Analysis Artifact design intermediate

Total Cost of Ownership Model

The full multi-year cost of an option including the people, the exit and the things that never appear on an invoice.

The comparison finance will actually make. Its credibility rests on the rows nobody wants to estimate — engineering time to operate, migration effort and exit cost — because leaving them out is what makes the cheap option look cheap.

The shape

Three-year comparison, ₹ lakh.

Cost line A: Managed SaaS B: Self-hosted OSS C: Build
Licence / subscription 420 0 0
Compute + storage 60 310 340
Network + egress 15 45 50
Implementation (one-off) 90 160 520
Migration off current (one-off) 70 70 70
Engineering run — 0.3 / 1.8 / 2.4 FTE 108 648 864
Support contract 0 90 0
Training + ramp 20 45 30
Compliance + audit effort 15 40 55
Exit / portability provision 40 15 10
Three-year total 838 1 423 1 939
Time to first value 6 weeks 4 months 11 months
Unit cost at 3× volume 1 980 1 610 1 520

Sensitivities. Option A's licence is per-seat: at 3× users its total exceeds B. Option B assumes 1.8 FTE, which is one person plus cover — if that person leaves, B's real cost is closer to C's. All options assume the current volume forecast; the crossover for A and B sits at roughly 2.2× today's volume.

When you produce it

For any build-versus-buy decision, any vendor selection above a material threshold, and any migration business case. Produce it before the technical recommendation, so the recommendation is informed by it rather than defending it.

Who reads it

Finance and procurement, who will test the assumptions. Executives choosing. Architects, who need to understand which technical decisions actually move the number — usually the operating headcount, not the infrastructure.

What good looks like

  • Engineering time is a line item, costed at a real rate. It is normally the largest number and it appears on no invoice.
  • Exit cost is included. It is the price of the option to change your mind.
  • Sensitivities are stated, with the crossover point named — that is what makes the model useful when reality differs from the forecast.
  • Unit economics at a different scale, not just today's total.
  • Time to value is on the same page; a cheaper option arriving eleven months later is not obviously cheaper.
  • One-off and recurring costs are separated.

Common mistakes

  • Infrastructure only, comparing cloud bills and calling it TCO.
  • Assuming the self-hosted option needs no one. It needs at least two people, because one person is not a rota.
  • Ignoring exit, then discovering the proprietary format three years later.
  • False precision. These are estimates; show ranges and say which assumptions the conclusion is sensitive to.