The agentic new product launch manager: coordinating innovation from concept to shelf

A consumer-goods company is launching a high-protein breakfast drink. The concept was compelling, the business case forecast €18 million of first-year net revenue, retailers expressed interest, and the executive committee approved it. Twelve months later, the launch date is six weeks away.

Marketing believes the product is ready. Packaging is still reviewing a front-label claim. Regulatory has approved the formula but not the latest artwork. The factory completed one successful trial, but line speed remains below the business-case assumption. Procurement has the bottle, not the preferred cap colour. Finance has not finalized standard cost. Demand planning is using an analogue from another category. Sales has entered retailer forecasts assuming national distribution from launch week — while two retailers have confirmed listings, three have expressed interest without activating anything in their systems, one will list only the chocolate flavour, and another requires shelf-ready packaging that the current case design does not provide.

The consumer GTIN exists. The case GTIN is still being corrected. The e-commerce images show the old package. The retailer portal carries an unapproved nutrition claim. Media begins in Week 38; the largest production run is Week 35; the first bottles have twelve months of shelf life. The commercial forecast bundles retailer pipeline fill, store inventory, consumer trial, repeat purchase, and promotional loading into one number, and nobody can say how much of the year-one forecast is genuinely incremental — the business case assumes 15% cannibalization while the demand team believes 35%.

At the final readiness meeting every function reports status. Marketing green, sales amber, supply amber, regulatory green, finance amber, master data green, retailer readiness amber. The consolidated report reads overall readiness: 82%.

The percentage is not merely optimistic — it is structurally misleading. The missing case GTIN may prevent the product from beingordered at all. The unresolved shelf-ready packaging may block one of the largest listings. The unapproved claim creates compliance risk. The lower line speed changes cost, capacity, and launch inventory simultaneously. The media campaign may create demand in stores where the product does not yet exist. The project is not 82% ready; it contains several dependencies that could make the entire launch fail.

A launch is not one project milestone. It is a network of synchronized commitments, and a delay in one part can invalidate all the others.

Most companies already have Stage-Gate, PLM, project management, commercialization workflows, readiness checklists, demand planning, retailer portals, and PIM systems. But the product lives in PLM, the forecast in demand planning, retailer commitments in CRM or spreadsheets, product data in ERP or PIM, artwork in a digital-asset system, regulatory approvals in email, production readiness in manufacturing systems, the business case in finance workbooks, and the plan in project software — with the executive committee receiving a presentation assembled from all of them. The organization has systems. It does not have one continuously governed launch state.

The objective is not to launch every approved idea on time. It is to launch the right products, with the right evidence and operating readiness, into the right customers and channels, at a level of investment proportionate to uncertainty — and to learn quickly enough to protect both upside and downside.

A new product launch is the cross-functional conversion of a validated opportunity into a product that can be manufactured, ordered, distributed, purchased, experienced, repurchased, and managed profitably. Two halves have to work: innovation determines what to create, for whom, why it wins, and whether it can be built; commercialization determines how it is produced, sold, delivered, supported, and measured. A desirable product fails through poor commercialization; a flawlessly executed launch fails on a weak proposition. And critically, the launch is not complete when the product ships — a first shipment is pipeline fill, DC inventory, store inventory, and launch stock. It proves nothing about trial, penetration, satisfaction, repeat, sustainable velocity, portfolio incrementality, or retailer profitability.

Before deciding how much process to apply, classify the launch. The ten recurring archetypes each carry different risks: renovation (existing-user rejection, production transition, old and new stock overlap, product identification), line extension (cannibalization, assortment complexity, retailer space, low incrementality), price-pack innovation (value-ladder distortion, pack switching, unit economics, channel conflict), new-to-brand (brand stretch, capability gap, analogue selection), new-to-category (low consumer understanding, retailer classification, regulatory uncertainty, limited analogues), new brand, market extension, channel extension, customer-exclusive product (dependence, volume concentration, residual inventory), and regulatory or mandatory change, where the objective is continuity rather than growth. GS1’s GTIN Management Standard also distinguishes new product introductions from product changes, defining when new identification is required — net content, functionality, brand, pack or case quantity, predefined assortments, and certain promotional configurations.

Stage-Gate exists to progressively reduce uncertainty while allocating increasing resources only to opportunities that still justify investment. Stages are for learning, not paperwork — each one answers the question the next investment decision needs: is there a meaningful opportunity, can we create and capture value, can we build it reliably, does the complete proposition work, are market and organization ready? And gates are investment decisions, so a gate should never ask whether every template field is complete but whether the evidence justifies the next commitment of scarce money, people, capacity, and brand equity. Outcomes are go, conditional go, hold, recycle, redirect, or kill — and kill decisions create value, protecting capital, capacity, inventory, commercial attention, and retailer credibility. A pipeline without attrition is a tunnel, not a funnel.

