Skip to Content
by
Sign In
The FPT tower at night
GGround floor. Every assessment starts here, with what the customer sends.
FleziLift · Evidence-first migration assessment

Map it first.Then migrate.

Every migration starts with a pile of exports and documents that never agree. FleziLift turns it into one AS-IS baseline where every number has a source and every gap becomes a question.

  • Cloud Tools Measure
  • AI Cross-Checks
  • Experts Decide
Sign In to FleziLift Take the Lift
The Lift

Four levels of confidence.
One ride up.

Most first proposals start at L1. That's fine, if the report says so. FleziLift grades every record by the strongest source behind it, and shows the way up.

  1. 1Share CMDB & exports
  2. 2Run Azure Migrate, AWS Transform or Migration Center discovery (free)
  3. 3Validate with app owners

Accuracy bands adapted from AWS Prescriptive Guidance fidelity examples; the L0–L3 mapping is FleziLift's and indicative.

L0Desk
≤ 25–40%
RFP & documents only
L1Inventory
~ 50%
CMDB & spreadsheets
L2Measured
75–80%
Discovery tool, 30+ days
L3Validated
~ 95%
Confirmed by app owners
indicative accuracy of the AS-IS
L0 Desk · Documents Only

Every deal arrives as a pile, and the pile never agrees with itself.

Discovery exports, tool reports, CMDB extracts, RFPs and workshop notes: the same estate, a different total in every file.

Illustrative · Sample Data
Discovery exportRVTools · vCenterServers in scope186Mar
CMDB exportcmdb_export.xlsxServers in scope203Jul, last year
RFPRFP_v3.pdfServers in scope~190no date
App registerapp_register.xlsxApplications42Aug
Workshop notesworkshop_notes.docxApplications39Sep
Architecture docarchitecture.pdfDatabases282023
203 ≠ 186~190, undated42 ≠ 39 appsnames differ
  1. 01Counts DisagreeThe same estate gives a different total in each file.
  2. 02Different DatesExports taken months apart, rarely labelled.
  3. 03Names Don't MatchHostname, VM name and CMDB record all differ.
  4. 04Size, Not UsageAllocated vCPU and RAM, no utilisation history.
  5. 05No App, No OwnerServers nobody can map to an application or owner.
  6. 06Late SurprisesIntegrations, licences and shared DBs surface late.

Today an SA fixes this by hand, file by file: slow, and easy to pick the wrong number.

L1 Inventory · CMDB & Spreadsheets

When sources disagree, a rule decides. A person signs.

The priority order is set before the first file arrives, strongest source first. Every decision is logged and becomes an appendix of the report.

  1. 1Customer rule
  2. 2Measured export
  3. 3Row-level inventory
  4. 4Summary document
  5. 5Workshop note
Cross-ChecksIllustrative · Sample Data
MetricSource ASource BSource CWe UseRule AppliedStatus
Servers in scope203CMDB export186RVTools · vCenter~190RFP text186Measured export beats CMDB and proseSigned by SA
Total vCPU1,120RVTools · March1,096Tool report · January—1,120Newest snapshot of the same estate winsSigned by SA
Applications42App register57CMDB39Workshop notes42Customer's own register governs; 15 CMDB entries to confirmAssumption A-07
Databases24CMDB31Discovery scan28Architecture docOpenScan may count dev copies — ask the DBAQuestion Q-12
✓An SA signs every decision. The log becomes an appendix of the report.
?No source, no number. An open conflict turns into a question with an owner.
L2 Measured · Discovery Tools

Cloud tools already measure. We cross-check.

AWS, Microsoft and Google give their assessment tools away. FleziLift never re-measures what they give us: it reconciles what they measure with everything else the customer sends.

AWS Transformassessment from RVTools, CMDBMigration Evaluatoragentless collector, TCOAzure Migrateappliance, CSV or RVTools importDr Migrateapp-to-server mapping, TCOMigration Centerinventory, TCO, Gemini quick assessmentRVTools · vCenterVMs, hosts, disksAWS Transformassessment from RVTools, CMDBMigration Evaluatoragentless collector, TCOAzure Migrateappliance, CSV or RVTools importDr Migrateapp-to-server mapping, TCOMigration Centerinventory, TCO, Gemini quick assessmentRVTools · vCenterVMs, hosts, disks
Cloud toolsmeasure · size · price
FleziLift AIread · match · draft
FPT expertsjudge · assume · sign
Defensible
AS-IS

Tools measure. AI cross-checks. Experts decide. No number reaches the customer unsigned.

