Data analysis for companies: methods and where to start

Line drawing of a spreadsheet grid and a bar chart under a magnifying glass, with one bar highlighted in blue

The five types of analysis, the business question each one answers, and a practical way to begin with one decision rather than a data project.

In short

Data analysis is the work of turning business data into answers that people act on. Companies use five main methods: descriptive (what happened), diagnostic (why), predictive (what is likely next), prescriptive (what should we do) and AI-assisted analysis of text and other unstructured data. Start with one recurring decision, the data behind it and a named owner.

Most companies already hold more data than they use: sales in the CRM, invoices in the ERP, tickets in the support tool and a folder of spreadsheets that someone updates every Monday. What is usually missing is not data but the habit of asking it precise questions. This guide from Devies, a Gothenburg-based product and engineering consultancy, is for managing directors, CFOs and department heads who want data analysis (dataanalys, often searched as dataanalys företag) to change decisions rather than produce more reports.

What is data analysis?

Data analysis is the process of inspecting, cleaning and modelling data to answer a specific business question and support a decision. The question comes first; the data, the method and the tool follow from it.

Three neighbouring things are often mixed up with it:

  • A report is a fixed presentation of figures, produced on a schedule. It tells you the state of things but rarely why.
  • A dashboard is an interactive report for monitoring. It is useful once you know which numbers matter.
  • An analysis is a piece of work with a question, a method and a conclusion, such as "why did repeat orders fall in the west region in the third quarter?"

A useful test is whether the work ends in a sentence that starts with "so we should". If it ends in a chart and no recommendation, it was reporting, not analysis.

Analysis also differs from the roles that support it. Data engineers build the pipelines that deliver data, and data scientists build statistical and machine learning models. Our comparison of data science, data engineering and data analytics explains who does what and which role to hire first.

Which data analysis methods should a company know?

A company needs to know five methods of analysis, each answering a different question, and most of the value in a mid-sized business comes from doing the first two well. Microsoft's training module on the types of data analytics uses the same five categories and shows how they build on each other.

Method

Question it answers

Business example

What it needs

Descriptive

Question it answers

What happened?

Business example

Sales by channel and month

What it needs

Clean, consistent history

Diagnostic

Question it answers

Why did it happen?

Business example

Which discount levels close deals

What it needs

Related data on causes, such as discounts, campaigns and owners

Predictive

Question it answers

What is likely to happen next?

Business example

Expected order intake next quarter

What it needs

Enough stable history and a way to test forecasts

Prescriptive

Question it answers

What should we do?

Business example

Which price change maximises profit

What it needs

A model of the trade-off and agreed goals

Cognitive or AI-assisted

Question it answers

What do large volumes of text, images or signals tell us?

Business example

Themes in thousands of support tickets

What it needs

Access to unstructured data, plus human review

The methods are steps on a ladder. A forecast built on revenue figures that finance and sales define differently will predict the wrong thing very precisely. Our explainer on predictive analytics covers when forecasting is worth the effort and what it requires.

Diagnostic analysis is where many companies stop too early. Knowing that margin fell is descriptive; knowing that it fell because large orders were discounted twice, once in the quote and once at invoicing, is diagnostic, and it is the finding that changes behaviour.

How do you start with data analysis in a company?

You start by choosing one recurring decision, writing the question it depends on and answering it with the data you already have, before you buy tools or build a platform. The steps below take a few weeks for a first question, not months.

Loop of six stations from a decision through question, data, quality check and chart to action, back to the decision
  1. Pick a decision, not a dataset. Choose something that is decided repeatedly and costs money when it is wrong: purchasing volumes, pricing, staffing, which customers to call. "Explore our CRM data" is not a starting point.
  2. Write the question precisely. Name the measure, the period, the segment and the comparison: "Which product groups lost gross margin between the first and second half-year, and how much of the change came from discounts?"
  3. Decide in advance what would change the decision. Agree with the decision owner what result would make them act differently. This keeps the analysis honest and short.
  4. Find the data and its owner. List the systems involved, who owns each and how far back the history goes. A missing field discovered now is cheaper than one discovered after the presentation.
  5. Check quality before you analyse. Look for duplicates, gaps, changed definitions and outliers. Our guide to measuring and fixing data quality shows which checks to run first.
  6. Use the simplest method that answers the question. A spreadsheet or a BI tool such as Power BI is enough for most descriptive and diagnostic work. Move to statistical models only when a simpler comparison cannot answer the question.
  7. Act, measure and make it repeatable. Record the decision taken, follow up on the result, and if the question recurs, automate the data and the calculation so the next analysis starts from a trusted base.

