Mereka x AVPN - client strategy dashboard

Decision-safe operating model for scale, cost, AI support, LTP delivery, and M&E.

A shareable view of the core AVPN assumptions: 300,000 registrations are programme-period scale, 5,000 concurrent sessions are a readiness ceiling, and the retainer works only with explicit usage caps and campaign approval rules.

Product frameGateway + M&E layer
Readiness ceiling5,000 sessions
Maintenance model$3,000/month
Operating ruleCapped usage
Recommendation Approve an elastic baseline, not permanent peak capacity.

Design and test for 5,000 sessions, but run ordinary months lean.

Evidence 300,000 is programme-period scale.

Monthly active usage and gateway minutes drive real concurrency.

Cost boundary $700 infra + $250 AI are capped assumptions.

Campaigns, grounding, and stronger models need explicit approval.

Decision ask Confirm caps, campaign windows, and LTP data rules.

These choices make the retainer governable through 2029.

Baseline scenario selected.

Decision-safe baseline

AVPN Strategy Dashboard Decision Pack

This print view expands every tab into one decision pack: executive readout, demand reality, cost assumptions, AI/language controls, LTP operations, M&E, delivery roadmap, risks, decisions, and source caveats.

Monthly registration pace
8,108
300,000 learners spread over 37 programme months.
Projected peak sessions
58
Far below the readiness ceiling.
Modelled cloud + AI cost
$509
Cloud + AI within planning assumptions.
Decision read
Lean baseline
Default operating mode for ordinary months.

Demand reality

Normal gateway use is materially lower than the 5,000-session engineering proof target.

Baseline-ready

Budget guardrail

Modelled cloud and AI against the $950 infra + AI envelope.

Infra + AI envelope used54%

Operating rules

A
Elastic baselineDefault month stays lean; campaign capacity is planned and time-bound.
B
Hard capsMonthly, per-session, and per-language AI limits prevent surprise spend.
C
Common schemasLTP uploads need templates, validation, queues, and change-order boundaries.

Client-ready position

The platform should be sized around an elastic baseline and load-tested against the 5,000-session readiness target. Campaign peaks and higher AI usage should be pre-approved through clear caps, while monthly dashboards make cost, language quality, and LTP verification visible to AVPN.

Consultant read

The answer is not "more servers." It is a governed operating model with transparent assumptions, visible thresholds, and pre-agreed decisions.

Demand logic

The concurrency estimate is driven by active learners, gateway minutes, and concentration.

Programme target300,000

Registrations across the programme period.

Programme months37

Used for planning pace.

Monthly pace8,108

Not the same as simultaneous users.

Current MAU25,000

Scenario-controlled monthly active learners.

Gateway time10 min

Time spent in the AVPN gateway, not external course time.

Peak sessions58

Computed scenario peak.

Formula

Average concurrent = MAU x gateway minutes / active minutes per month. Peak concurrent = average concurrent x peak concentration factor.

Scenario ladder

This is the mental model that should replace "300k equals 5k always online."

ScenarioPeak sessionsInterpretation
Normal operations~130Lean baseline planning.
Peak campaign month~1,406Pre-warm, monitor, and cap usage.
300k target spread month~2,500High but still below 5k.
Synchronized deadline surge~5,625Stress envelope, not retainer baseline.
Formal proof target5,000Load-test and document readiness.

What would change the answer

TriggerWhy it mattersCommercial response
All learners pushed through one synchronized deadlinePeak factor rises sharply.Temporary campaign capacity window.
Gateway becomes full LMS hostingSession time and storage increase.Separate architecture and cost model.
Always-on 5k reserve requestedIdle capacity becomes the cost driver.AVPN-funded guaranteed-capacity option.
External course providers return inconsistent dataVerification queues and support load rise.LTP schema governance and change-order rules.

Base cost assumptions and submitted budget stack

These lines must stay distinct in client conversations.

BucketAmountMeaning
Development$154,000 one-timeBuild phase, June to November 2026.
Maintenance retainer$3,000/monthOperating retainer across the maintenance period.
GCP infrastructure sub-line$700/monthBaseline cloud platform allowance.
Vertex/Gemini AI sub-line$250/monthCapped AI usage allowance, separate from dev-hours.
Other tool/API sub-lines$450/monthEmail, video testimonial, survey, translation, and related tools.
Commercial guardrail

The retainer covers agreed operations within approved assumptions. It is not unlimited cloud, AI, translation, email, video, survey, or feature scope.