  1. 1
    Cloud Tools Own MeasurementWe never re-measure what they give usTheir assessment is free — we use it
  2. 2
    AI Never Invents a NumberEvery value points to a file, row or pageNo source means a question, not a guess
  3. 3
    Experts Own Every AssumptionConflicts are settled by an SA, not by AISigned, dated, visible in the report
Where FleziLift Fits

A cross-check and evidence layer between what customers send and what cloud tools need.

  • Discovery ExportsRVTools · LiveOptics
  • Tool ReportsAzure Migrate · Dr Migrate
  • CMDB & Spreadsheetsoften months out of date
  • RFP & Architecture DocsPDF, Word, diagrams, pictures
  • Workshops & Q&Anotes, emails, answers
AIFleziLiftcross-check & evidence layer
  1. 1Normalizeone schema for every format
  2. 2Cross-checkmatch assets, surface conflicts
  3. 3Gradeconfidence from source fidelity
  4. 4Explaingaps, questions, assumptions

FPT experts review every conflict and sign the baseline.

  • Evidence-graded AS-IS workbook + report, every value sourced
  • Question Pack a ranked customer pack of 15, with defaults
  • Clean Import Files nextready for the cloud agents
L3 Validated · Confirmed by People

Four steps. Every gate signed by an expert.

The FleziLift agents draft each step from the customer's files. Nothing moves to the next step until an FPT solution architect has reviewed it and signed.

02

AS-IS Assessment

One canonical inventory, cross-checked and graded L0–L3.

8 agents · who does what, and why to trust it ↓
You Get

AS-IS Workbook · Report · HLA Diagram · Marked-up Customer Diagram · Question Pack

✓ SA settles conflicts and signs the baseline
AS-IS first.Every TO-BE decision stands on it. TO-BE stays advisory.
How the Agents Work

19 agents in five flows. Code owns the facts.

One flow of agents per step. They read the customer's files and a curated knowledge base, draft, and hand over to code that checks every number. A reviewer agent closes each flow; an FPT architect signs each gate.

StepAgentsWhat They DoWhy You Can Trust It
0Project Setup1 agent
Project ProfilerReadsA summary of every uploaded documentDoesDerives the customer, drivers, target cloud and constraintsGivesThe project profile the other agents work from

What they doProfiles the customer, drivers, target cloud and constraints from the uploaded files.

Why you can trust itProject-scoped

Reads only this project's files: code binds the retrieval to the project, not the agent. The profile is shown in the portal and every passage it used is cited.

1Scope & Questions4 agents
ScopingReadsSOW, contract and RFPDoesSets what is in, out and still ambiguous, anchored on the contractGivesScope draft and clarification itemsGap DiscoveryReadsThe scope, the project files and the common knowledge baseDoesFinds the decisions the files cannot settle yetGivesCustomer-specific questions, each with its sourceQuestion PrioritizerReadsThe whole question list and the scopeDoesRanks every question P1–P3 by what its answer changesGivesA customer pack of 15, each with a default and its riskGap ReviewerReadsThe question batchDoesChecks evidence, depth and relevance of every questionGivesApprove, or revise with reasons

What they doSets what is in, out and unclear from the SOW and RFP, turns what the files cannot settle into questions, and ranks them.

Why you can trust itCited, then signed

Each question names the file behind it, and a separate reviewer checks evidence and relevance. One agent ranks the whole list; code keeps the customer pack to 15, each with a default, and an SA approves before anything reaches the customer.

2AS-IS Assessment8 agents
Data NormalizerReadsThe customer's exports and documentsDoesExtracts servers, databases and applications into one schemaGivesThe normalised inventoryInventory AnalystReadsThe normalised inventoryDoesClassifies each workload by type and complexityGivesWorkload analysis and dependenciesSemantic EnricherReadsInventory and project filesDoesInfers tier, environment, business domain and clustersGivesAn overlay; it never changes a factCandidate ScorerReadsWorkloads and a readiness rubricDoesScores readiness, risk factors and evidence gapsGivesReadiness notes for reviewArchitecture ×3KB ExtractorReadsThe customer's architecture documentsDoesExtracts systems, zones and linksGivesArchitecture facts, each citedArchitectReadsThe facts and the inventoryDoesPlaces every system in its zone and tierGivesThe current-state layoutSynthesizerReadsThe layoutDoesNames the flows and writes the diagram's storyGivesThe high-level diagram, in draw.ioAssess ReviewerReadsThe whole AS-IS draftDoesChecks completeness, scope and qualityGivesApprove, or revise with reasons

What they doBuilds one inventory from every export, analyses and scores the workloads, and draws the current-state diagram.

