AI does the first review. Humans make the call.
Sparqbox uses Claude to read every submitted idea against your weighted criteria and produce a structured first score. Coordinators review the AI's reasoning, score independently, and the weighted total fires the decision. AI accelerates the work — it never owns the verdict.
Four steps. Same scoring rubric for every idea.
1. Configure criteria
Set 4–8 weighted criteria per category. Customer impact, effort, strategic fit — whatever the bucket needs.
2. AI reads every submission
Claude scores the idea against each criterion with a short justification. The output is a structured first review, not a paragraph.
3. Humans score independently
Coordinators see the AI's score per criterion, can agree, override, or ask the submitter for more. Human scores are what count.
4. Weighted total fires the decision
Approve, reject, or needs-discussion based on per-category thresholds. The submitter gets feedback within 48 hours, every time.
AI recommends. People decide.
Sparqbox uses AI as a first reviewer. It reads every idea, scores it against your criteria, and gives reviewers a fast, consistent starting point. What it does not do is decide. It never sets a status, assigns a reviewer, or closes an idea.
We hold this line on purpose. As AI takes over more of the execution, the scarce and valuable part of any decision becomes the person accountable for it. Sparqbox keeps that person in control. The AI scales the work. The human owns the judgement.
It is also why bias detection in scoring is on our roadmap: surfacing where a score may be skewed so the people deciding can see it. The decision stays human.
Your ideas stay your ideas.
We pay Anthropic for inference under their zero-data-retention policy. Your idea content is processed in the EU, then discarded. Nothing is stored at the model provider, nothing is used for training, and nothing is shared between tenants.
EU inference
All AI calls route through EU regions. No data crosses the Atlantic.
Zero training
Anthropic's zero-data-retention policy applies — your ideas never train a model.
Tenant isolation
AI calls are scoped per-company. No cross-tenant context, ever.
Quick answers.
Can AI evaluate ideas fairly?
AI can score ideas consistently against your criteria, which removes some human inconsistency. In Sparqbox the AI only recommends; people decide, so the judgement stays accountable.
Why does Sparqbox not let AI make the decision?
Because a decision needs someone accountable for it. AI handles the first-pass scoring at scale. The human verifies and decides. That is the line we will not cross.
Which model does Sparqbox use?
Claude (Anthropic) for the first review. We choose the right model per task — fast and cheap for simple scoring, larger for nuanced ideas. The choice is invisible to you.
Can I turn AI scoring off?
Yes. Some categories work better with humans-only review. You can disable AI per category in settings — the rest of the workflow (criteria, weighted scoring, mandatory feedback) stays the same.
How accurate is the AI?
Accurate enough to save coordinators 30–50% of triage time. Not accurate enough to remove the human reviewer. That's by design — the weighted total only fires on human scores.
What happens if the AI gets it wrong?
Coordinators override with their own scores. The AI's score is shown for context but never counts. Mistakes are visible, easy to correct, and don't propagate.
Every idea deserves an answer.
Give your team the one thing a suggestion box never will: a real decision, every time.
