NederlandsProcess automation with RPA and AI.
Eliminate repetitive work, improve data quality and speed up throughput — by combining RPA and AI intelligently. Once prerequisites are clear and APIs are documented, the technical build is surprisingly straightforward.
Two technologies, one goal: less manual work.
RPA and AI solve different problems. RPA follows rules — perfect for stable, structured processes. AI makes judgements — essential for unstructured input like email, PDFs or free text. The biggest gains come from combining both.
Robotic Process Automation
Software bots that precisely replicate human actions — filling forms, transferring data between systems, building reports. Predictable, fast and cheap to scale.
AI & LLMs
Models that understand language, images and context. They classify, summarise, extract and make data-driven decisions — even when input is messy.
RPA + AI together
The bot handles the steps; AI does the thinking. Email arrives → AI reads and classifies → RPA transfers to ERP, writes note and triggers follow-up workflow. End-to-end, with audit trail.
Where RPA & AI deliver immediate time savings.
Six processes we frequently see at organisations. Not all automation needs to start big — often one well-defined process is enough to build the business case.
Purchase invoice processing
AI reads invoice (PDF/email), identifies supplier, line items and VAT. RPA books in ERP, links to PO and queues for approval. Exceptions: human-in-the-loop.
New employee onboarding
From the HR system, RPA generates contracts, accounts, authorisations and hardware tickets. AI personalises standard texts and checks for missing fields.
Incoming email triage
AI classifies intent and urgency, extracts customer and order data. RPA opens ticket, fills fields, sends confirmation and routes to the right department.
Order & packing slip processing
Scan packing slips, compare to order, flag discrepancies. RPA books receipt and triggers invoicing. No more manual typing in WMS.
KYC & document review
AI reads contracts and KYC documents, extracts fields and flags risks. RPA archives in DMS and queues review tasks for the compliance officer.
Ticket resolution & routing
AI categorises tickets and suggests responses. RPA executes standard actions (password reset, mailbox permissions) and escalates to second line when needed.
Technology is rarely the problem — preparation is.
A successful RPA or AI project succeeds or fails based on what you agree before the build. Our experience: once the process, owner and APIs are clear, we typically build the first working version in days, not months.
- 01Map the process, not the wish
We walk through the process as it actually runs — including exceptions, workarounds and the spreadsheet nobody knows about.
- 02Owner & KPIs defined
One process owner from the business, one technical lead. With measurable KPIs: throughput time, error rate, FTE reduction.
- 03APIs & access verified
Which systems have a clean API? Where do we need UI automation? Are credentials, scopes and rate limits sorted? This is the critical success factor.
- 04Build, measure, scale
First, get one process end-to-end to production. Then expand. We build in sprints with the business — no big bang.
One workflow, four clear layers.
A well-functioning RPA + AI solution is a chain of clearly separated responsibilities. Each layer has one task and a clear contract with the next.
- 1Trigger & intake
What starts the process? Email, scan, form, schedule or API event. We choose the most reliable signal as the starting point.
- 2AI layer for understanding
Document AI, classification and extraction. Here AI makes the unstructured world machine-readable — with confidence scores per field.
- 3RPA & orchestration
The bot executes the steps. High confidence: automatic. Low confidence or exception: human-in-the-loop with context.
- 4Systems of record
ERP, HRM, CRM and DMS. Preferably via API; UI only when absolutely necessary. With audit log and idempotent write actions.
Meet your automation partners.
Two senior partners who guide the entire journey — from process inventory to production and maintenance. With experience in Finance, Manufactoring and Logistics.
Translates processes and business goals into achievable automation. Works closely with process owners, finance and HR to choose the right first step.
Book an intake with Marten →
Designs and builds the infrastructure for the RPA process in combination with AI and ERP systems. Hands-on knowledge of APIs, queues and exception handling.
Plan a tech deep-dive with Tanmay →When these are in order, the build becomes surprisingly straightforward.
The weight of a successful RPA + AI project sits before the first line of code. Our experience: once the prerequisites are clear — and APIs are documented — technical realisation is usually the least complex step. Eight points we work through with you as standard.
The actual process — not how it should be — is documented, with start, end, decision points and exceptions.