Why you can trust itEvidence L0–L3

An SA confirms how each file is read. Code then completes the inventory from those files, grades every value L0–L3, reads the customer's network diagrams (pictures too, by OCR) against the server list and computes the key findings. When files disagree, a written priority decides; with no source, it becomes a question, not a number; an AI insight whose quote is not in its file is dropped. An SA signs.

3Migration Strategy4 agents
Dependency MapperReadsWorkload brief and project filesDoesMaps the links the AS-IS did not recordGivesDependency map, each link citedSignal AnalystReadsThe customer's documentsDoesReports what they say about each serverGivesSignals with file and locatorStrategy AdvisorReadsThe rule decisions and the signalsDoesExplains hard cases, proposes exceptions with a sourceGivesAn overlay on the rule resultStrategy ReviewerReadsExplanations, exceptions and signalsDoesChecks every claim for evidenceGivesApprove, or revise with reasons

What they doMaps the links the AS-IS missed, reports what the documents say about each server, and explains the 7R result.

Why you can trust itRules decide

A written 7R rule tree in code decides, and it follows the customer's own target design where there is one. An exception needs a reason and a cited source, or it is dropped. Checked against golden sets of past bids.

4Wave Plan2 agents
Wave PlannerReadsThe waves code built, and the project filesDoesExplains each wave and finds cited constraintsGivesRationale and go/no-go checksPlan ReviewerReadsNarratives, checks and movesDoesChecks them against the plan's dataGivesApprove, or revise with reasons

What they doExplains each wave, sets its go/no-go checks and finds the customer's blackout dates.

Why you can trust itPolicy in code

Code builds the waves from a written policy: move groups never split, prerequisites first, a pilot by AWS's criteria. The estimate is a work breakdown: every count comes from the facts and names its source; the rates per unit are FPT assumptions.

Hover or tap an agent to see what it reads, does and hands on.

  • KnowledgeEach project's files in an isolated space, and a curated, dated common base: AWS, Microsoft and Google methods, lifecycle dates, FPT practice.MCP · LightRAG · pgvector
  • Code ChecksThe evidence layer, the 7R and wave rules, and integrity checks: no number leaves unchecked, no fact is rewritten.Rules from a policy
  • PeopleA reviewer agent closes every flow; then an FPT architect signs each gate in the portal.Signed gates
  • PlatformOne gateway to the approved models, flows that resume at every gate, and a trace of every call and passage.LiteLLM · LangGraph · Postgres
How We Earn Trust

Trust is designed in, not prompted in.

Every output carries its evidence, and accuracy is measured every release.

Factsnumbers, configurationscomputed by code, never by the AI
Judgmentstier, criticality, app mappingAI with a rubric, checked against SAs
Recommendations7R, waves, sizingrules first, AI explains, people decide
  1. 1Groundedno source, no claim

    Only the customer's files and a curated, dated knowledge base · Every claim carries a quote and its source · “Unknown” is allowed, and becomes a question

    Customer sees: every number linked to its source row
  2. 2Verifiedcode checks the AI

    Code re-checks each quote and recomputes each number · The writer and the reviewer are separate · Two runs disagree → flagged for an SA

    Customer sees: the checks each section passed
  3. 3Measurednumbers, not adjectives

    Golden set: past bids with known answers · Field accuracy, citation support, conflict recall · Re-run on every model, prompt or KB change

    Customer sees: an evaluation card with every release
  4. 4Accountablea person signs

    A named SA signs conflicts and the baseline · Audit log: who, when, which model and prompt · Every item labelled AI-drafted or SA-verified

    Customer sees: a signed baseline and its history
  5. 5Protectedyour data stays yours

    Enterprise model endpoints, no training on your data · Approved region, access control, PII masking · ISO/IEC 42001 as the management target

    Customer sees: a clear data-handling statement
NIST AI RMF 1.0Govern · Map · Measure · Manage, 2023NIST AI 600-1GenAI profile: confabulation, 2024ISO/IEC 42001AI management system — our target, not a certificationNature, 2024semantic entropy detects confabulationAnthropic Guidancesay unknown, quote, cite, verifyRagasfaithfulness scored claim by claim
R Rooftop · The View

The AS-IS you can defend.

One workbook is the system of record. The report and the diagram are generated from it, and every value points to the file, row or page it came from.

