Phased legacy ERP migration for uninterrupted transformation
Phased migration of legacy ERP systems in the public sector helps avoid downtime and ensures process continuity by focusing on automation and…
Low-code BPM, RPA, AI agents in workflow. Automation as an engineering discipline, not "buy a robot".
The BPM practice covers design and automation of business processes through a combination of low-code platforms (Scriptum on UnityBase), RPA for legacy scenarios, AI agents for unstructured data processing. Covers approval, request handling, financial operations, HR flows.
In 2026 the key trend is AI agents inside workflow: not as a human replacement, but as a first-level processor that removes 60–70% of routine work.
Do not evaluate AI automation on pilot — evaluate on stable production phase after 3–6 months. Pilots are too clean to be true.
Spreading low-code to business users without architecture guardrails creates shadow IT with zero documentation. The architect remains mandatory.
RPA pays off only when an API is unavailable and the legacy will be decommissioned in <2 years. Otherwise — invest in proper integration.
It is more honest to say "you do not need this yet" than to sell an engagement that will not deliver ROI.
Interviews with process owners, mining actual flows from email/ticketing, identifying exceptions and shadow processes.
Platform choice (Scriptum / Camunda / RPA), workflow design accounting for exceptions, ERP/CRM/ECM integration, governance.
Usually we start with request handling or contract approval. Doomed-to-fail scenarios — high-frequency, high-stakes processes with unknown exceptions.
Adding AI agents for request classification, data extraction, routing. Only after the basic workflow is stable.
Monthly process mining, identifying new bottlenecks, optimizing AI models for new request types.
Project lead: InBase (Scriptum, UnityBase) and Softengi (AI agents, RPA).
Brought in when needed: Softline (integration with government systems).
They automate the current process "as is", without optimization. Result — the same inefficient process executing faster. Chaos scales.
What we do instead: process redesign before automation. First we ask "is this step needed at all", then — how to automate it.
They expect the AI agent to design the workflow itself. In production the agent makes decisions that break compliance. They blame AI, not design.
What we do instead: AI agent inside a workflow designed by an architect. Clear boundaries: what AI can decide, what escalates to human.
Instead of 2 weeks on API integration — 3 days on an RPA bot. A year later: 200 bots, each breaking on legacy updates, support team grows linearly with bot count.
What we do instead: RPA — only when (a) API is unavailable and (b) legacy will be decommissioned in <2 years. Otherwise — invest in proper integration.
No precise savings percentages — actual numbers depend on the client's starting point. Instead — concrete architectural decisions and organizational changes.
The agency processed 5000 requests/month manually, time from registration to response — 14 days. Scriptum for workflow + AI classification for routing. Time dropped to 3 days.
AI agent checks customer documents against 12 sources (government registries, sanctions lists, internal blacklists). Escalation to human — only for conflicts and borderline cases. 70% of applications processed without human.
Previously — procurement approval from request to tender took 6 weeks. Scriptum with parallel approval tracks for different categories. Time dropped to 12 days.
Nine recent expert articles — from thematic overviews to specific architectural decisions.
Scriptum · UnityBase · Camunda · Bonita · Mendix · OutSystems
UiPath · Microsoft Power Automate · Automation Anywhere · Blue Prism · Robocorp
OpenAI · Anthropic Claude · LangChain · LangGraph · Microsoft Copilot Studio · n8n
Celonis · UiPath Process Mining · SAP Signavio · ARIS Process Mining
n8n · Make · Zapier · Apache Camel · MuleSoft
BPMN 2.0 · CMMN · DMN · ISO/IEC 27001 · GDPR
Scriptum — for custom low-code business applications, especially if you already use UnityBase. Camunda — for complex BPMN scenarios with a developer-first approach. UiPath — for RPA with legacy that lacks APIs. Projects often use a combination.
In 2026 AI agents close first-line (typical queries, FAQ, routing) — 60–70% of requests. Second/third-line — still human, requires judgment and context. Complete human replacement — still marketing, not reality.
Pilot on one process — 2–3 months. Platform for a mid-size organization — 6–12 months. AI layer on top of working BPM — additional 3–6 months.
Process mining — analysis of actual process flows from event logs across systems. Identifies shadow processes, deviations from intended workflow, bottlenecks. Pays off for organizations with >5 enterprise systems and complex processes.
Three metrics: (1) time-to-completion (from request to completion); (2) error rate (errors per 1000 instances); (3) cost-per-instance (including infrastructure and licenses). ROI calculated 6+ months after production launch — not on pilot.
Real projects rarely fit in one competency. See which other areas we work in.
30-minute discovery call with a BPM architect. We will see if it is worth it, where to start, how long it will actually take.