Weak:   Go, subject to supply readiness.

Strong: Conditional go.
        Condition: line trial must achieve 85% of target
                   speed with approved quality yield by 15 August.
        If not achieved: delay national production and
                   activate the regional launch option.

None of this requires rigid sequential development. Within stages, teams can run rapid prototypes, design sprints, formulation iterations, consumer co-creation, pilots, and test-and-learn. Governance decides whether investment continues; the development method decides how learning happens. Gate evidence should be risk-based too — a low-risk line extension may need analogue evidence, a limited product test, retailer validation, and a simplified case, while a new category needs extensive research, technical pilots, regulatory advice, supply qualification, probabilistic scenarios, and a staged market launch.

The first question is not whether the idea is attractive but whether it belongs in this portfolio and strategy. A launch serves a portfolio role — penetration driver, premiumization, new occasion, affordability, margin enhancer, category entry, channel access, brand modernizer, defensive renovation, regulatory replacement, or test platform — and that role determines the success metrics. Beyond the individual case sit three portfolio constraints that gates routinely ignore. Crowding: too many launches create development overload, retailer fatigue, production complexity, fragmented marketing, poor launch support, and internal cannibalization — the question is not how many concepts the company can create but how many launches it can execute properly. Capacity: R&D, packaging, regulatory, line-trial, customer-selling windows, marketing budget, production, and master-data resources are all finite, so a project can be individually attractive and collectively impossible. And dependency: two launches may share an ingredient, a production line, retailer shelf space, a marketing audience, a launch period, or a packaging supplier — competition the agent should make visible rather than discovering at the trial.

Concepts should begin with a problem, not a product. “We have a new protein technology” is weaker than “busy working adults skip breakfast because current protein options are inconvenient or taste medicinal.” Technology may enable the solution; it is not the need. From there the discipline is sharpness — a proposition that tries to solve every need is harder to develop, communicate, test, sell, position, and measure. But the most consequential rule in this section is about evidence quality: stated interest is not demand. Purchase intent is affected by research setting, novelty, socially desirable responses, incomplete competitive context, unrealistic distribution assumptions, and the absence of a real budget constraint. Behavioural evidence — test-market purchase, digital conversion, sample redemption, repeat order, controlled retail test, home-use trial, conjoint or choice modelling, actual willingness to pay — carries more weight. And test against the real environment: direct competitors, private label, substitutes, existing own products, the expected shelf, and actual price points.

Then validate the product separately, because concept strength is not product strength — a great concept can be delivered through a disappointing product, and a strong product can hide behind a weak concept. Blind testing isolates product performance; branded testing measures the total proposition, and a product can win blind and lose branded because expectations differ. Benchmark against the category leader, private label, the target competitor, the existing portfolio, and the concept promise. Use home-use testing where performance depends on repeated use, household context, preparation, or real routines. Run failure-mode testing — misuse, extreme storage, transport, temperature, opening, resealing, portioning, damage, and consumer misunderstanding. And define acceptance thresholds in advance: must-win attributes, acceptable parity, minimum quality, unacceptable defects. The closing question is whether the experienced product credibly delivers the promised benefit.

Technical development turns the proposition into something reproducible: specification, bill of materials, recipe and routing — and then the reality that a laboratory prototype is not a commercial product, since scaling changes taste, texture, appearance, yield, stability, cost, and cycle time. Industrial trials need an objective, batch size, acceptance criteria, measurements, deviation handling, and a repeat requirement, and trial success is multidimensional: a batch can meet product quality while failing target speed, yield, labour, packaging integrity, changeover time, or cost. Shelf-life evidence must precede launch inventory, and supplier readiness means an approved supplier, specification, quality agreement, capacity, lead time, MOQ, contingency, and risk assessment for every critical input.

Regulatory readiness is part of product design, not a final artwork check. Requirements vary by country, category, ingredient, process, claims, consumer group, pack, and channel, and qualified specialists must define them. For prepacked food in the EU, Regulation 1169/2011 governs mandatory food information — product naming, ingredients and allergens, net quantity, date marking, storage, nutrition declaration (energy, fat, saturates, carbohydrate, sugars, protein, salt, generally per 100 g or 100 ml), and responsible business information — with mandatory information understandable in the market of sale, which for multi-market launches means market-specific artwork, multilingual packs, stickers, or separate SKUs. Nutrition and health claims must fit the applicable legal framework, and novel foods may require authorization and safety assessment before market placement, through a process with pre-submission, submission, suitability assessment, risk assessment, and post-adoption stages. Some approvals exceed the commercial timeline, so regulatory work belongs on the critical path — and a claim approved for formula version 3 must be reassessed when the formula moves to version 4.