AS-IS_Workbook.xlsxAI Draft · Pending Review
SummaryApplicationsServersDatabasesDependenciesSourcesCross-ChecksQuestions
HostnameOperating SystemvCPURAMLevelCheckSource
erp-app-01Windows Server 2012 R2832L1Verifiedcmdb.csv · row 2
erp-db-01Windows Server 2012 R216128L1Verifiedcmdb.csv · row 4
hr-app-01RHEL 7.9416L1Conflict, settledcmdb.csv · row 7
web-portal-01Windows Server 201648L0No source—
file-01Windows Server 2016416L2Verifiedrvtools.xlsx · vInfo 11
Every row: Source (file · row · page) and Level (L0–L3)Fictional Sample
Current-State Diagram · .drawio
AS-IS Report · Word
AS-IS Assessment1212/12L1
  • TodayAS-IS Workbook8 tabs, Source and Level on every row
  • TodayAS-IS ReportWord, ready for the proposal
  • TodayCurrent-State Diagramdraw.io, editable
  • TodayQuestion Packowners, due dates, default assumptions
  • NextCloud Import FilesAzure Migrate · AWS Transform · Migration Center
After the AS-IS

From a baseline you can defend to a plan you can run.

Once the AS-IS is signed, FleziLift proposes one of the 7 Rs for every server and turns the decisions into waves. Rules decide, the AI explains with sources, and an FPT architect approves. It follows AWS Prescriptive Guidance, Microsoft's Cloud Adoption Framework with Azure Migrate, and Google Cloud's migration guidance.

Keep scrolling: the panels slide past.
1 · How a Strategy Is Decided

Six gates. The first match decides.

AWS's default tree, applied by code to every server. The AI may propose an exception, only with a reason and a source, and every decision records the gate that made it.

  • Rehost
  • Replatform
  • Repurchase
  • Retain
  • Retire
  1. 1Customer stated a treatment?As Stated
  2. 2Retire signal?Retire
  3. 3Out of scope?Retain
  4. 4Cannot run on the target?Repurchase or Retain
  5. 5Target design, old OS or pattern?The Design's Service
  6. 6Otherwise: the postureRehost
2 · Readiness, Risk and Confidence

Can it move as it is, and how sure are we?

Rehost readiness · Azure Migrate's four states
  • ReadyNo finding
  • ConditionalFix first: out-of-support OS, iSCSI, a cited licence limit
  • Not ReadyCannot run on the target: AIX, mainframe, over the VM limits
  • UnknownOS or size missing: a question to the customer
Risk · probability × impact

Two drivers before a critical server rates High.

Confidence · the AS-IS evidence level
  1. L2–L3Highmeasured, or confirmed by the owner
  2. L1Mediuminventory or configuration export
  3. L0Lowdocuments only, or no source

Support dates come from a table in code, never from the AI.

3 · The Wave Plan

Foundation and a pilot first. Then waves that move together.

  • A move group is never split: a shared database moves together
  • Shared services go first, in Wave 0, instead of gluing everything
  • Dev and test move before production, each environment in its own waves
  • Gates from T-28 to a T-1 go/no-go, then hypercare
Wave 0 · FoundationDCs, appliances, shared services
Wave 1 · Pilot≤ 5 servers, one test estate
Early Waves≤ 10 servers each
Later Waves≤ 15 servers, up to 3 in parallel
  1. T-28 Scope Frozen
  2. T-21 Replicating
  3. T-14 Test Passed
  4. T-7 Runbook Approved
  5. T-1 Go/No-Go
  6. T-0 Cutover
  7. T+1–3 Hypercare
  • Golden SetsEvery rule change is replayed on anonymised past bids with known answers.
  • EvidenceEvery reason names its source; an exception without one is dropped and recorded.
  • Tree FitThe report shows how many servers follow the tree; below AWS's 70%, it says so.
FPT Know-How

Every number has a rule, and every rule has a source

The Handbook explains every tab, figure and column of the four steps; the Method walks through the strategy and the waves. Both stay inside FPT: sign in to read them.

HandbookWhat each part means, how FleziLift decides it, and what it rests on: AWS, Microsoft and Google guidance where they give it, FPT practice where they do not.Open the Handbook Method: Migration Strategy and Wave PlanThe decision tree for the 7 Rs, readiness, risk, move groups, the pilot and the cutover gates.Read the Method
You have reached the top

See what your customer
really has.

Sign in to open a project, upload what the customer sent, and let the agents draft the first evidence-graded AS-IS for your review.

Sign In to FleziLift