The point of the first round is not the answer alone. It shows where the data is weak, which definitions people disagree on and how long a real question takes, which is the evidence you need before investing in a larger platform.

What does data analysis need from your data and your people?

Good analysis needs three things in place: data with stable definitions and enough history, people who understand both the numbers and the business, and clear rules for what data may be used for which purpose.

On the data side, agreed definitions matter more than volume. If "active customer" means one thing in the CRM and another in finance, every analysis starts with an argument. History matters for comparisons and forecasts, and access matters because analysts who wait two weeks for an export stop asking questions.

On the people side, the most valuable analysts sit close to the business. They know that a spike in returns in January is normal, and they can tell the difference between a data error and a real change. A named decision owner on each analysis keeps the work tied to action.

Personal data adds legal limits. Under the GDPR, Regulation (EU) 2016/679, Article 5 requires that personal data is collected for specified purposes, not processed in ways incompatible with them, and limited to what is necessary. Customer-level analysis is often possible, but check the purpose and the legal basis first, and use aggregated or pseudonymised data where individual records are not needed.

Devies saw the value of putting analysis directly into operational tools with Cetasol. According to the case, the work on the iHelm platform included improving data quality in the interface and a new performance page where users compare vessels and operators on fuel efficiency, speed patterns and other operational KPIs, designed to support fleet management and performance analysis.

Which mistakes make data analysis fail?

Analysis usually fails for organisational reasons rather than technical ones: the question is vague, the definitions move, or nobody owns the decision the analysis was meant to support.

  • Analysis without a decision. Work that answers "what is interesting in this data?" produces slides, not change.
  • Treating correlation as cause. Customers who use a feature may buy more because they were already your best customers. Check for other explanations before you act.
  • Averages that hide segments. A stable overall margin can hide one region falling and another rising. Break results down before you conclude.
  • Moving definitions. If revenue is calculated differently in two reports, fix the definition before you analyse the trend.
  • Platform first, questions later. A data platform is valuable once several recurring questions depend on the same data. Building it before you know those questions is the most expensive way to start.

Devies

Want a second opinion?

Devies, a Gothenburg-based product and engineering consultancy, works on questions like this with Nordic product and IT teams. Tell us where you are and we will suggest a sensible first step.

Talk to Devies

Frequently asked questions

What are the main types of data analysis?

The main types are descriptive analysis, which shows what happened; diagnostic analysis, which explains why; predictive analysis, which estimates what is likely to happen next; and prescriptive analysis, which compares options to recommend an action. Many sources add a fifth, cognitive or AI-assisted analysis, which processes text, images and other unstructured data at a scale people cannot handle manually. Each method builds on reliable results from the ones before it.

Do small and mid-sized companies need a data analyst?

Not always a full-time one at the start. Many companies begin with a business person who is strong in Excel or Power BI and a clear question from management. A dedicated analyst becomes worthwhile when several departments ask recurring questions, when the same data is reconciled by hand every month, or when decisions are delayed because nobody has time to look at the numbers properly.

Which tools are used for data analysis in companies?

Most business analysis is done in spreadsheets and BI tools such as Microsoft Excel and Power BI, often on top of data from ERP, CRM and finance systems. SQL is used to query databases, and Python or R are used for statistical modelling and machine learning. The tool matters less than a precise question, clean data and agreed definitions; switching tools rarely fixes an unclear question.

How long does a first data analysis project take?

A focused first analysis on data you already have usually takes a few weeks: time to agree the question, gather and check the data, analyse it and present a recommendation. It takes longer when data is spread across systems with different definitions, because that has to be resolved first. Treat the first round as a way to learn where your data is weak before planning anything larger.

Sources

Here’s more

Join Our Newsletter

We handle your data according to our privacy policy.

Shopping Cart

We use cookies

We use cookies to improve your experience on this website. You may choose which types of cookies to allow and change your preferences at any time. Disabling cookies may impact your experience on this website. You can learn more by viewing our Cookie Policy.

Check out

Contact Form text

We will contact you text

We handle your data according to our privacy policy.