Current scenario vs allowance

Only cloud and AI are modelled here; other tool/API lines remain separate.

Within envelope
Cloud model$433

Against $700/month infra line.

AI model$76

Against $250/month AI line.

Vendor subtotal$959

Cloud + AI + fixed tool/API lines.

Infra usage62%

Of $700 allowance.

AI usage30%

Of $250 allowance.

Maint. total$2,559

Labour + vendor/tool lines.

Cost assumption controls

These variables drive the AI line and reveal when a change order is needed.

Chatbot users7,500

MAU expected to use the assistant.

AI conversations11,250

Monthly text conversations priced by model.

Support turns67,500

Used for grounding and operational load.

Token AI cost$76

Input/output token spend before grounding.

Grounding cost$0

Search/grounding uplift after free allowance.

Maintenance bridge

Shows whether the selected scenario fits the $3,000/month retainer.

Retainer-safe
Monthly budget$3,000

Submitted maintenance retainer.

Monthly model$2,559

Calculated all-in maintenance line.

Monthly delta$441

Positive means remaining headroom.

32-month budget$96,000

Submitted maintenance total.

32-month model$81,888

If this scenario held every month.

Overage/month$0

Requires approval if above $0.

Unit cost assumptions

The visible formula layer behind the monthly model.

Formula visible
AssumptionValue usedWhy it matters
Chatbot usage volumeMAU x adoption x sessions/userConversations, not registrations, drive the text-token AI line.
Conversation token profile10,000 input tokens + 2,000 output tokensText-only planning assumption. Halve or double costs if real conversations are half or 2x this size.
Default volume routeGemini 3.1 Flash-Lite: $0.25 / 1M input + $1.50 / 1M outputRecommended default for most production learner-support conversations.
Escalation routeGemini 3.5 Flash: $1.50 / 1M input + $9.00 / 1M outputUse when ambiguity, risk, confidence, or quality requires it.
Routed policy85% Gemini 3.1 Flash-Lite + 15% Gemini 3.5 FlashNear-premium UX without paying premium pricing on every conversation.
Grounding/search upliftFirst 5,000 grounded prompts free, then $14 / 1,000 promptsExternal grounding is the fastest way to break the $250 AI line.
Cloud formula$430 base + peak/MAU uplift above planning thresholdsSeparates elastic baseline from campaign or stress capacity.

Development budget breakdown

One-time build cost by workstream, separated from maintenance.

$154k build
WorkstreamBudgetClient meaning

Total cost and unit economics

Converts the selected monthly model into a full programme view.

Inside envelope
Submitted envelope$250,000

$154k build + $96k maintenance.

Modelled TCO$236,888

Build + selected monthly model x 32.

TCO delta$13,112

Positive means remaining programme headroom.

Cost/target registration$0.79

Modelled TCO over 300,000 registrations.

Monthly cost/MAU$0.10

Selected maintenance model over MAU.

AI cost/1k convos$9.63

AI line divided by text conversations.

Line-item cost calculator

Monthly and 32-month values are calculated from the selected scenario.

LineBudget/moModel/moDelta/mo32-month modelStatus

Scenario cost comparison

ScenarioPeakCloudAIMaint./moStatus

AI route sensitivity

RouteAI/moDelta vs $250Read

Published model price comparison

Text-only estimate using 10,000 input + 2,000 output tokens per conversation.

Token-only
ModelCost / conversation1k convos5k convos10k convosStrategic read

Batch / Flex economics

Useful for offline reporting, classification, and bulk processing; not the default for live learner chat.

Model1k convos5k convos10k convosUse when

Voice and self-host caveats

Option1k5k10kRead

Approval math

These rows translate usage choices into monthly overage and decision language.

Decision triggerAdded cost/moIf 32 monthsClient decision

Included vs separately approved

Inside retainer when cappedSeparately approved or pass-through
Business-hours support, QBR/reporting, 5 dev-hours, dependency/security checks.24/7 coverage, new feature scope, major integrations, or additional dev capacity.
Elastic baseline cloud and ordinary storage/monitoring within the $700 infra line.Always-on 5k reserve, dedicated cluster, campaign pre-warm, or load-test windows.
Routed AI support within the $250 AI line and approved fallback behaviour.High-quality model everywhere, broad web grounding, unusually high chatbot adoption, or no per-session cap.
Standard email, video testimonial, survey, and translation tool allowances.Large notification blasts, heavy media usage, new survey tools, or expanded translation volumes.
Common LTP upload schema, validation, queue ageing, and agreed reporting.Provider-specific parser logic, custom LTP workflows, or remediation of poor source data.

