Alteryx Intelligence Suite: Machine Learning and Text Mining for Finance Data

Alteryx Intelligence Suite is an add-on for Alteryx Designer. According to the Alteryx Help page for the suite, it "offers machine learning, text mining, and computer vision capabilities" on top of Designer's existing features. For finance teams, the parts most often discussed are the machine learning tools (for example, for predicting a numeric or categorical outcome) and the text mining tools (for example, for sorting free-text descriptions into categories). This guide sets out what the official documentation says each part does, what it requires, and where finance data raises extra care. It does not repeat a general Designer workflow walkthrough.

If you are new to the underlying ideas, it helps to read a plain-language primer first, such as our guide to machine learning for accountants, before looking at any one tool.

What the suite contains, according to Alteryx

The Alteryx Help page for the suite groups its tools into three categories. The page does not show a publication date, and its version labelling is inconsistent (see the caution near the end of this post), so treat the lists below as a snapshot and check the current documentation.

• Machine Learning: Assisted Modeling, Build Features, Classification, Data Health, Feature Types, Fit, Predict, Regression and Transformation tools. The Machine Learning overview page also lists Machine Learning Predict and Machine Learning Send, which relate to Alteryx Machine Learning projects.
• Text Mining: Named Entity Recognition, Part-of-Speech Tagger, Sentiment Analysis, Text Classification, Text Pre-processing, Text Summary, Topic Modeling, Word Cloud and Zero-shot Text Classification.
• Computer Vision: Barcode, Image Input, Image Output, Image Processing, Image Profile, Image Recognition, Image Template, Image to Text and PDF to Text.

This post concentrates on the first two categories, because they are the ones most relevant to numbers and narrative text in a finance function.

The machine learning tools

The Machine Learning overview page describes the tools in one line each. A few are worth understanding before you open the software.

Assisted Modeling

The documentation describes Assisted Modeling as a guided option for building predictive models. It walks you through setup and model selection and then adds a pipeline of machine learning tools to the canvas. The steps the page lists are: select the target and method, select an automation level (step-by-step or automatic), set data types, clean up missing values, select features, and select algorithms. For a categorical target the page names Logistic Regression, Decision Tree, Random Forest and XGBoost; for a numeric target it names Linear Regression, Decision Tree and Random Forest. Results can be compared in a leaderboard, and the page says you can export an HTML report or export the pipeline code to a Jupyter Notebook in a Python tool.

The idea of a guided process is useful for finance users, because the choices it asks you to confirm (data types, missing values, which features to include) are exactly the places where a model can go wrong. Our suggestion: treat every recommended setting as a proposal to review, not a decision already made for you.

Data Health

The Data Health tool checks missing values, outliers and sparsity. It produces column-level scores and a report, passes outliers downstream, and can add a column of suggestions for improving data health. The page states that the upstream data must have at least 30 rows. For a finance team, this is a sensible first step before any modelling, and it connects directly to wider data quality discipline; see our guide to financial data quality management for the controls that sit around a tool like this.

Feature Types, Fit and Predict

Feature Types identifies the type of each feature in your data automatically, and you can override its choices. The Fit tool outputs a model and is described as the final tool in a pipeline, and the Predict tool applies a pipeline you have built to new data. The tool pages describe where these sit in a workflow, but they do not tell you whether a given model is appropriate for your purpose. That judgement stays with you.

For wider context on where machine learning is applied across the sector, see our overview of machine learning in finance.

The text mining tools

Finance data is not only numbers. Expense descriptions, invoice lines, customer comments, audit notes and variance commentary are all free text. The documentation describes the text mining tools as follows:

• Text Pre-processing: cleans up text data.
• Text Classification: trains a model that categorises text from labelled examples.
• Zero-shot Text Classification: assigns scored categories to text based on a category list you define.
• Topic Modeling: identifies and categorises topics in a body of text.
• Sentiment Analysis: determines whether text reflects positive, negative or neutral sentiment.
• Named Entity Recognition, Part-of-Speech Tagger, Text Summary and Word Cloud: entity recognition, grammatical tagging, summarising and visualising text.

Text Classification in more detail

The Text Classification page says the training data must have a column with text and a column with the text's label, and both must be String data type. It offers an Auto Mode that searches Multinomial Naive Bayes and Linear SVC for the best algorithm, and it uses TF-IDF vectorisation. The page lists English, French, German, Italian, Portuguese and Spanish as supported languages, and notes that Linear SVC might be slow on a large training dataset. The outputs are a model, which you connect to the Predict tool for new text, and evaluation metrics.

Topic Modeling in more detail

According to the Topic Modeling page, the tool uses latent Dirichlet allocation (LDA). You choose how many topics to model and the type of report. The page suggests considering the Text Pre-processing tool upstream of Topic Modeling (check the current Topic Modeling page, as documentation changes between versions). The outputs include a new column for each topic, a report, and a model object for use with new data. The same six languages are listed.

Illustrative example: sorting free-text descriptions

This is an illustrative example with made-up numbers. Suppose a finance team has 2,000 expense lines whose descriptions were typed by hand, and a reviewer has already labelled 300 of them with a category. A Text Classification workflow would use those 300 labelled rows as training data, then apply the trained model to the remaining 1,700. Our suggestion is to check a sample of the model's categories by hand, keep the labelled sample separate from the rows used to judge accuracy, and record who approved the final approach. The documentation shows how the tools work; it does not say how accurate any model will be on your data.

Things to check before using it on finance data

• Data quality first. Poor labels and inconsistent descriptions will carry through to the output. Run Data Health or an equivalent check, and keep a record of what you changed.
• Personal data in free text. Free-text fields can contain names and other personal details. Our suggestion is to ask your data protection contact whether the text can be used for this purpose before you start.
• Human review. A classification or topic output is a draft. For anything that feeds a reported figure, define who reviews it and how exceptions are handled.
• Version and installation. The suite page says it is not included in the Designer installer, so it is installed separately, and you must install the matching Designer version first. It also says your computer must meet the same system requirements as Designer.
• Compatibility. The page states that workflows built with the suite's tools in Designer 25.2 or earlier are not compatible with Designer 26.1. If your team has older workflows, check before upgrading.
• Access. The documentation says access depends on your edition and user type, and that administrators can restrict access by role. Confirm with your administrator what is enabled for you rather than assuming.

A caution about the documentation itself

On the Alteryx Help pages used for this post, the "Publication date" field was blank, and the version information was inconsistent: the breadcrumb showed 2026.2 while the version selector marked 2024.1 as current. Menus, tool names and availability change by version and licence, so check the current Alteryx documentation for the version you actually run, and do not rely on a summary like this one for configuration details.

Learnsignal's offices are off Baggot Street in Dublin, and we offer in-person training from there. For current details, see our CPD pages.

Frequently asked questions

Is Alteryx Intelligence Suite part of the standard Designer installation?

No. The Alteryx Help page says the suite is not included in the Designer installer and is installed separately, with the matching Designer version installed first.

Which languages does the text mining documentation list?

The Text Classification and Topic Modeling pages list English, French, German, Italian, Portuguese and Spanish. Check each tool's page for its own language notes.

Does Data Health fix my data?

The documentation describes it as a tool that analyses missing values, outliers and sparsity and reports scores and recommendations. Deciding what to change, and documenting that decision, remains a finance team responsibility.

Can I rely on a model's output for reporting?

The documentation explains how the tools work, not whether a result is fit for a particular reporting purpose. Our suggestion is to test on a held-back sample, review exceptions manually, and keep a record of the checks.

The Alteryx documentation is the authority on what each tool does in your version, so use it alongside your own data quality checks.