- Process flow covering 100% of happy-path steps
- Top-5 exceptions and how they are handled
- Volume per day/week and peak periods
- Definition of done for one process instance
- Bot works on 80% of cases — the rest becomes messy manual work
- Scope creep: every exception becomes a mini-project
- No baseline to measure improvement against
Someone with authority, daily process knowledge and the ability to make decisions on exceptions. Not IT — the business.
- One named owner with decision-making authority
- Weekly ritual to discuss exceptions
- Mandate to adjust standard rules
- Decisions don't get made; bot stays 80% ready
- IT and business point at each other during outages
- No buy-in — users bypass the bot
The biggest accelerator. Does every system have a clean, stable API with decent documentation? Then we build fast, robust and maintainable.
- API spec (OpenAPI/Swagger) or working examples per endpoint
- Authentication, scopes and rate limits mapped
- Sandbox or test environment available
- Data contract: which fields are mandatory, which optional?
- Idempotency & error codes — how does the API behave on retries?
- UI bots that break on every release of the source system
- Unexpected rate limits halt the process
- Unnecessarily high maintenance costs long-term
The bot has its own technical identity with the minimum required permissions — not a colleague's password.
- Own service account per bot (no personal credentials)
- Least-privilege scoping per system
- Secret management via vault (no plaintext)
- Rotation policy on tokens and passwords
- Bot runs on the account of a colleague who leaves
- Audit cannot distinguish human vs. bot
- Security incidents are harder to isolate
The bot can't be better than the data it works with. Duplicate customers, missing fields or inconsistent code lists lead to errors you have to fix elsewhere.
- Master system per data entity (customer, supplier, product)
- Mandatory fields & validation rules documented
- Data cleanup plan for polluted tables
- Bot neatly delivers error-upon-error at scale
- Reconciliation costs more time than the old manual flow
- Trust in automation is gone within a quarter
Which data may the AI layer see? Which models may we use, and where do they run? This must be resolved before the build — not after.
- DPIA for processes involving personal data
- EU hosting requirement on models (GDPR / EU AI Act)
- PII masking before external model calls
- Retention periods for logs and intermediate results
- Customer data ends up in public model training
- Audit blocks the production go-live
- Fines and reputational damage
Which cases may the bot handle independently, and when must a human review? How quickly does that human respond? This is a process agreement, not a technology choice.
- Confidence threshold per decision
- Work queue for exceptions + SLA
- Escalation path on outage or unknown error
- Exceptions pile up into a backlog
- Nobody knows if the bot is running or stuck
- Customer or employee gets a response too late
A bot is software. Someone needs to look after it. Ideally the same person who owns the process, supported by dashboards that are clear at a glance.
- Dashboard with volume, throughput time and error rate
- Alerts on outage, AI output drift or API errors
- Monthly improvement meeting with the business
- Bot runs incorrectly in silence; nobody notices
- Improvements are never made — ROI flattens
- After staff turnover: nobody understands the bot anymore
Your processes and data stay in Europe.
GDPR and EU AI Act are not an afterthought — they are built into the design. Bots run on controlled infrastructure, AI models on EU-hosted infrastructure, with full audit trail per process instance.
- 1 Step 1. Your data and model in the EU — fully GDPR-compliant.
- 2 Step 2. AI policy & training — awareness, prompts, agents and spreadsheets.
- 3 Step 3. Security, quality, consistency and governance implemented.
Personal data processed in accordance with GDPR requirements.
Risk-based approach in line with the European AI Regulation.
Automatic detection and masking of privacy-sensitive data.
Full audit trail and documented governance.
A healthy integration as the foundation under every bot.
Our 'healthy integration' approach ensures data is exchanged cleanly, traceably and repeatably between systems — before we apply automation. No brittle scripts, only durable connections.
- Clear data contracts between source and target system
- Idempotent write actions & clean error handling
- Logging at process instance level, not API level
- Fallback strategy for API outage or UI changes
Ready to get started?
In a free 60-minute scan we look at one process you have in mind. You get a go/no-go on feasibility, an initial business case estimate and — if it fits — a concrete first sprint.
Ready to automate your first process?
KODIFY uses RPA and AI to eliminate repetitive work and get the people in your organisation back to meaningful work. Senior experts, embedded in your team — from first scan to production and continuous improvement.
