Classification
Which category does it belong to?
Assigns a case to known categories, such as a document type or image class.
Machine learning · Concepts and applications
What is machine learning and how is it used?
A clear explanation of how a model learns from examples, makes predictions and can help with specific tasks.
Learning from examples
A model looks for patterns in data.
Instead of writing a rule for every situation, we provide examples with known information and outcomes. The model finds relationships in that data and uses them to estimate what could happen in a new case.
From example to decision
The work involves more than training.
Each step helps check whether the approach answers the question and whether its result can be used responsibly.
- 01
Define the question
Agree on the outcome to estimate and the decision it could support.
- 02
Prepare the data
Review data quality, coverage, permissions, labels and relevance to the task.
- 03
Train and evaluate
Learn patterns from examples and measure performance with separate data.
- 04
Test in context
Review errors, limitations and integration before using the model in operations.
Three common approaches
The question helps determine the approach.
The goal defines what is predicted and how it is evaluated.
Regression
What value might it take?
Estimates a numeric value, such as demand or time, from available signals.
Clustering
Which cases are similar?
Groups data by similarity without starting with predefined categories.
Examples by sector
One idea can support different tasks.
These diagrams are illustrative: they show types of problems, not T2G results or specific projects.
What type of document is this?
A classifier can sort documents to help a person review them.
Which record should be reviewed?
A system can flag unusual transactions against historical patterns to support a review.
Which signals appear in the image?
Computer vision can analyse visible features to support expert classification.
What demand should we expect?
A model can estimate future ranges from historical data and relevant conditions.
Ideas to explore
Where could ML help your business?
Nine starting points for a conversation with T2G. These are ideas: feasibility depends on available data, the process and evaluation against the current way of working.
01
Accounting businesses
Suggest expense categories
Propose categories from previously classified transactions for the team to confirm or correct.
Support reconciliation
Rank possible matches between invoices and transactions when references are incomplete.
Prioritise unusual records
Highlight uncommon amounts or combinations to guide a quality review.
Data to start withHistorical transactions, documents and team corrections with consistent criteria.
People keep responsibility for judgementConfirm classifications and matches. An anomaly alone does not prove an error or fraud.
02
Legal teams and firms
Organise documents
Suggest document or matter types to support routing and review.
Find similar files
Retrieve internal records by content similarity to help people find relevant information.
Estimate workload
Explore ranges of hours or task volumes using previous matters with comparable characteristics.
Data to start withAuthorised documents, reviewed categories and task histories with context.
People keep responsibility for judgementVerify documents and sources. Similarity does not establish legal relevance or replace a lawyer’s judgement.
03
Finance teams
Explore cash-flow scenarios
Estimate ranges of receipts and payments from historical records to complement planning.
Review unusual transactions
Flag changes from usual patterns and prioritise alerts for the responsible team.
Classify operational requests
Suggest a team or request type for recurring enquiries, with review for uncertain cases.
Data to start withTime series of receipts and payments, transactions and dated requests with recorded outcomes.
People keep responsibility for judgementCompare scenarios and review alerts. These ideas support operations, rather than automated investment or credit decisions.
When to explore it
Start with a task you can evaluate.
Machine learning may be relevant when there is a specific question and data related to it.
- There are enough representative examples to analyse.
- The expected result can be checked against clear criteria.
- The team can assess errors and agree how to respond.
Next step
Have a task you would like to assess?
We can review the question, data and criteria to decide whether machine learning is an appropriate approach.
Book a diagnostic session directly with the partners