TRAJECTORY

Making change
steerable.

Intervention Intelligence

Test the decision before you commit. Hold it to its prediction after.

See a recorded case
01 · The problem

Everyone holds a different piece.

Operations sees
capacity
Finance sees
cost
Leadership sees
execution
Trajectory sees
the chain
02 · How it works

One decision.
Start to close.

  1. 01
    Sense
    What we observe
  2. 02
    Understanding
    What might explain it
  3. 03
    Impact
    What is at stake
  4. 04
    Scenarios
    What if we move a lever
  5. 05
    Prescribe
    What to do about it
  6. LOCKED
    ·
    Signed
    Baseline, prediction and limits frozen
  7. 06
    Steering
    Readings on the path
  8. 07
    Close
    Scored against what was signed
  9. 08
    Learn
    Feeds the next decision
Back to Sense: each closed decision sharpens the next
03 · Understand · a recorded case, 12 Nov 2026

From correlation
to causality.

Kestrel Foods · illustrative company

15 measures from the case. Every pair tested for a link.

  1. 105
    Pairs of measures tested
  2. 47
    Survive false-discovery control
  3. 14
    Hold with everything else controlled for
  4. 4
    Drive on-time-in-full
How the narrowing happens¹
Timing
The cause has to lead the effect.
Third factors
A link another measure explains is set aside: 19 here.
Chance
Corrected for all 105 pairs tested at once.
The 4Forecast bias · Order pick accuracy · Expediting spend · Promo depth
04 · Steer · same case

On-time-in-full to 96%.

The goal: orders delivered complete and on time, 92.5% to 96% in 12 months. The move: central forecasting with dynamic safety stock, signed 9 Jun 2026.

On track, or off track.

Early sign
Forecast bias · due month 0Moved as expected
Result
On-time-in-full · due month 2Inside doing nothing's range: not yet credited
Readings
Every 14 days
Tripwire
Not beyond doing nothing in timeRe-evaluate the move

Drift caught in weeks, not reporting cycles.

Reporting cycle
Found at the next report
month 3 to 12
Steering
Found at the next reading
day 14
Correct course sooner: less time and money spent on a move that isn't working.
Signed move vs doing nothingOn-time-in-full % / month
DOING NOTHING · 93.15%SIGNED PATH · 96%EXPECTED Δ
Readings, months 1, 2, 3, 4, 5Doing nothing, 90% rangeInside the range: not yet credited to the move
05 · Rules

Built to
refuse.

Each rule pinned by an automated test.
i.

Every stage can refuse

Too little evidence: it stops and names what is missing.
ii.

Causality gates value¹

A figure claims no more than its evidence grade.
iii.

Your data has the last word¹

Every proposed link is tested against your own series.
iv.

The prediction is locked

Scored against what was signed, not what was hoped later.
v.

Limits are not traded

A move that crosses a declared limit is withheld, reason shown.
vi.

AI writes, code computes

Every number, graph and refusal is computed. An AI failure never changes one.
Refused · a lever path the data cannot settle
Withheld · a move that crosses a declared limit
Held · a theme raised by 4 people, not 5
7,900+ automated tests·Each refusal pinned·Versioned system specification·Append-only readings·Each organisation sees only its own data
06 · People

Honest perspectives.
Protected identities.

1:1
Individual invitations, AI follow-up questions
0
Transcripts or quotes stored
5+
People before a theme is reported
0
Small groups named · disagreement still shown
07 · Team
Daniela de Paiva
Founder & CEO

Organisational strategy, impact measurement and systems innovation, across international development and business.

Stuart Wilkins
Co-founder & COO

30+ years in financial services, risk, governance and transformation, including Lloyds Banking Group and Nedbank.

Anuradha P
Technology & Operations

Computer engineer and MBA. 15+ years across data systems, business analysis, finance and operations.

¹ Methods: lagged Granger tests with false-discovery control; a structural equation model fitted on the links that survive; double machine learning where the data allows; synthetic controls for site-level effects; Monte Carlo forecasts against doing nothing. Every estimate carries its evidence grade. Estimates, not guarantees.