Infra mode choices

ModeEstimate/monthStatus
Elastic baseline: Cloud Run + managed services$588Fits $700
GKE Autopilot baseline$677Tight
Pre-warmed campaign month$1,708Approve separately
5k load-test / stress month$4,668Not steady state
Dedicated 10-node Kubernetes equivalent$2,903Different commercial model

Cost escalation triggers

1
Always-on reserved peakIf AVPN wants guaranteed 5k capacity every month, price it as a reserved-capacity option.
2
Higher-cost AI defaultGemini everywhere breaks the $250 AI line under many campaign assumptions.
3
General web groundingExternal search should be an approved uplift, not default behaviour for every turn.
4
New languages or custom LTP parsersThese are scope and operations changes, not invisible retainer work.

AI route comparison

Same traffic, different model policies.

Model routing keeps high-volume support away from expensive defaults while preserving escalation quality.

Support policy

LayerDefault behaviourControl
FAQ firstApproved catalogue and platform guidance answer common questions.No token spend where static support is enough.
Low-cost routeBounded navigation and discovery support.Per-session and monthly caps.
Escalation routeAmbiguous, sensitive, or quality-critical cases.Stronger model, sampled review, budget visibility.
Guided help modeIf caps are approached, assistant degrades gracefully.Learners are not shown blunt budget failure language.

Language quality risk

Lower-resource languages need review queues and explicit fallback rules.

Quality controls

1
Grounding signalTrack whether answers came from approved FAQ/catalogue content or fallback.
2
Feedback by languageCapture unhelpful reports, issue type, and time-to-resolution.
3
Human samplingUse native-speaker or LTP-assisted review before declaring a language production-ready.
4
Post-incident reviewMaterial AVPN/LTP-reported quality issues should receive a documented review within 5 business days.

LTP verification load

The 25-30 LTP challenge is workflow governance, not mainly compute.

Operating model

1
Common upload templateRequire shared learner identifiers, completion status, completion date, and evidence fields.
2
Validation before importReject schema errors early; route ambiguous records to an on-hold queue.
3
Audit trailTrack who uploaded, what changed, and why records were accepted or rejected.
4
Change-order boundaryProvider-specific parser logic beyond the agreed schema should be separately approved.

Monthly LTP operating rhythm

CadenceSurfaceDecision it supports
WeeklyUpload success, rejected rows, schema exceptions, queue age.Which LTP needs support before reporting quality degrades.
MonthlyCompletions, certificates, unresolved verification cases, LTP responsiveness.Which delivery partners create programme risk.
Campaign windowDaily throughput, peak queue depth, error types, support tickets.Whether to extend campaign capacity or intervene operationally.
QuarterlyData quality trend and schema-change requests.Whether the common template needs governance updates.

M&E dashboard spine

These are the client-facing reporting modules the platform should make visible after launch.

ModuleQuestions answeredCore fields
Learner funnelWhere do learners drop off?Registration, recommendation, click-through, start, completion, certificate, survey.
Programme reachWho is being reached across markets?Country, cohort, language, approved demographic fields, LTP, course.
LTP deliveryWhich partners are operationally healthy?Uploads, completions, rejected rows, verification ageing, support exposure.
AI and language qualityWhere is the support layer risky?Usage, escalation, fallback, feedback, language tier, issue ageing.
Finance guardrailsAre costs inside the approved envelope?Cloud, AI, translation, email, survey, campaign uplift, approval status.

Reporting promise

A
Monthly operating reportCost, learner funnel, LTP queue health, language risk, and incidents.
B
Donor-ready exportClean CSV/XLSX export for agreed outcomes and disaggregation fields.
C
Exception registerSeparate normal programme movement from data quality, integration, or cost exceptions.
D
Governance reviewUse the data to decide capacity, AI caps, LTP support, and language QA investments.
Positioning

This is not just a dashboard. It is the governance layer that makes the fixed retainer, LTP network, and multilingual AI support manageable.

Delivery roadmap and gates

A five-month build only stays credible if each phase has a visible proof gate and a named decision it resolves.

Gate-led delivery
Month 1 Confirm operating assumptions

Lock scope, billing stance, AI policy, languages, and LTP data rules.

  • Architecture decision record
  • Common LTP schema
  • AI caps and fallback policy
Month 2 Build core gateway

Implement learner registration, recommendation/routing, identity, and admin foundations.

