Feb 9, 2026
Not all automation is the same. How to choose the right one?
We explain the most commonly used types of automation. Finally, we provide a real-world example of report automation.
AI is driving an unprecedented productivity boom. As a consequence, companies are looking to automate processes and recurring tasks and implement AI agents. We were already going through a digital revolution, but now it feels like “automate or risk being left behind.”
The combination of urgency and uncertainty is dangerous for decision-making, and strategic mistakes are being made when implementing automation.
In this article we discuss different types of automation and when each is most suitable.
What types of automation and tools should you evaluate before you start?
Let’s begin by understanding different ways of automating. It’s not the same thing to have a bot that exports an Excel file and emails it as it is to have a pipeline that automatically retrains a credit risk model.
For simplicity, we can classify automation into four broad types:
1. Operational task automation (RPA and workflows)
Tools like UiPath or Automation Anywhere allow automation when APIs aren’t available.
Other platforms like Make or n8n let you orchestrate flows between systems via API without writing much code. These are very useful and recommended when a process is repetitive, structured, and stable.
If the process changes frequently or depends on complex human validations, the automation becomes fragile.
2. Data flow automation (ETL/ELT)
This type focuses on automating the movement and transformation of data.
It includes:
- extracting data from databases, APIs, or files,
- transforming data and validating business rules,
- loading it into databases so the information is up-to-date and high-quality.
This requires architecture design, SQL skills, and knowledge of cloud services.
3. Analytics automation (models and rules)
This automates decision-making using statistical or machine learning models developed in Python with libraries like scikit-learn, XGBoost, or statsmodels.
Typical applications include:
- risk scoring,
- customer segmentation,
- demand forecasting,
- anomaly detection.
Success here depends on deep understanding of the business, statistics and math, result validation, and monitoring model performance over time.
4. AI-based automation (RAG and agents)
Here large foundational models are connected to internal information through architectures like RAG (retrieval-augmented generation), which combine information retrieval with text generation.
Platforms like AWS Bedrock, Azure OpenAI, or frameworks such as LangChain make it possible to build assistants that:
- query document repositories,
- summarize information,
- generate contextualized responses.
Simple cases can be implemented quickly, but robust enterprise use requires data governance, access control, and quality monitoring.
What kinds of problems are companies referring to when they say “we need automation”?
"The team works weekends closing reports"
"Data doesn’t match across departments"
"Decisions are late because information doesn’t arrive"
"There is pressure to use AI to stay competitive"
The need is often clear, but companies tend to stall because they don’t know how to move forward. In that context:
- If the problem is manual operational time, consider RPA or process redesign.
- If the issue is inconsistent data across systems, data flow automation (ETL) and automated validations should be the focus.
- If the challenge is slow or subjective decisions, it’s worth evaluating analytics models.
- If the problem is access to dispersed information, an agent built with RAG architecture might make sense.
This logic is straightforward, but it’s worth emphasizing: choosing the wrong type of automation impacts budgets, exhausts teams, and creates barriers for future innovation.
A poor technical choice becomes a strategic problem.
Automation is about changing how work and information flow. It directly impacts how you operate.
When you choose the right type of automation, you reduce friction, improve response times, increase traceability, and build trust in data and processes, both internally and externally.
Now, if the wrong choice is made, it will most likely result in more complex processes, unnecessary tools that confuse and do not contribute.
Let's conclude with a real-world example of data flow that replaces weeks of manual work.
The finance department consolidates information from the ERP, CRM, accounting system, and various Excel and CSV files received from other departments.
Previously, everything was done manually, downloading files from the systems, copying and pasting historical data into Excel spreadsheets, and replacing information used in previous reports.
The flow that can be visualized below:
1. Automatically extracts data from the ERP, CRM and accounting system.
2. Applies cleaning and validation rules and relates it to the files received from other areas, which always have the same structure.
3. Standardize formats and create new fields as needed for the report.
4. Load information into the data warehouse.
5. Populate the dashboard with the periodic report.
Result: decisions can now be made quickly and with reliable information.
Felipe Uribe Velásquez
Partner - Analytics and Data Engineering
Camilo Monsalve Maya
Data Engineering Consultant
