Deployment Value Review

A deployment-value review

Discount the ROI to what transfers.
Gate each worker on its data.
Sequence so each deployment compounds the next.

Paste a logistics account brief. The tool discounts HappyRobot's published ROI multiples to what actually transfers for that account, with a named haircut on each one, gates every AI worker on the data foundation it needs, and sequences the rollout so each deployment lands on the data the last one produced.

Live engine

Pick a backtest, edit the brief, run the review.

Pick a backtest or paste your own brief, then Run review for a live extraction by gpt-5.4.

Method

What the model does, what the look-up adds, what the engine does

Four stages, left to right. The model only reads the published figure; the engine owns every haircut and every sequencing call.

01

Model · extract

Reads the messy brief and pulls out findings — data foundations and AI-worker deployments, each with the figure HappyRobot published and how to read it. It never discounts.

02

Look-up · prior rollouts

Checks each deployment against prior HappyRobot rollouts. Confirms a declared sequencing dependency, or surfaces one the brief never mentioned.

03

Engine · discount + gate

Pressure-tests each published figure against the account's call mix and stated volume, then gates every worker on the data it needs and sequences the rollout.

04

Output · plan

The discounted figures up top, then a three-phase rollout. Foundations first, deployable workers in parallel, gated ones after. Deterministic from the same findings.

Extraction, ROI transfer, and readiness stay separate. The model extracts; the look-up grounds it in prior rollouts; the engine discounts the figures and sequences the plan. That logic lives in engine.py, the records in corpus.py, both covered by evals.

Framework

Five ROI rules. One readiness gate.

HappyRobot publishes rosy multiples. The engine reads each one against this account and either holds it, haircuts it, or restates it — with a named, auditable reason. Then a separate gate sequences the rollout.

R1
A recovery ratio is not a return multiple
A ">100x" on collections is dollars recovered over cost — restated against the real AP backlog, never headlined as a multiple.
reclassify
R3
A cycle-time win is speed, not ROI
"A week to under 30 minutes" is a speed and coverage gain — counted as such, never expressed as a return.
reclassify
R2
The ~5x is documented on inbound
Carrier-sales' ~5x came from inbound carrier sales. On an outbound-heavy account the unit economics differ — the figure is flagged rather than applied at full strength.
haircut
R4
No stated volume, no sizing
A return or throughput figure with no call/load volume stated can't be sized here — shown as HappyRobot's headline until it's sized to this account.
haircut
R5
Anchored and on-shape — it holds
A figure tied to the account's stated volume and matching the documented deployment shape is defensible. The engine passes it — a number that survives the account is one worth carrying.
holds

Then every worker is gated

  • Deploy now. Its data foundation is in place. Ship.
  • Fix foundation first. A sound deployment, sequenced behind the data it lands on.
  • Re-scope. A useful play exists nearby; the named one isn't it.
  • Don't pursue. Wrong-shaped, or value too ambiguous to defend.

Merit, ROI transfer, and readiness are three separate calls — so a figure can hold while its deployment still waits.

Future State

How this runs in production

The live demo implements the three highlighted steps; the rest is the automated pipeline around them — same boxed language, LIVE badges, a diamond for the human gate.

Future State

Account calls CCaaS · recordings → transcription
Ops stack & volumes TMS · telephony · load & rate feeds · carrier master
Canonical record one account record · backend
PII / commercial-confidential redaction privacy by design — before any model call
LIVE LLM extraction brief → foundations + published figures
LIVE Prior-rollout retrieval RAG in prod · fixture-backed here
Prior-rollouts library the look-up layer
LIVE ROI + readiness engine discount · gate · sequence — deterministic core
LIVE Phased rollout foundations first · 3 phases
Business case defensible figures · drafted from the holds
Human review HITL · strategist + account approve
Into existing rituals Slides · account plan · rollout backlog
approved
shipped rollouts feed the library
Live in this demo Production pipeline around it

Account calls are transcribed and the ops stack and volumes are pulled in over connectors; everything lands in one account record, PII is redacted before any model call. Then the three live steps run — the model extracts foundations and published figures, the look-up grounds them against prior rollouts, and the deterministic engine discounts each figure and sequences the plan. That fans out into the phased rollout and a business case built only from the figures that held, a person approves before anything ships, and the shipped rollout feeds back into the library — so the look-up gets smarter each time.