Coding worked — but it was slow and invisible. No way to copy across lines, no way to see total status, and errors only surfaced on submit. Multi-line orders meant repeating the same coding 20+ times.

• Line by line, every time — even when all lines shared the same cost center.
• No bulk action — no copy, no template, no propagation.
• Invisible errors — discovered only at submission, causing delays and rework.

Three validation rounds with Inframark, Stena Line, and Alleima — different volumes, different structures. The same story every time: coding wasn't broken, it was repetitive.

Three moves carried the redesign. Inline coding panel: click a row, panel opens in place — no context switch. Bulk actions: select multiple lines, apply once. Coding status per row: complete / partial / missing visible at a glance before submit.

100% task success across all three validation sessions. Significant reduction in clicks on multi-line orders. Phase two — coding templates and copy-from-previous-order — is planned based directly on what surfaced in testing.

Most approvers had no reason to open Medius except to approve. The friction — login, navigation, context switch — was enough to delay decisions by hours. Some approvers batch-approved without reviewing details just to clear the queue.

The card had to communicate enough context to decide — and nothing more. Lines, supplier, total, budget status, latest comment. Every field tied to a real approval input and tested against Outlook desktop, web, and Teams rendering quirks.

Security was a design layer, not an afterthought. Cards tie to Azure AD / Entra; forwarded emails are rejected by the back-end; every action is audit-logged with channel = "Actionable email".

Approvers can now approve or reject purchase requisitions without ever opening the platform. The interface feels as casual as a one-tap reply; the audit trail satisfies enterprise compliance. Zero logins required.

Duplicate suppliers created reporting noise, split invoice history, and caused compliance gaps. The original flow let users jump straight to a blank creation form — no search, no check, no warning.

• No pre-creation search — duplicates were easy to create accidentally.
• Manual data entry — VAT numbers, addresses, industry codes entered by hand and often incomplete.
• No risk signal at creation — credit risk and company health were only visible after the fact.

Search-first design: the modal opens with a search field instead of a blank form. Typing a company name triggers a fuzzy-match check against existing suppliers in real time — showing potential duplicates with a match confidence score before the user can proceed.

If the user continues to create, enrichment data from an external provider (Dun & Bradstreet) auto-fills company details, industry classification, headcount, and credit risk signal — reducing manual entry and improving data quality from day one.

Duplicate supplier creation dropped significantly in UAT. Onboarding time fell as auto-enrichment replaced manual data entry for core fields. Procurement teams gained a risk signal at the point of creation — not weeks later during an audit.

Finance, procurement, and legal each managed contracts separately. There was no single source of truth, no system to flag expiring agreements, and no way to understand what a clause actually meant without involving a lawyer.

• Fragmented storage — contracts spread across email, shared drives, and spreadsheets.
• No deadline monitoring — renewals and expirations tracked manually in calendars.
• Opaque language — legal terms inaccessible to the procurement and finance users who needed to act on them.

Stakeholder workshops with finance, procurement, and legal mapped the real workflows. Three distinct personas emerged — each needing the same contracts, but different signals from them.

The solution: a centralised repository with smart intake forms, automated spend and status dashboards, and an AI layer that flags risky clauses, translates legal language into plain English, and supports multiple languages for global teams.

Critical deadlines and risky clauses flagged automatically. Manual contract monitoring significantly reduced. All three user groups — finance, procurement, legal — working from a single, searchable source of truth for the first time.

ERP XT was built around internal accounting logic, not user mental models. Module names like "Cash & Bank" and "Accounting" meant nothing to a bakery owner or freelance consultant. Workflows required 6–8 steps for tasks users expected to take two.

• Terminology mismatch — accounting language inaccessible to non-expert users.
• Dense forms — cognitive overload on screens used infrequently.
• Too many steps — bank statement import required excessive clicks with no progress signal.

24 users across three groups — internal consultants, active clients, external partners. Remote sessions and in-person prototype testing revealed the same pattern: plain language and fewer steps mattered more than visual polish.

We restructured forms into logical sections, replaced jargon with plain-language labels, added contextual tooltips, converted unclear icon buttons to descriptive text, and introduced a 3-step progress pattern for the most complex flows.

Faster task completion, fewer errors, and a marked reduction in support requests related to navigation and terminology. Beta feedback consistently highlighted language clarity as the most valued improvement.