Schedule risk management for deep-tech & hardware startups

Hardware schedules carry more compounding uncertainty than software, and the deterministic Gantt hides all of it. A long-lead casting, a custom ASIC, or a vendor-built cryostat can land six weeks late, and there is no slack-free way to recover. A qualification campaign that should take three runs takes five. A single test chamber or one cryo engineer becomes the real constraint even though the dependency network suggests otherwise. None of this is visible in a plan that shows one bar and one date.

The cost of that false precision is paid at the board level. When the only number on the slide is a deterministic milestone date, a single slip cascades silently through the rest of the plan, and the team discovers it months later. Founders end up re-baselining by hand every time a vendor moves, and investors learn — correctly — to distrust the dates.

Technology-readiness-level progression is the spine of a deep-tech program, but in most tools a gate is just a milestone diamond with no logic behind it. CritPath AI models decision gates — Go, No-Go, Pivot, or Defer — as real objects in the schedule. A TRL gate is where you decide whether a subsystem has earned the right to advance, and the consequence of that decision is built into the plan rather than tracked in a separate deck.

The differentiator is retroactive rescheduling. When you record a gate outcome — say a TRL-5 review slips, or a subsystem pivots to a backup approach — CritPath re-cascades the entire downstream schedule automatically, recomputes the critical path and chain, and updates your probabilistic finish date. You stop re-planning by hand after every gate and start seeing the real consequence of each decision the moment you make it.

Investor and grant milestones deserve probabilistic backing, not a hopeful single date. CritPath AI runs a Monte Carlo simulation over your schedule — PERT-Beta task distributions, explicit risk-event injection, optional duration correlation — and returns P50, P80, and P90 finish dates plus a criticality index and a tornado chart that ranks which tasks actually drive the spread.

That changes the conversation with a board. Instead of committing to a date you privately know is a coin flip, you commit to a P80 milestone you can defend, show the few drivers behind it, and explain exactly what would have to go right to hit the aggressive P50. The tornado chart tells you where to spend de-risking dollars — a second vendor, an extra test article, a parallel path — because it shows which uncertainty moves the milestone most.

Padding every task to feel safe makes a hardware plan longer without making the date safer — the safety gets wasted to student syndrome and Parkinson's Law. Critical Chain Project Management (CCPM), from the Theory of Constraints, does the opposite: it strips estimates to an aggressive median and pools the protection into explicit buffers where it matters most.

For a hardware program this maps cleanly onto reality. A project buffer protects the committed delivery date at the end of the critical chain. Feeding buffers protect that chain wherever a long-lead procurement or a parallel subsystem merges in, so a late casting or a slipped board spin is absorbed instead of cascading. CritPath sizes those buffers from your Monte Carlo distribution — so the buffer reflects your actual P50-to-P80 spread — and tracks consumption on a fever chart, giving you a single green/yellow/red signal of whether the program is on track without staring at a wall of task dates.

The CritPath AI copilot, built on Claude and Gemini, is grounded in your actual schedule — the dependency graph, the buffers, the Monte Carlo output — not a generic chatbot pinned on top. Ask it which task is driving your P80 slip, what happens to the integration date if the magnet vendor moves three weeks, or which gate decision is consuming the most buffer, and it answers by reasoning over the real engine.

Founders and CTOs can also configure AI employees with custom skill sets and use AI assistance to decompose a program into a work breakdown structure, draft reports, and build skills. During the beta, autonomous AI-agent runs are off; AI assistance — the copilot, the AI work-breakdown decompose, reports, and skill wizards — is on, so the help is real but always under your review.

Legacy schedule-risk tools — Primavera Risk Analysis, Deltek Acumen, Safran, Full Monte — have the math but cost tens of thousands of dollars, run on Windows desktops, and assume a dedicated analyst and a construction-shaped world. Modern PM tools like Monday, Asana, and Linear have the UX but zero probabilistic schedule-risk math. Neither was built for a deep-tech founder who needs both.

CritPath AI is $10 per user per month for every standard feature — unlimited projects, CPM, Monte Carlo, CCPM/TOC, decision gates, WSJF and Cost of Delay, EVM, and AACE 132R-23 Level 4 with an append-only audit log — with AI copilot usage billed separately by metered usage. SOC 2 Type II, SAML SSO, and on-prem are on the Enterprise roadmap, not yet shipped. The result is owner-grade schedule risk that a founder can stand up in an afternoon and a board can actually trust.