Grounded AI, not guesses.
Every figure Dexor speaks is derived from your real data — the AI narrates, it never invents. It sees only the subtree you're entitled to see. And the deterministic baseline is free: no model, no latency, no per-use cost.
Intelligence before you spend a single credit.
Ranked at-risk flags and plain-language findings — critical below 60% attainment, warning to 85%, ahead at 100%+ — plus forecast-driven risk and statistical anomaly detection, all computed deterministically from your team-performance tree. Immediate, defensible, and free. The AI layer is optional and sits on top.
- Attainment < 60% · Lahore-Northcritical
- 60–85% · Karachi-Eastwarning
- ≥ 100% · Faisalabadahead
No model · no latency · no per-use cost.
Ask your team anything. In plain language.
“How is Ayesha’s territory tracking against target?” Ask Dexor answers synchronously and streams as it thinks — gathering its own data through internal tools (dashboard snapshot, forecast outlook, org roster), looping until it can answer, then grounding every figure. Greetings route to a fast, cheap path. History is kept for 90 days. And the caller can only ever see their own subtree — structurally.
Precise questions. Auditable answers. Delivered to your inbox.
A structured ask — scope, subject, period, optional metric — is validated, authorized to your subtree, queued, and answered into your inbox. The figures are the deterministic summary scalars, verbatim; the model only writes the prose around them.
Find the finding.
Hybrid keyword + semantic search across your team’s insights and per-rep sales summaries — no LLM in the loop, usable at near-zero cost. An optional RAG mode narrates an answer grounded only on the retrieved hits, which are always returned alongside for transparency. Off by default, so there’s never a surprise cost.
- Insight · Karachi-East coverage -12%hit
- Sales summary · Sana M. · week 26hit
Keyword + semantic · RAG optional · off by default.
The weekly narrative, already written.
One grounded, 2–4 sentence executive summary per manager, generated on a weekly schedule and waiting in the Insights view — no live model call on read. Managers with nothing to report are skipped; idle tenants burn nothing.
Attainment holding at 82% but momentum turned negative (-6%). Lahore-North is the drag at 58% with rising no-shows; Faisalabad is ahead of target. Priority: coach Ahmed R. on coverage this week.
No live model call on read · idle tenants burn nothing.
Alerts that name the problem, the reason, and the fix.
A statistical anomaly detector surfaces outliers as insight items. On top, the notice engine decides if something matters, what to say — in a strict Problem / Reason / Recommendation format, grounded in real figures — and who to notify, targeting exact people on the roster. Deduplicated over 24 hours, pushed live, dismissible. Signal, not spam.
- Problem — Lahore-North attainment dropped to 58%
- Reason — coverage -18% MoM, no-shows up to 4
- Recommend — re-plan Ahmed R.’s week around missed doctors
Deduped over 24h · pushed live · dismissible.
Right-sized models. Built to stay up.
Each tenant selects from a curated allowlist — an efficient default for everyday questions, and a stronger option where depth matters. Inference runs on a primary provider with automatic failover to the direct Claude API on outage or throttle — covering the full agentic loop, not just one-shot answers. Streaming and prompt caching keep latency and cost down.
- Efficient defaulteveryday questionsselected
- Stronger modeldeeper analysisoptional
- Primary throttled → direct Claude APIfull agentic loopfailover
AI that can't blow your budget.
You buy AI as a predictable pool of prepaid Compute Credits — never an open-ended metered bill. Dexor manages the cost controls behind the scenes to keep every tenant margin-safe, and a live meter always shows how much of your pool each feature has used, so there are no surprises.
Targets & forecasting
The deterministic team-performance tree the AI narrates — attainment, coverage, and the quarter verdict.
Targets & forecastingPlatform & admin
Per-tenant AI configuration, seat governance, and the tamper-evident audit trail behind the intelligence layer.
Platform & admin