rarely outperforms a well-prompted GPT-4o.
AI-Generated Marketing Copy at Volume
The content is detectable, the quality is inconsistent, and search engines are getting better at identifying it. Use AI to draft, human to finish. Never publish raw LLM output.
Chatbots Without Proper Fallback
A chatbot that cannot gracefully hand off to a human — or that confidently gives wrong answers — is worse than no chatbot. We have seen multiple clients lose customers because their AI support tool hallucinated refund policies.
The Right Starting Point
Before any AI investment, ask: what decision or task happens frequently, is costly in human time, and has a reasonably defined correct answer?
That intersection is where AI pays off. Everything else is a science project.
We are happy to do a free AI opportunity audit for your business. Book a call if you want an honest assessment of where AI can actually move the needle.