AI that explains your numbers — not invents them

Ask a general-purpose AI “what’s my project’s CPI?” and it will happily give you one. Confident, plausible, and completely made up. That’s the trap with AI in project delivery: the output sounds like analysis, so people trust it — right up until a fabricated number lands in a board deck.

There’s a better pattern, and it’s not subtle: compute, then explain. The software does the math deterministically. The AI only turns those real figures into readable prose. Two jobs, cleanly separated — and it changes everything about whether you can trust the result.

Why the split matters

• Determinism. The same inputs always produce the same CPI, the same critical path, the same capacity picture. Ask twice, get the same answer — because a formula computed it, not a probability distribution.
• No invented facts. When the AI is handed the numbers and told “explain these,” it can’t conjure a risk that doesn’t exist or a date nobody set. It works from what’s genuinely on the record.
• Defensibility. You can trace every figure back to its source. “Where did this come from?” has an answer that isn’t “the model felt strongly about it.”

Where AI genuinely earns its keep

Used this way, AI is a fantastic translator, not an oracle. It turns a computed critical path into “here’s where your schedule risk is, and two moves to protect the end date.” It drafts a status summary from the real KPIs so a PM isn’t staring at a blank box on a Friday. It reads a capacity picture and says, in plain English, who’s overloaded and where the headroom is. The facts are the app’s; the fluency is the AI’s.

Where it shouldn’t go

Notice what’s not on that list: deciding whether to approve a change, manufacturing a decision that nobody made, or inventing progress to fill a gap. If a field records something that actually happened — a decision, an approval, an outcome — a human enters it. AI drafts the words around the facts; it doesn’t get to make the facts up.

The test for any AI feature in your toolchain is one question: can it invent a number? If yes, be very careful what you do with the output. If no — if it can only explain figures the system already computed — then let it write all day. That’s AI you can put in front of a sponsor.