Every ML project I've estimated has taken longer than I predicted. Every single one.

After 250+ projects, here's why and what I do about it:

ML projects have fundamental uncertainty that software projects don't. You don't know if the data is good enough until you try. You don't know if the approach will work until you test it. You don't know how much iteration the model will need.

My estimation framework: take your initial estimate and multiply by 2.5. That sounds ridiculous but it's empirically accurate. Here's why the multiplier works:

Data collection and cleaning: always 2-3x what you expect. Model iteration: "the first approach didn't work, let's try another" happens in most projects. Edge case handling: the last 10% of quality takes 50% of the time. Deployment and integration: always has surprises.

I now communicate ML timelines in ranges, not points. "This will take 4-10 weeks" with clear milestones: "Week 2 we'll know if the data is sufficient. Week 4 we'll have a baseline model. Week 6-10 is iteration and deployment."

Clients and managers appreciate the honesty. And when you deliver at week 7, you're early instead of late.

Under-promise. Over-deliver. Always.