Designing global talent systems

I spent years designing and redesigning teams that sat in India, Indonesia, the Philippines, Nigeria, and earlier in the US and Europe. The documented pattern is consistent: joined pre-revenue as the third person leading operations and data leadership across multiple countries, built a 15+ person global data organization, stood up a 70+ person call center from scratch across India and the Philippines, and later led cross-border restructuring that produced 300–400% cost savings. The stated goal was almost always the same — lower cost per unit of output while keeping quality and speed acceptable. The real work was never the unit cost.

Pure cost arbitrage fails in predictable ways. The failure modes are not mysterious, but they are easy to ignore when the spreadsheet looks good.

The arrangement that looks obvious

You locate work where the fully loaded cost of a capable person is a fraction of the cost in the primary market. You build a capability matrix so that you know, in theory, who can do what. You create reporting lines, documentation standards, and some form of review cadence. On paper the savings are large — three to four times is common once you include benefits, office, management overhead, and attrition.

The matrix is useful. It forces a conversation about actual skills instead of titles. It makes gaps visible. It becomes the basis for hiring, training, and succession. But a matrix is a static map. The system that actually runs is the daily flow of decisions, context, and correction across time zones and cultural defaults.

A capability matrix in practice

Here is the shape of the tool that actually got used (simplified and anonymized). These matrices later appeared again during knowledge-transfer and succession work:

Skill / Capability                  | Person A | Person B | Person C | Person D | Depth
------------------------------------|----------|----------|----------|----------|--------
Owns end-to-end process design      |    3     |    1     |    0     |    2     |  Thin
Can train others on the process     |    3     |    2     |    1     |    2     |  
Can execute without supervision     |    3     |    3     |    2     |    3     |  
Can diagnose production failures    |    3     |    2     |    1     |    2     |  
Can write usable documentation      |    2     |    3     |    1     |    2     |  
Can hold full business context      |    3     |    1     |    0     |    1     |  Critical gap

Scoring was usually simple:
0 = cannot do
1 = can do with heavy guidance
2 = can do independently
3 = can teach and improve it

The matrix was never a hiring form. It was a living record of systemic risk. The most useful version was updated after projects, not once a year, and the conversation always started with the “Depth” column: where is the system one resignation away from losing a capability?

Where the cost actually lives

Three taxes appear almost immediately.

1. Coordination tax

Every decision that used to require one conversation now requires three. Context has to be written down or it evaporates. The person who holds the full picture of a process is rarely the person executing the next step.

Pattern observed across multiple builds:
A change that previously took a 15-minute conversation in one location started taking 1.5–2 days of back-and-forth once the work was split. The delay was not malice or incompetence. It was the cost of re-establishing shared context every time. Decision latency became a first-class metric because it was the tax that showed up first in the numbers.

2. Leadership bandwidth tax

Someone still has to own the standard. When the standard is enforced from a different geography, two things happen: either the standard softens to match local norms, or the person enforcing it becomes a bottleneck.

Pattern observed:
In multi-year setups, a single senior person frequently ended up spending more than half their time translating intent and reviewing work product rather than improving the system. Throughput looked fine on paper; the actual constraint was that person’s calendar. The original cost model had never priced the leadership layer as a scarce resource.

3. Process redesign tax

You cannot simply relocate the same process. The version that worked when everyone shared the same cultural assumptions and ambient knowledge does not survive the move.

Pattern observed:
Teams that tried to move an existing process “as-is” usually saw quality drift within 60–90 days. The ones that succeeded treated the move as a redesign: shorter documentation, explicit escalation paths, and a deliberate period where the original location still owned the standard while the new location built fluency. The first six to nine months were almost always more expensive than the spreadsheet predicted.

What the large restructurings actually required

The restructurings that produced the documented 300–400% savings required more than lower wages. They required:

• Explicit process ownership that could survive the departure of any single person
• Written standards short enough to be used and specific enough to be enforceable
• A small number of people who could hold the full context and were willing to spend a large fraction of their time transferring it
• Acceptance that the first six to nine months would be more expensive, not less, while the new system stabilized
• Willingness to reassign or exit people who could not operate inside the new design, even when their individual cost looked attractive

Without those pieces the savings were temporary. The system slowly re-created the original cost structure through rework, escalation, and the quiet return of work to higher-cost locations.

Business KPIs that actually moved

These were the measures that proved useful when talking to operators and boards (not the usual HR dashboard):

The wage differential always looked attractive. These KPIs told you whether the system was actually becoming cheaper or merely relocating the cost.

Implementation timeline

A realistic sequence, distilled from the multi-year pattern of standing up and later restructuring these systems:

Months 0–2 — Design & baseline
Map the current process and decision rights. Build the first version of the capability matrix for the critical skills. Identify the true single points of failure. Set the target cost model including coordination and leadership overhead, not just wage rates. Decide what “done” looks like for capability transfer.

Months 2–5 — Seed & dual-run
Hire or identify the first cohort in the new location. Run dual ownership: the original location still owns the standard while the new location executes under close review. Update the capability matrix every two to four weeks. Measure decision latency and rework from day one. Expect higher short-term cost.

Months 5–9 — Stabilize & transfer
Shift primary ownership of individual processes only when depth ≥ 2 on the matrix. Document the short, usable standards that actually get used. Force the leadership bandwidth question into the open: who is spending their time translating versus improving? Begin measuring % of work completed without escalation.

Months 9–15 — Optimize & harden
Remove remaining single points of failure. Tighten the cost-per-unit metric so it reflects fully loaded reality. Use the matrix as a living risk register rather than an annual HR exercise. At this stage the 300–400% savings become durable only if the process and leadership layers have been redesigned, not just relocated.

Ongoing
Treat the matrix as operational infrastructure. Update it after major projects or attrition events. Revisit the original cost model every six months; pure wage arbitrage will slowly erode if coordination and leadership costs are ignored.

The pattern that held up

The durable pattern was never “move the work to the cheapest capable location.” It was “design the system so that capability can be distributed without destroying the decision quality that made the original team valuable.”

That requires treating talent systems the same way you treat any other production system: with explicit interfaces, observable failure modes, and a cost model that includes the coordination and leadership layers rather than pretending they are free.

The spreadsheet will always show the wage differential. The real design work is pricing everything the spreadsheet leaves out.