To choose data analysis software for a small or mid-sized business, start with the questions management asks every week and with who will ask them, and only then compare tools. The criteria that matter most are the connection to the data the business already has, how independent the people asking can be, whether every figure can be checked, and control over who sees what.
This guide explains the types of solution available, the criteria that separate a good choice from a tool nobody uses, the questions to ask the supplier and a simple way to test before deciding.
What types of data analysis software are there?
In most small and mid-sized businesses, data analysis runs through one of four kinds of solution, and often through several at once.
| Type of solution | Where it works well | Where it tends to fall short |
|---|---|---|
| Spreadsheets | One-off analysis, few sources, one person in charge | Copies and competing versions, manual cross-referencing, errors that are hard to find |
| ERP reports | Tax reports, listings and standard indicators | Questions the report did not anticipate and combining other sources |
| Dashboard tools | Following the same indicators over time | Every new question depends on whoever can build the dashboard |
| Natural language analysis | New questions asked by managers, without waiting for a report | Requires checking how answers are verified and who can see each piece of data |
None of these types is good or bad in itself. The choice depends on what the organisation asks and who asks it.
Which questions does the business need to answer?
Before watching any demonstration, write down the ten questions the management team asked over the last two months that took longest to answer. For example, which customers reduced their orders this quarter, which products lost margin, or how the average collection period has changed.
Then split them into two groups.
- Recurring questions, asked every week or every month, which call for a dashboard or a saved report.
- New questions, which come out of a meeting or a deviation and need an answer the same day.
If most are recurring, a good set of dashboards will do. If many are new, the deciding criterion is how quickly a non-technical person can get to the answer.
Which criteria matter when choosing?
- Connection to the data you already have. Check which database engine runs your ERP, CRM or invoicing software, and which files the team uses. A tool that forces you to migrate data or export everything by hand loses its appeal within weeks.
- Independence for the people asking. See who can ask a new question without asking IT or someone who knows SQL for help.
- Verifiable answers. Every figure should be traceable to its source. Ask how you can see the query behind a result and the data it returned.
- Access control. Margins, salaries and customer data are not for everyone. Check whether you can define what each person is allowed to query.
- A log of what was asked. Knowing who asked what, and when, supports auditing and the team's trust.
- Ongoing monitoring. Beyond answering questions, the tool should let you save dashboards and warn you when an indicator drifts.
- Where it runs. On the organisation's own infrastructure or the supplier's, depending on security requirements.
What should you ask the supplier?
- Which databases and files does it connect to, and what happens if our ERP runs on another engine?
- Do we need to copy or migrate the data somewhere else?
- How do we check a figure that appears in an answer?
- How do we restrict each person's access to each table?
- What is logged for each question, and is the returned data stored?
- How long does it take from first access to the first result on our own data?
The answers to these questions say more than any feature list.
How can you test before deciding?
Imagine a distribution company with 80 people, where the sales director asks every Monday for an analysis of sales by customer and waits two days for it. A good trial lasts a few weeks and follows four steps.
- Connect one real source, such as the sales database or a file exported from the ERP.
- Ask the ten questions from the initial list, through the people who ask them day to day.
- Check three answers against figures the business already knows.
- Measure how long each question took compared with the current process.
If the answers match and arrive the same day, the tool works for you. If the team goes back to the spreadsheet after a week, it does not, however complete it may look.
Where does EngiAnalytics fit in?
EngiAnalytics is Engibots' conversational data analysis platform and one of the options worth considering when many of your questions are new. Analysts ask in natural language, in Portuguese, Spanish or English, and receive the answer as a chart, text or table. The SQL query behind each answer can be opened and copied alongside the returned data, and if the sentence in the answer quotes a figure that is not in that data, the reader is warned.
It connects to MySQL and PostgreSQL databases, including ERP and CRM databases running on those engines, and to CSV and Excel files, with no migration and no replacement of systems. Access profiles define, table by table, what each person can query, and every question is logged without storing the returned data. Personal dashboards and alerts follow the operation day to day, and the platform can run on the organisation's infrastructure or on Engibots'.
If you would like to see how it would answer your own questions, the quickest way is a 20 minute conversation using your examples.
Frequently asked questions
What is the difference between BI software and natural language analysis?
Traditional BI software shows dashboards built in advance for the indicators that were planned. Natural language analysis answers new questions written as you would ask a colleague, without waiting for someone to build a dashboard.
Is a spreadsheet not enough to analyse company data?
It is enough for one-off analysis with few sources. When several sources have to be combined every week, copies and competing versions make answers slow and hard to check.
Which criterion matters most?
It depends on the questions. If most are new, independence for the people asking matters most. In every case, each figure in an answer should be traceable to its source.
How can you test a tool before deciding?
Connect one real source, ask the ten questions that took longest to answer in recent months and check three answers against figures the business already knows.
What data does EngiAnalytics connect to?
MySQL and PostgreSQL databases, including ERP and CRM databases running on those engines, and CSV and Excel files, with no migration and no replacement of systems.