ELU NOVA
← All field notes
GCC build-outs

Build-Operate-Transfer: what "transfer" actually means

5 min read

The phase everyone plans least

Build-Operate-Transfer is a sound structure for standing up an offshore capability centre. A partner establishes the entity, hires the team, runs it to steady state, and hands ownership to the client. The client gets the speed of a managed build with the ownership of a captive.

Three of the four phases receive detailed planning. Assessment is modelled. Build has a project plan. Operate has a reporting cadence and service levels.

Transfer, in a large share of agreements, is a paragraph.

That paragraph is where BOT engagements fail — not dramatically, but expensively. The transfer date arrives, the entity changes hands, and the client discovers that what transferred was a legal shell and a headcount rather than a functioning capability.

The four things that actually have to transfer

People. Not the headcount — the specific people, particularly the leadership. A centre's capability sits disproportionately with a handful of senior engineers and one or two managers who understand both the work and the client's business. If those people are employed by the partner and choose not to move, the transfer moved chairs. What to write down: which roles are designated as transferring, what retention arrangements apply through the transfer window, and what the partner's obligation is if a designated role departs within a defined period afterwards.

Process. The documented way the centre works: onboarding, code review, release, incident response, quality checks, escalation. In practice much of this lives in the heads of the people running it, and it is only when someone new has to follow it that the gaps appear. What to write down: that process documentation is a continuous deliverable during Operate, reviewed at defined intervals, not produced in the final quarter. Documentation written in month thirty is archaeology.

Platform. Repositories, pipelines, environments, tooling, licences, credentials. Two specific traps: tooling licensed to the partner rather than to the centre, and infrastructure inside the partner's cloud accounts. What to write down: an asset register maintained throughout Operate, with the licensing and account position for each entry stated, and a requirement that anything not transferable is flagged at the point it is introduced rather than discovered at the end.

Client relationships and context. The centre's knowledge of your business — why a rule exists, which stakeholder cares about which report, what went wrong in year one. This is the least documentable and the most valuable. What to write down: an overlap period during which the partner's principals remain reachable, with a defined scope and a defined end. Not an open-ended retainer, and not a hard stop on the transfer date either.

Failure modes

Leadership discontinuity. The centre head was a partner employee and does not transfer. Everything else can be documented; this cannot be replaced quickly.

Documentation written retrospectively. Produced in the last quarter to satisfy a clause. It describes an idealised process rather than the operating one, and the first incident after transfer proves it.

Platform lock-in by accident. Nobody intended it. A tool was convenient, a licence was in the partner's name, an environment was in the partner's account. Individually trivial; collectively a six-month untangling.

A transfer date nobody planned backwards from. Transfer is treated as an event rather than a phase. It should be a phase with its own plan, its own owner and its own milestones, beginning well before the date.

Compliance discontinuity. Data protection obligations do not pause during a transfer. In India, the DPDP Rules 2025 — notified 13 November 2025, with an eighteen-month compliance runway to 13 May 2027 — mean the entity's obligations as a data fiduciary continue uninterrupted through a change of ownership. Who holds them, on which date, should be written down rather than assumed.

Questions to settle at assessment

  1. Which named roles are designated as transferring, and what happens if one leaves within six months of transfer?
  2. Is process documentation a continuous deliverable with review points, or a final-phase task?
  3. Who maintains the asset register, and is transferability assessed when an asset is introduced?
  4. What is the overlap period after transfer, what does it cover, and when does it end?
  5. Which entity holds the data protection obligations on each side of the transfer date?
  6. What triggers an early transfer, and what triggers an extension? Both should be defined, and both should have a price.

The test

A transfer has succeeded when the centre operates for a quarter without the partner, and nothing degrades.

That is a demanding standard and it is the right one. A partner confident in their transfer plan should be willing to have it written into the agreement as the acceptance criterion, rather than treating the transfer date itself as completion.

Everything above is cheaper to agree at assessment, when nobody has anything at stake, than at month thirty, when both parties do.

--- ---

# Sources for this batch

  • Anthropic MCP announcement, 25 November 2024; OpenAI adoption March 2025; Microsoft Build announcement May 2025; Linux Foundation Agentic AI Foundation, 9 December 2025; specification revisions March, June and November 2025
  • MIT Project NANDA, The GenAI Divide: State of AI in Business 2025, July 2025 — including the 67% blended-team vs 22% IT-only success rates
  • Gartner press release, 25 June 2025 — agentic project cancellations, agent washing, ~130 genuine vendors
  • ISO/IEC 42001:2023, published December 2023
  • Digital Personal Data Protection Rules 2025, notified 13 November 2025, MeitY

# Publishing notes

  • A7 and D7 both contain candid statements about Elu Nova's own position. Keep them. A vendor-selection piece that exempts its author is worth nothing, and the self-assessment in D7 is the most trust-building paragraph in either batch. Founders should approve the wording before publication.
  • D7 is the strongest link-earning candidate in the program. Consider a downloadable one-page version of the twelve questions, gated only by nothing — no email capture. A free checklist that circulates internally at a prospect is worth more than a lead.
  • A8's arithmetic (10,000 documents, 90% STP, 1,000 exceptions, eight minutes each) should be rendered as a small inline calculation block, not buried in prose.
  • C1 should link to C2, C7 and E3 once those publish. Update the hub links as each goes live.

Find out where AI actually pays off in your business.

A focused discovery audit with a senior principal. You leave with a prioritized, costed roadmap — whether or not you work with us.

contact@elunovatech.com · Response within one business day