Bricks
Typed semantic checks for conversations
A Brick is one reusable question about a conversation. It returns a typed value, confidence, reason, and evidence instead of another paragraph to read.
Definition
next_step_confirmed
Return true when the buyer and seller agree on a specific next action with an owner or date.
Output
{
"value": true,
"confidence": 0.91,
"reason": "Legal review agreed",
"evidence": "Legal will review Friday"
}
Definition
A Brick is not a prompt
Prompts are useful for one-off questions. Bricks are building blocks: versioned checks that make the same check safe to run across every call, email, chat, and transcript.
One concept
Each Brick evaluates a single business signal, such as whether a next step was agreed or whether budget was confirmed.
One typed result
The value comes back as a boolean, score, category, or string list, with the same field shape every time.
Evidence attached
A Brick can return the exact transcript span that supported the answer, so a person can audit the signal.
Reusable logic
Build a Brick once, then reuse it in discovery, forecast risk, QA, coaching, or data workflows.
Versioned safely
Change a rubric by creating a new version, while historical runs stay tied to the logic that produced them.
Ready for CRM and BI
The output is designed for CRM fields, warehouse tables, BI models, and workflow automation.
How it works
Define the check. Run it everywhere.
A Brick turns repeatable judgement into a clear contract. The contract names the signal, defines the output type, sets the rubric, and returns evidence with the result.
Name the signal
Start with one question that matters to the workflow, such as next step confirmed, pricing objection raised, or stakeholder identified.
Set the type
Choose the shape downstream systems should expect. Keep the output small enough to map into a column, field, or trigger.
Attach the rubric
Describe what should count, what should not count, and which evidence needs to be present before the Brick returns a value.
Reuse in Kits
Add the Brick to any Kit that needs the same signal. The same check can support forecasting, coaching, QA, and reporting.
Output types
Small outputs. Useful everywhere.
Bricks avoid long write-ups. They return the smallest reliable value a system can consume.
Boolean
Use true or false when the workflow only needs to know whether a signal is present.
Score
Use a numeric score when the signal has quality, completeness, or strength.
Category
Use a fixed category when the result must fit a known set of labels.
String list
Use a list when a conversation can contain multiple names, products, issues, or competitors.
Evidence-backed value
Use evidence when a result may need review, audit, or a human-facing explanation.
Grounded check
Use a grounded Brick when the answer depends on your pricing, policy, product, or process documentation.
Why Bricks
Stop measuring calls as one score.
A single score hides the signals teams actually need. Bricks keep each signal separate so it can be trusted, trended, and routed.
Traditional call scoring
Semarize Bricks
Controls
Built for repeatable evaluation
Bricks are designed so teams can change evaluation logic deliberately without breaking historical reporting or downstream integrations.
Independent versions
Update one Brick without changing every Kit that has ever used it. New runs can use new logic while old runs remain reproducible.
Rubric control
Define what counts, what does not count, and what evidence is required before a value should be returned.
Kit composition
Group Bricks into Kits when you need a complete rubric, scorecard, risk screen, or evaluation framework.
Ready to build
One check used across every workflow.
Start with the signal your CRM, dashboard, or workflow is missing, then turn it into a Brick.
No card required. Start testing in minutes.
Questions
Frequently asked questions
Short answers about what Bricks are, how they differ from prompts, and how they fit into the Semarize product model.
From the blog
Further reading
Practical pieces on stable evaluation contracts, typed rubrics, and why Bricks are safer than freeform prompt scoring.