Phase 8 of 10 · Build

Technology and infrastructure

What does the centre need to run on day one, and to grow afterwards?

The big picture

Technology failures in a GCC are usually decision failures: the wrong tools too early, the right ones too late, or a good stack never integrated into how work gets done. The choices made at launch set the centre’s ceiling for years.

The decisions that matter

  • The connectivity model: full remote access, a replicated instance, or hybrid
  • Which data can be processed where (data classification and residency)
  • How automation is governed
  • The data layer under reporting
  • Facilities: flexible space or a direct lease

What good looks like at the end

  • A data classification and residency map signed off by Legal, security and Finance
  • A connectivity decision made with the CIO and CISO
  • Five-layer readiness confirmed as a launch gate
  • A bot registry, a test environment and an upgrade protocol

Who owns it

The CIO and CISO with the GCC Head

When

In parallel with process migration planning, never after it. Technology readiness is a go-live gate.

Five-layer infrastructure readiness

LayerWhat ready looks likeCommon gap
1. Core process systemsERP access, user IDs and licences in place; user acceptance testing completeAccess without testing; process gaps found on day one
2. Connectivity and integrationModel decided; network tested at full load; integration points mappedTested at low volume; breaks under live load
3. Productivity and collaborationSame tools as the parent; knowledge management liveA different platform from the parent; two-speed communication
4. Analytics and reportingData layer designed; definitions standardised; BI on a clean sourceReports built on ERP extracts; reconciliation problems at first month-end
5. Process intelligenceProcess mining connected; event data flowing; baseline setDeferred; no performance baseline for year one

Use this as a pre-launch gate, not a post-launch diagnostic.

What specialist advisers say

Curated from published work by Everest Group. Summarised in our own words; follow the link for the full piece.

In detail

The approach

  1. Map data before provisioning anything

    Classify data and map residency with security, data protection and Legal: what can be processed in the location, what needs controls (encryption, access logging, masking), and what cannot move at all.

  2. Decide connectivity explicitly

    Remote access is fastest; a replicated instance gives independence but adds synchronisation risk; hybrid suits centres doing both processing and analytics. Changing later is a mini-implementation.

  3. Face the ERP reality

    With a fragmented ERP landscape, either rationalise first or scope the centre around units that are already standardised and expand as harmonisation happens.

  4. Govern automation from the first bot

    One platform at first, a live bot registry, a test environment that mirrors production, and a protocol that routes system upgrades through the automation team.

  5. Build the data layer before reports

    One source of truth with standard definitions, so month-end is spent interpreting numbers, not reconciling them.

  6. Switch on process intelligence at launch

    Connect process mining to the ERP from day one so there is a baseline to measure every later improvement against.

  7. Plan facilities for the operating model

    Design the workspace for the centre you intend to be. Flexible space suits early scale; a direct lease usually wins somewhere around 150 to 200 people.

Readiness signals

The signs that need attention before a centre moves on from this phase.

  • The data residency map is signed off and feeding the design
  • The connectivity model was decided with the CIO and CISO
  • All five layers meet their “ready” definition
  • A bot registry, test environment and upgrade protocol exist
  • Reporting runs from a standardised data layer

Where organisations go wrong

  • Infrastructure planned after process migration

    What works: Run them in parallel, with technology readiness as a gate.

  • A decentralised “bot farm” with no registry

    What works: Govern automation as a system from day one.

  • Process mining switched on in month twelve

    What works: Connect it at launch; the first year’s data is otherwise lost.

Questions for leadership teams

  • Was our connectivity model chosen explicitly, or defaulted into by whoever set up access first?
  • Is our reporting sourced from a clean data layer, or from extracts reconciled by hand?
  • What happens to our automations at the next ERP upgrade?

Read more