"Automation" brings different things to mind: a software robot clicking a screen (RPA), integration connecting systems, rule-based workflow or AI understanding text. Each of these approaches suits a different kind of process. A wrong choice leads to brittle automation, needless cost or faulty results.
Approaches and their differences
1. Workflow and rule-based automation
"If this happens, do that" logic: when a form arrives, send an email, request approval, create a record. With structured input, clear rules and systems with APIs it is the sturdiest and cheapest method; see digital workflows and low-code and no-code.
2. Integration (API)
Systems exchanging data directly with each other: an order into the ERP, a booking into the PMS, a payment into accounting. Durable, fast and reliable; see API integration. It should always be the first choice where possible.
3. RPA (robotic process automation)
Software robots that mimic the clicks, copy, paste and data entry a person does on screen. Used to work with old systems that have no API. Suited to rule-based, repetitive and structured work.
- Strength: Automation without touching old systems, a quick win.
- Weakness: Brittle; it breaks when the interface changes. It does not understand unstructured data. It needs maintenance.
4. AI
Understanding unstructured input such as text, documents, images and audio; classification, extraction, summarisation, suggestion and prediction. Flexible for vague and variable processes; but it is probabilistic, sometimes gives wrong results and needs verification; see AI process automation, AI document processing and AI agents.
5. Combined (intelligent) automation with AI
In practice most good solutions are a combination: AI reads the document and extracts the fields, rule-based workflow validates and routes, integration or RPA passes the data to the system, and at a critical step a person approves.
Decision frame
Ask the process these questions:
- Is the input structured? If form, table or API then workflow or integration; if free text, documents or email then AI.
- Are the rules clear? If clear and unchanging then rule-based or RPA; if dependent on interpretation and context then AI.
- Does the system have an API? If so, integration; if not and it cannot be changed, RPA.
- Volume and repetition: If high and frequent, automation investment pays back fast; for rare work it generally does not.
- Cost of error: If error is costly, human approval and verification; if low, more autonomy.
- Is the process changing? If it changes often, avoid brittle RPA; choose a flexible approach.
- Is the process healthy? Automating a broken process speeds up the breakage; improve the process first; see the digital transformation roadmap.
Examples
- A form arrives, then a CRM record and email: Workflow or integration; see the CRM guide.
- Entering invoices into an old accounting program (no API): RPA or, if available, file import.
- Reading and matching incoming supplier invoices: AI document processing plus validation rules plus transfer to accounting; see the e-Invoice guide and choosing accounting software.
- Classifying customer emails and preparing reply drafts: AI plus human approval; see AI for customer service and an AI chatbot for your website.
- Synchronising a booking to the PMS and channel manager: Integration; see the channel manager guide and hotel management system.
- Compiling a weekly report: Workflow plus data integration plus AI summarisation; see KPIs and management dashboards.
- Stock order suggestion: Forecast model plus rules plus approval; see demand forecasting and stock planning.
Risks and cautions
Security and authorisation
Automation robots and AI agents reach systems with credentials. Apply separate service accounts, least privilege, secure storage of passwords, action logs and MFA; see the password manager article and zero trust. For AI agents, do not trust external content against prompt injection risk.
Data privacy
Which data goes to which service; see AI and data privacy and the KVKK compliance checklist.
Maintenance and ownership
Each automation needs an owner, documentation and a monitoring and alert mechanism. Ownerless robots break silently; see patch management for dependency updates.
Accuracy and human approval
Especially in financial, legal and customer communication areas, put approval at critical steps. Route low-confidence results to a person.
Data quality
Automation works with data; duplicate and faulty records break automation; see data quality and master data management.
Employees
Automation is meant to reduce repetitive work; plan with employees, train them and explain the change of their jobs openly.
Cost and return
- Cost: Licence, development, integration, maintenance, monitoring, training.
- Return: Hours saved, fewer errors, speed, customer satisfaction.
- Payback period: Calculate as transaction volume times time saved; for low-volume work automation is expensive.
For general budget logic see how to plan a software budget. For the ready-versus-custom decision see custom software or off-the-shelf?.
Implementation plan
- List processes and measure them (time, volume, error).
- Prioritise: High volume, low complexity, clear rules.
- Choose the right approach (integration, workflow, RPA, AI or a combination).
- Pilot: Small scope, human-approved.
- Measure: Time, error, cost.
- Scale and govern: Ownership, monitoring, documentation.
For general starting points, where to start with AI and using ChatGPT at work help.
Common mistakes
- Applying AI or RPA to every problem,
- Building a brittle RPA solution when an API exists,
- Automating a broken process,
- Not putting in human approval and verification,
- Giving robot accounts excessive permissions,
- Not defining ownership and monitoring,
- Neglecting data quality,
- Starting without calculating the return.
ujans runs process analysis, choice of the right automation tool (integration, workflow, RPA, AI), secure implementation and monitoring together.