Engagement Leadership · Control room

Estimation & Scoping Studio

SIMULATEDVerified Jul 2, 2026

Estimate the same engagement three ways, watch them disagree, and price the unknowns as line items — because AI work blows up in data discovery and evaluation, not modeling.

Same instrument · three industries — pick a use-case to reconfigure the run

The three disagree by 4 weeks

That spread is the conversation. The analogous number is cheapest and most wrong; bottom-up misses the unknowns it hasn't imagined; PERT's range is the honest answer — 32–40 weeks.

Staffing & schedule · from PERT

Engagement lead×1

Senior engineer×2

ML engineer×1

Data engineer×1

QA / eval×1

Effort 36wDuration ~8wkTeam 6

Change control

Add 2 dispute categories + a new data source — +12w, concentrated in data + eval.

Baseline

38%

Absorbed

38%

With change order

38%

Steering-committee takeaway: AI estimates blow up in data discovery and evaluation, not modeling. Price the unknowns as line items or eat them later.

Resume echo — consulting delivery estimation across HCLTech/Genpact/Deloitte.

How this is built

Bottom-up = sum of a WBS with AI-specific line items flagged. Analogous = past baseline × complexity factor. PERT = (O + 4M + P)/6 with ±(P−O)/6 as the range.

Duration = effort ÷ effective team capacity (4.8 FTE-weeks/week); margin = (revenue − cost)/revenue at $12k bill / $7.5k cost per person-week. A scope change absorbed silently drops margin; a change order re-prices it.

Stack: Next.js (static) + shared design system; client-side only.

Limitations: WBS and rates are illustrative; capacity ignores ramp and dependency stalls. It frames the estimate and the change-control discipline, not a full resource-levelled plan.