Recommendations are Nagata's core output. Each one is a specific, evidence-backed action your team can take to reduce waste, consolidate tools, or make a smarter buying decision.
How they're generated
Click Generate recommendations on the Recommendations page. Nagata runs a set of rules across your data and produces a prioritized list.
Rules are fully deterministic — no AI model decides whether to flag something. Every recommendation shows exactly which data it used and what formula it applied. Re-generating is safe: Nagata updates existing recommendations rather than creating duplicates.
The seven rules
1. Ownerless high-spend vendor
An active vendor is spending $5,000 or more per year and has no owner assigned.
2. Inactive vendor with a renewal
A vendor is marked as inactive but still has an active contract with an upcoming renewal.
3. Duplicate tools
You have multiple active vendors in the same category (e.g., two analytics platforms).
4. AI spend growth
Month-over-month AI API spend grew by 50%+ and the dollar increase is at least $200.
5. AI duplicate provider
You're paying material amounts on multiple models from the same provider.
6. AI expensive workflow
A single LLM workflow is consuming more than 40% of your total monthly AI spend.
7. AI high user
A single user is responsible for more than 50% of monthly AI API spend.
What each recommendation shows
| Field | What it means |
|---|---|
| Type | Keep, cut, downgrade, consolidate, renegotiate, review, or replace with agent |
| Priority | Low, medium, high, or urgent — scales with the dollar amount at stake |
| Reason | A plain-English explanation of why the rule fired |
| Supporting data | The exact inputs the rule used and the formula it applied |
| Estimated savings | Projected dollar impact (null when there isn't enough data) |
| Confidence | 0–100% likelihood this recommendation is actionable |
Filtering
On the Recommendations page, filter by status, priority, type, or owner. Start with Urgent and High priority to find the highest-dollar-impact actions first.