Workforce Capacity Planning for IT Services (200-500 Seats) — 2026 Framework

Why standard capacity models break at 200-500 seats

Standard capacity models assume a stable bench-to-bill mix, a clean utilisation number, and a roughly linear ramp curve. None of those assumptions hold for IT services shops in the 200-500 seat range. The bench fluctuates with the project pipeline. Utilisation diverges from realisation because not every billable hour is captured, billed, or collected at the contracted rate. Ramp curves bend sharply by skill, account type, and team manager. A plan that looks right on the spreadsheet misses by 15-25% on actual recoverable hours — and that gap is exactly where the margin lives.

Three structural realities pull the standard model apart at this size.

The utilisation-vs-realisation gap

Utilisation is the share of an employee's working hours assigned to billable work. Realisation is the share of those billable hours that actually convert to invoiced revenue at the contracted rate. The two are not the same. A shop running 75% utilisation against 85% realisation is converting only 64% of available hours into billed revenue. Standard models collapse the two into a single number, which hides the leakage. Once the gap is named, the conversation moves from "can we push utilisation higher?" to "where is the leakage and what is recoverable?" — which is the more useful question.

The bench mix

At 50 seats the bench is roughly homogenous and roughly knowable — five people you can name. At 250 seats the bench is a portfolio with skill heterogeneity, vintage skew, and account-history bias. Some of the bench is high-conversion (specific Java backend skill on the bench when the pipeline has Java backend asks). Some is low-conversion (legacy stack with no near-term pipeline). Treating the bench as a single number averages the high-conversion with the low-conversion and produces a forecast that overestimates near-term conversion and underestimates long-tail conversion.

The ramp curve

A new joiner ramping into a billable account at 200-500 seats does not ramp linearly. The first two weeks are induction. The next four to six weeks are shadowing and partial billing. Full billable rate only lands around the eight-to-ten week mark for most account types — longer for regulated work, longer again for offshore-to-onsite transitions. Standard models often use a single ramp factor; at 200-500 seats the variance across roles is large enough that the average is misleading on both ends.

The 4-variable capacity model

The 2026 model uses four inputs, each pulled from a system the shop already runs. The discipline is in keeping the inputs honest — particularly the leakage number, which most shops underreport because the timesheet system rounds it away.

Variable 1 — Billable hours captured

The honest measure of how much billable work actually makes it onto the invoice. The cross-check is the gap between timesheet submission (what employees write down) and productivity intelligence signal (what work systems show happened). The gap is informative — it points to timesheet hygiene issues, scope-under-billing, and rounding losses. The pattern at 200-500 seats is typically 5-12% under-capture against the productivity intelligence baseline. The discipline is closing that gap without turning the timesheet review into a surveillance exercise.

Variable 2 — Recovered leakage

Leakage is the difference between hours worked and hours billed. Recovered leakage is the share of that difference recoverable in the next quarter. Recovery actions include scope-creep re-billing, write-off discipline, rate-card refresh on rolling contracts, and account-level conversation on under-billed work. The number is small per account and large in aggregate — a typical 320-seat shop has recoverable leakage in the 4-8% revenue range, which is usually larger than the operating margin.

Variable 3 — Bench-to-bill conversion lead time

Not the average — the mean and the long tail. A 30-day mean bench conversion with a 90-day long tail tells you that capacity planning needs both a short-horizon forecast (the next-30-day fillable bench) and a long-horizon view (the legacy-stack bench that will not convert without re-skilling). Segment by skill (frontend, backend, data, ML, cloud, mobile) and by account type (offshore steady-state, offshore burst, onsite augmentation).

Variable 4 — Attrition replacement lead time

From notice through hiring through onboarding through ramp to full billable rate. The number runs longer than most CFOs assume — for mid-senior engineering roles in India IT services the typical end-to-end lead time is 90-120 days when measured honestly. The capacity model needs the attrition replacement pipeline as an explicit input, not a buffer line.

A quarterly capacity-planning template

Quarterly cadence with weekly variance checks. The quarterly cycle sets the model; the weekly variance catches drift early. Monthly is too coarse for ramp-curve and attrition surprises; weekly without a quarterly anchor produces noise the COO cannot act on.

The discipline that matters is the mid-Q checkpoint. Shops that defer re-baselining to the next quarter end up planning the next quarter on broken assumptions. The 5% trigger is approximate — for some shops 3% is the right line, for some 7% — but the principle of an explicit re-baseline trigger is universal.

DPDP and India compliance overlay

The capacity-planning data architecture processes personal data of employees at material granularity — timesheets, productivity signals, ramp curves, attrition pipeline. India's DPDP Act applies. The compliance overlay has three concrete obligations to design in from the start.

1. Section 4 consent or recognised legitimate use. The lawful basis for processing capacity-planning data. Employment context covers most of it under recognised legitimate use, but specific AI-driven evaluation may need explicit consent depending on how the implementing Rules land.
2. Section 8 DPIA when the data feeds AI inference. If capacity data feeds an AI system that makes significant evaluation, allocation, or retention decisions, a Data Protection Impact Assessment is the safe default. Examples: AI-driven bench-allocation recommendations, AI-driven attrition-risk scoring, AI-driven utilisation flagging.
3. Section 10 Significant Data Fiduciary obligations. If the shop is designated SDF (typically by data volume or sensitivity), periodic-audit and DPO obligations attach. Build the audit trail into the capacity-planning system from the start.

The implementing DPDP Rules are expected to be notified late 2025 or 2026, so build the architecture to flex with the final Rules — name the Data Fiduciary, retain DPIA documentation, keep the AI inference layer separable from the raw capacity data. The deeper DPDP scoring framework lives in the 14-question DPDP CISO worksheet. [needs-legal-review]

FAQ