AI Beyond the Sales Pitch: A Project Manager's Experience

Curtis, a senior project manager working across several sectors, opens by separating narrow AI, which performs one defined task, from the broader and agentic systems often promised in sales pitches. She contrasts automation, which repeats fixed steps, with AI that recognises patterns, and explains that pattern matching is not the same as reasoning. Her central framework is a three-part check: the data source, the desired output, and explainability, meaning the ability to trace how an answer was produced.

Through case studies and audience exercises, she applies that framework to real projects. An insurance first-notice-of-loss pilot showed why emotionally sensitive calls should stay with people. A lottery rules assistant succeeded after a phased rollout, first to internal testers and then to call-centre staff. A celebrity chat project stalled on the cost of a premium model used for long conversations. A university enrollment tool surfaced translation and access-control problems, while an HR assistant worked well once it lived inside Slack, where staff already were.

She closes with practical advice. When an answer looks wrong, trace it back to its source document, since outdated content is a common cause. Do not default to the most expensive model; switching to a cheaper one after tuning saved budget in one inventory project. Define an acceptable answer rather than a perfect one, bring in content, developer and in-between experts, and expect testing to stall when knowledge stays within the original team.