Skip to content
LUNTA

What you receive

The documents, not the description of them.

Every exit gate in our delivery system is defined by a named artifact. Four of them carry the weight of the whole programme, so rather than describing those four we publish their formats — contents, structure, and a worked extract of each.

No form, no gated download, no “request the sample pack”. If the format cannot survive being read by a stranger before they have spoken to us, it is not a format worth defending.

Formats — not client artifacts

Every figure, verdict, and finding on this page is illustrative. We do not publish client deliverables, including anonymised ones: anonymisation is rarely as complete as it looks, and a client who agreed to a case study did not agree to be re-identified by someone who knows the industry.

Artifacts 01, 02, and 04 follow one illustrative engagement end to end — a service-desk deflection programme — so you can watch a loop close on a number. 03 is a verdict memo from an unrelated engagement, in a different operation, that did not pass: a page showing only the case that succeeded would reproduce exactly the selection bias this method exists to remove.

  1. 01

    Required by the Diagnose exit gate

    See the phase

    Diagnostic report

    Ground truth before investment: where value actually sits in the operation, what the data can honestly support, and a ranked verdict on every candidate — including the ones we advise against. It is written to be usable by a firm that is not us.

    • Value map across the operation, with the measurement source for each figure
    • Data-reality assessment per candidate — what exists, what is retrievable, what is fiction
    • Ranked opportunity portfolio with a build / buy / wait / do-not-build verdict each
    • Effort, run-cost, and risk profile for the leading candidates
    • Written evaluation thresholds the pilot would have to clear, with their baselines
    • Executive readout: fifteen slides that survive being presented without us in the room
    Diagnostic report — §4 Ranked opportunity portfolio — extract · illustrative
    Tier-1 service desk deflectionBuildVolume, ticket structure, and resolution history support a measurable pilot. Baseline is clean and already instrumented.
    Contract review assistantWaitThe clause corpus is not retrievable to the standard this use case needs. Revisit after the records migration completes; re-scoring is a two-day exercise, not a new diagnosis.
    Demand forecast copilotDo not buildThe reported pain is a decision-rights gap, not an information gap. Forecasts are already accurate enough; they are overridden downstream. An AI layer here would automate the step that is not broken.
    Field engineer knowledge assistantBuyThree vendors already meet the requirement at a lower total cost than building. We name them and the evaluation we would run to choose between them.

  2. 02

    Annexed to the Pilot statement of work, signed before work starts

    See the phase

    Evaluation threshold schedule

    The single document that makes a gate a gate. Specific numbers, on specific measures, against a measured baseline, signed by both sides before a line of pilot code exists — so the threshold belongs to the programme rather than to whoever needs the result.

    • Measures, chosen so that clearing them means the operation changed
    • Pass thresholds, set against the baseline measured during Diagnose
    • The measurement method, including sample size and who adjudicates
    • Named signatories on both sides, and the date signed
    • The defined consequence of a miss — a written no-go, not a conversation
    Evaluation threshold schedule — Schedule 2 — Pilot evaluation thresholds · illustrative
    First-pass resolution on production tickets≥ 62%41% · measured in Diagnose
    Answer accuracy, adjudicated sample (n = 500)≥ 95%—
    P95 latency at operating volume≤ 4 s—
    Cost per resolved case at volume≤ $0.40$3.10 · human-only

  3. 03

    Required by the Pilot exit gate — issued either way

    See the phase

    Go / no-go verdict memo

    The honest no, as a document. What was tested, what the evidence showed, the result against every threshold, the recommendation, and — the part that gives it a shelf life — precisely what would have to change for the answer to change.

    • Scope of what was tested, and explicitly what was not
    • Result against each threshold in the signed schedule, pass or miss
    • Failure taxonomy: what the system got wrong, how often, and how badly
    • Recommendation, stated in one sentence in the first paragraph
    • What would have to change for the recommendation to change
    • Cost of the alternative, so stopping is compared against something
    Pilot
    Claims triage assistant · 6 weeks
    Recommendation
    No-go. Do not proceed to Scale.
    Correct triage tier, adjudicated sample (n = 400)
    71% — threshold ≥ 90% · miss
    Handler time saved, matched sample
    9% — threshold ≥ 35% · miss
    P95 latency
    2.6 s — threshold ≤ 6 s · pass
    Why
    Performance is bounded by retrieval, not by the model. In 61% of the misses, the evidence a correct tier depended on was never returned to the model at all; a stronger model does not fix a corpus the index cannot reach. The two passing measures are not evidence of a system that works — they are evidence that the parts we did not depend on behaved.
    What would change the answer
    Indexing the adjuster notes that currently exist only as scanned attachments, then re-running this evaluation unchanged. That is an ingestion change, not a new build. We would expect the high 80s and would not restart the pilot before the measurement clears 90%.
    Cost of the alternative
    Proceeding to Scale on the current evidence: 14–18 weeks of engineering against a capability that would still mis-tier three claims in ten, in a process where a miss is a regulatory and customer-harm exposure rather than an inconvenience.

  4. 04

    Required by the Full Value exit gate

    See the phase

    Value-realisation statement

    The programme closes on a number against the day-one baseline, with the measurement method signed by both sides — and with the things we are not claiming listed alongside the things we are.

    • Measured outcome against the baseline recorded in Diagnose
    • Measurement method, agreed and signed before the measurement was taken
    • Run-cost and ownership economics, so the number is net rather than gross
    • Adoption and operating-model status at close
    • What we are not claiming, and why
    • What we would do differently — written by us, about us
    Value-realisation statement — §1 Outcome against the Diagnose baseline — illustrative
    First-pass resolution, tier-1 tickets41%64%Same query, same ticket taxonomy, 12-week window
    Cost per resolved case$3.10$0.38Fully loaded variable cost per case; annual platform and model run cost deducted separately below
    Annualised net benefit—$1.9mVolume × unit-cost delta, less $0.4m annual run cost
    Headcount reduction—Not claimedRedeployment was the agreed outcome; no reduction was targeted or measured

Four rules every artifact follows

Written to be usable without us

Each document assumes the reader is your engineer, your auditor, or the next firm — not a member of our team with the context in their head.

Every figure names its source

A number in a LUNTA document carries how it was measured, over what window, with what sample. A figure without a method is an opinion with a decimal point.

The negative case is a deliverable

A no-go memo is written to the same standard as a go. It is the document that stops the same idea being re-piloted in eighteen months by a different team, with the same result and a fresh budget.

What is not claimed is stated

Every outcome document carries an explicit list of what it does not evidence. It is the shortest section and the one that makes the rest credible.

This page is formatted for print. Use your browser’s print or save-as-PDF to take it to a review — there is no form in front of it.

The first of these arrives two to four weeks after we start.

A diagnosis is fixed-scope and fixed-fee, and it stands on its own: you can act on it with us, with another firm, or not at all.

Start a diagnosisSee how engagements are contracted