GA AI Case Study – Modernising business-process software for human-and-AI work
At a glance
Global Advisors has progressively modernised a portfolio of internal business-process applications spanning employee administration, expenses, recruiting, performance, planning, payroll-related work, finance review and action approval. The programme replaced page-centric legacy logic with reusable presentation patterns, clearer service boundaries, explicit workflow states and interfaces that both people and agents can use.
The objective was not to put a chatbot over old screens. It was to make process state, decision rights, evidence and actions explicit enough for AI to assist safely.
Maturity: active partial migration; several modules are operational while legacy data paths and coexistence controls remain in use.
GA AI Case Study – Modernising business-process software for human-and-AI work
The situation
The legacy intranet delivered important business value but concentrated presentation, database access, business rules and workflow in individual page scripts. This made changes risky, user experience inconsistent and integration difficult. It also left AI with no safe unit of work: a model could draft text, but could not reliably determine current state, permitted actions, evidence requirements or approval authority.
A single replacement programme would have introduced significant cutover risk across many business processes. Global Advisors instead needed a migration pattern that could deliver visible improvements module by module while preserving continuity.
GA AI Case Study – Modernising business-process software for human-and-AI work
The architectural response
We established a shared presentation and control surface for migrated modules, but deliberately avoided recreating the legacy monolith. The target architecture separates:
- Experience: user forms, dashboards, workbenches and review queues.
- Process state: explicit statuses, transitions, ownership and deadlines.
- Business services: validation, calculation, integration and durable transactions.
- Decision rules: policy and approval logic that can be inspected and changed.
- Evidence: source documents, extracted fields, user edits, review actions and final records.
- AI assistance: bounded tasks such as classification, extraction, drafting, anomaly highlighting and next-action support.
- Human authority: decisions, approvals, exceptions and accountability.
Stable task and action interfaces allow a user interface, workflow engine or authorised agent to invoke the same governed operation.
GA AI Case Study – Modernising business-process software for human-and-AI work
What we did
The programme created a common migration vehicle and then used it across processes:
- built shared component, settings, administration and asset patterns;
- migrated user-facing forms and dashboards ahead of full backend extraction where this reduced risk;
- kept transitional access to legacy data explicit and replaceable;
- created service-like interfaces for checklist, action and approval work;
- introduced review workbenches for finance and administrative processes;
- separated submission, enrichment, review and finalisation stages;
- added structured capture and learning controls for processes where classification improves over time;
- made role, owner and state visible rather than implicit in page flow;
- moved integrations toward middleware and API contracts;
- kept module maturity distinct so one successful surface did not imply the whole portfolio was complete; and
- documented retirement paths for direct database and shared-helper dependencies.
GA AI Case Study – Modernising business-process software for human-and-AI work
How AI was included
AI participation was designed around bounded process steps, not autonomous end-to-end control.
Suitable tasks included:
- extracting candidate fields from semi-structured documents or messages;
- classifying an item into a controlled taxonomy;
- suggesting a match based on prior reviewed outcomes;
- drafting a summary, response or performance narrative from authorised facts;
- identifying missing information or an unusual pattern;
- routing work to the right queue; and
- helping a user understand the next permitted action.
The system retains the distinction between suggestion and decision. Proposed values are visible, editable and attributable. High-consequence actions pass through existing approval or posting controls. Feedback from corrections can improve future assistance without allowing the model to rewrite policy implicitly.
GA AI Case Study – Modernising business-process software for human-and-AI work
Difficult problems we had to solve
Migration sequence and architecture sequence differ
The cleanest diagram would replace every backend before changing the interface. The safest operational sequence often moved the user experience first while retaining legacy tables or helpers. We treated this as controlled coexistence with a defined extraction path, not as architectural completion.
Shared platforms can become new monoliths
A common module framework created leverage in assets, configuration and interaction patterns. It also created a temptation to place every rule and integration in the presentation layer. We kept the intended boundary visible: thin interaction at the front, process and integration logic behind contracts.
AI confidence does not equal process authority
A highly confident classification may still require review because the business consequence, not model certainty, determines control. We designed human checkpoints around the impact of the action and the reversibility of error.
Historical data encodes exceptions
Legacy processes contain corrections, workarounds and ambiguous records. AI trained or prompted from history can reproduce those inconsistencies. Taxonomies, examples and learning sets therefore needed curation, versioning and exclusion of unsuitable cases.
Idempotency matters when agents act
A user double-click is inconvenient; an agent retrying a financial or workflow action can create duplicate state. Action endpoints need replay protection, transaction boundaries and receipts so a caller can determine whether a requested change already occurred.
GA AI Case Study – Modernising business-process software for human-and-AI work
Controls and assurance
- Server-side validation and role checks on state-changing operations;
- explicit workflow states and permitted transitions;
- human review for consequential suggestions and exceptions;
- idempotent action contracts and durable receipts;
- separation of model proposals from policy rules;
- visible source, extracted value, correction and final decision;
- module-specific maturity and rollback planning;
- controlled coexistence with legacy records; and
- progressive replacement of direct data access with governed services.
GA AI Case Study – Modernising business-process software for human-and-AI work
Results
Global Advisors created a practical path for renewing important operational software without stopping the business or waiting for a perfect replacement platform. Multiple processes now operate through a more consistent experience and stronger control surface, with service-like actions that are suitable for authorised human or agent use.
The programme also produced a reusable judgement framework for AI-enabled process redesign: start with state, authority and evidence; choose bounded assistance points; preserve review where consequences require it; and measure the whole process outcome rather than the quality of generated text.
GA AI Case Study – Modernising business-process software for human-and-AI work
What we learned
AI exposes weak process architecture. If ownership, state, rules and evidence are implicit, adding a model magnifies ambiguity. The work required to make a process safe for AI often improves it for people as well.
We also learned that incremental modernisation is not a compromise when it is governed. A coexistence-first path can reduce risk and accelerate learning, provided transitional dependencies are visible and there is a credible route to remove them.
GA AI Case Study – Modernising business-process software for human-and-AI work
Why this matters for leaders
Leaders evaluating AI in core processes should ask:
- Is the process state explicit enough for an authorised machine actor?
- Which steps are suggestions, decisions or irreversible transactions?
- What evidence must accompany each action?
- How will retries and duplicates be handled?
- Can front-end renewal proceed safely before full backend replacement?
- How will roles change when routine preparation becomes machine-assisted?
The business case should be built around cycle time, quality, control effort and changed roles, instead of simply counting the AI features added.
GA AI Case Study – Modernising business-process software for human-and-AI work
Note
Global Advisors does not perform technical AI implementation or systems integration for clients. However, we have worked on architecting and implementing AI at a deep level in our own business since the beginning of 2024. This allows us to provide grounded AI strategic and architectural advice based on a deep hands-on knowledge of AI. We work with clients to build strategies, business and operating models to win in an AI enabled world. We help them make architectural and partner choices for implementation and work with them to change their businesses in response.
