Skip to main content
For data teams

The questions your backlog never gets to

Your queue of “can you look at X?” is forty deep. DecisionBox works through the exploratory analysis unattended, shows every step it took, and brings the business the findings nobody thought to ask for.

CriticalHidden in the detail

Network-wide text messages jumped 10.2× and calls 23.4× in a ten-minute window at New Year’s midnight, invisible in hourly reports

Peak of 360,131 messages at 23:10
Telecom dataset · 10,000 cellsConfirmed

A finding from a real discovery run. Nobody asked for it.

What it answers

Five questions that sit at the bottom of the queue

DecisionBox goes after questions like these on its own, run after run, without anyone writing them down.

  1. 01

    What patterns exist across two tables nobody has joined yet, and are any strong enough to build on?

    It explores the combinations your team never had time to model.

  2. 02

    What changed in last week’s data that nobody has asked about yet?

    New movements flagged before the business notices them.

  3. 03

    Which of our curated segments have quietly broken: stale definitions, inactive members, missing links?

    A health check on the groups everyone relies on.

  4. 04

    At what level of detail do the interesting patterns first appear?

    It looks below the daily and hourly views your reports stop at.

  5. 05

    Which findings worth flagging to the business exist in our data, without anyone asking?

    Given a playbook, it brings the business what matters, ranked.

From real runs

From a finding to a plan

Every finding that holds up comes with a recommendation: who it targets, what it is worth and the steps to get there.

HighMirror effect

Over the holidays the city lost 51% of its internet traffic while ski-resort areas gained 109%: an exact mirror

City −51% · ski resorts +109%
Telecom dataset · 3,131 cellsConfirmed
P1Recommendation

Add temporary capacity at 646 ski-resort cells for the winter holidays, sized to the surge

Expected impact60–80% fewer congestion events
Target646 cells
5 numbered action steps
HighWeekend pattern

Voice falls 40.4% at weekends while internet falls only 12%. Business-district capacity sits idle

Voice −40.4% · internet −12%
Telecom dataset · 10,000 cellsConfirmed
P3Recommendation

Shift weekend voice capacity to data in business-district cells

Expected impact+15–25% weekend data throughput
Target2,119 cells
5 numbered action steps
HighConcentration

The busiest 10% of cells carry 48% of all traffic, in the city and in the countryside alike

48% of traffic on 10% of cells
Telecom dataset · 63 daysConfirmed
P1Recommendation

Split or reinforce the nine central cells running at their limit

Expected impact+30–40% capacity headroom
Target9 central cells
5 numbered action steps

Every pair is from a real DecisionBox run. Source datasets are anonymised.

The honest list

What it does, and what it doesn't

It does
  • Reads the curated tables your team already builds. It doesn’t replace your modelling.
  • Shows every step and every query, so any finding can be checked and repeated.
  • Runs in your environment, with your credentials and your choice of AI.
It doesn't
  • Replace your modelling layer. It reads what your team builds.
  • Act as a data-quality monitor. Use a tool built for that job.
  • Find much in raw, uncurated logs. Point it at curated tables for signal.
Videos

See DecisionBox at work

All videos
Product tourThe AI analyst that finds what nobody askedA two-minute tour: discovery, validation, Ask and reports on a real marketplace dataset.
DecisionBox for business · Episode 1How to read a finding, and how much to trust itThe numbers, the validation verdict and the evidence behind a finding, before you share it.
DecisionBox for business · Episode 4From finding to decision: read a recommendationThe expected impact, the steps, the findings behind it and its validation.

Your data has all the answers.
DecisionBox asks the questions.

Let's run it on your data.