  • Staging environment live
  • Audit log baseline
  • First funnel dashboard
Month 3 Wire LTP and evidence flow

Validate uploads, completion evidence, certificates, queue ageing, and exception handling.

  • LTP pilot upload
  • Verification queue
  • Certificate workflow
Month 4 Harden AI and reporting

Operationalize language QA, cost dashboards, graceful fallback, and M&E exports.

  • Language review pack
  • Cost guardrail alerts
  • Donor-ready export
Month 5 Prove readiness and handover

Run load proof, UAT, support rehearsal, runbook review, and go-live decision gate.

  • 5,000-session load proof
  • UAT sign-off
  • Handover pack

Readiness gates

GatePass conditionOwner decision
Cost gateInfra, AI, translation, email and survey usage have caps and reporting.Approve baseline vs pass-through model.
Scale gateLoad test documents 5,000-session envelope and campaign pre-warm runbook.Approve launch/campaign window.
Language gateTiered language QA, fallback, feedback and incident review are live.Approve production language list.
LTP gateAt least one pilot upload proves validation, queue ageing and audit trail.Approve common upload schema.
Handover gateRunbooks, ADRs, backups, org-owned access and support routes are reviewed.Approve maintenance transition.

What AVPN should see before go-live

1
Working learner journeyRegistration, recommendation, LTP handoff, completion evidence and certificate path.
2
Operating dashboardFunnel, cost, AI quality, language risk, LTP queue health and exception register.
3
Proof packLoad-test result, UAT evidence, pilot LTP upload, language review sample and runbooks.
4
Decision logBilling model, AI model route, support coverage, campaign rules and change boundaries.

Risk register

RiskImpactMitigation
Retainer interpreted as unlimited usageCommercialHard caps, assumptions table, approval workflow.
5k treated as steady-state capacityCostLoad-test ceiling plus campaign pre-warm option.
Low-resource language answer is wrongTrustFAQ grounding, feedback, review sampling, fallback and incident process.
LTP CSVs are inconsistentOperationsCommon template, validation, on-hold queue, audit trail.
One-person technical dependencyContinuityRunbooks, ADRs, code review, org-owned access, named backup coverage.
Exact pricing drifts before submissionEvidenceRefresh vendor pricing before final commercial quote.

Decisions to close

1
Billing modelAVPN-owned GCP project, Mereka-managed capped usage, or hybrid.
2
5k interpretationConfirm load-test ceiling versus permanently reserved capacity.
3
AI policyApprove model routing, hard caps, grounding limits, and graceful fallback.
4
LTP data rulesApprove common upload schema and provider-specific change boundaries.
5
Support commitmentConfirm business-hours plus P1/P2 escalation or price broader coverage.

Assumption sources

AVPN TOR and clarification trailDefines the requested learning navigation/data-management platform, 300,000 registration scale, multilingual support, LTP network, and 5,000-session readiness expectation.
Submitted commercial proposalDefines the $154,000 build budget, $3,000/month maintenance model, $700 infra line, $250 AI line, and other tool/API allowances.
Mereka clarification response draftDefines the bounded AI support posture, fallback UX, continuity practices, maintenance framing, and lower-resource-language quality controls.
Internal finance and concurrency modelsTranslate registration volume into MAU, gateway minutes, peak concentration, cloud cost, AI route cost, and approval thresholds.

Vendor anchors to refresh

Exact prices and product capabilities should be refreshed before a final commercial submission.

Google Cloud Run pricingcloud.google.com/run/pricingUsed to anchor elastic baseline assumptions.
Google Kubernetes Engine pricingcloud.google.com/kubernetes-engine/pricingUsed to compare Autopilot and reserved-capacity alternatives.
Gemini Developer API pricingai.google.dev/gemini-api/docs/pricingUsed for Gemini API, Batch/Flex, voice, and 2.5-family text-token rates.
Gemini / Vertex generative AI pricingcloud.google.com/gemini-enterprise-agent-platform/generative-ai/pricingUsed to compare low-cost, routed, and high-quality AI policies.
Cloud Run GPU docsdocs.cloud.google.com/run/docs/configuring/services/gpuUsed for L4 and RTX minimum CPU/memory assumptions.
Google Gemma docsai.google.dev/gemma/docsUsed as an open-model route reference, with hosting costs treated separately.

Client caveat

This dashboard is a planning and decision-support artifact. Final contractual numbers should be reconciled against approved scope, region, billing ownership, selected AI model policy, campaign assumptions, and live vendor pricing.