Packaging is simultaneously product, communication, and supply chain, and its levels — consumer unit, inner pack, case, display unit, pallet — are linked but distinct trade items, each needing correct identification and master data. Two failure points deserve naming because they stop a launch dead while looking like administration. Barcode verification: a GTIN can exist while the printed barcode fails on size, contrast, position, quiet zone, print quality, or simply the wrong number — physical verification matters. And case-level readiness: retailer ordering depends on case GTIN, case quantity, dimensions, weight, and pallet information, so a perfect consumer unit with incomplete case data will not move through the supply chain. Alongside these, artwork versioning must bind artwork, formula, claim, market, language, approval, printer proof, and effective date — and the same approved facts must feed the physical pack, retailer portals, e-commerce pages, advertising, product sheets, and sales materials, because contradictory digital content is both a compliance and a trust problem.

A new product needs a value architecture, not a price: absolute price, unit price, pack role, price index, retailer margin, manufacturer margin, promotional price, and channel position. Elasticity is genuinely uncertain before launch — evidence comes from choice models, concept tests, willingness-to-pay research, analogues, category elasticities, and test markets, so use ranges. The retailer story needs shelf price, acquisition cost, cash margin, margin rate, category incrementality, velocity, space productivity, and promotional support. The launch must strengthen rather than accidentally break the pack ladder. And promotional architecture is a decision with a long tail: launching too deeply discounted creates the wrong reference price, which then constrains the product for its whole life. On cost, separate launch cost from steady-state cost from target cost, since early estimates typically assume target volume, yield, line speed, and supplier scale that the first months will not deliver.

The business case is a model of assumptions, not a revenue number. The retail structure decomposes to eligible stores × distribution × units per store per period × periods — but it must separate launch phases, and distribution itself is not one thing: retailer commitment, retailer system activation, DC distribution, store distribution, on-shelf distribution, and weighted distribution are six different states, and treating an expression of interest as distribution is how forecasts become fiction. Pipeline fill — DC stock plus shelf stock plus backroom stock plus safety inventory — is not recurring demand. Sustained consumption comes from target households × awareness × availability × trial rate × repeat rate × repeat frequency × units per purchase, and trial and repeat fail for entirely different reasons: trial depends on awareness, distribution, shelf visibility, proposition, price, promotion, sampling, and brand, while repeat depends on product experience, satisfaction, availability, price, occasion recurrence, habit, and competitive response.

New products have no direct history, so forecasts combine analogues, product attributes, consumer research, distribution, commercial plans, category dynamics, and launch execution — with planning systems supporting reference products, weighting factors, market-specific launch dates, and phase-in curves that shift toward the product’s own history as it accumulates. Analogues should be selected on category, brand, benefit, price tier, pack, channel, distribution, launch support, consumer target, and occasion, often combining several (category, brand, pack, channel, launch-support analogues) and normalized for market size, distribution, price, inflation, media, retailer mix, category growth, and seasonality. The trap is social rather than statistical: teams select the most successful historical launch, so the agent should compare similarity, surface failure cases and base rates, and record the selection rationale.

The forecast must then separate sell-in, sell-out, pipeline inventory, consumer consumption, returns, and cannibalization, and be built per retailer rather than spreading one national curve uniformly — different customers have different launch dates, store counts, packs, prices, promotions, and shopper profiles. Distribution ramps should reflect actual rollout (20% of stores in week 1, 50% by week 4, 75% by week 8, 90% by week 12), and launch lag deserves its own tracking, since commercial approval, system activation, DC availability, store distribution, and shelf availability happen on different dates. When retailers supply forecasts, validate what the number actually represents — order volume, sell-out, full-year demand, or a single event. And on ownership: marketing owns proposition assumptions, sales owns customer commitments, demand planning owns the integrated signal, and supply should not decide demand by constraining it.

Supply readiness is much more than inventory. It spans materials, approved suppliers, capacity, line qualification, quality, labour, maintenance, warehousing, transport, product data, and contingency — and the most common silent failure is capacity assumption drift: the business case assumes 10,000 units per hour, the trial achieves 7,800, and cost, capacity, inventory timing, and launch feasibility all move at once. A production readiness review should confirm formula, BOM, recipe, materials, equipment qualification, trial results, quality plan, updated cost, released master data, reserved capacity, and available contingency. Manufacturing master data — material master, BOM, recipe, routing, work centre, quality plan, batch rules, shelf life, costing, sourcing, packaging hierarchy — is a launch deliverable, not administration. Where possible use postponement (common base product, late-stage label, market-specific sleeve, customer-specific case) to reduce risk, and prevalidate every contingency rather than discovering at the moment of need that the alternate supplier requires its own technical, claim, regulatory, cost, consumer, and retailer review.

A retailer listing is a project, not a yes.It may require buyer approval, category rationale, product and case data, images, cost price, shelf price, margin, forecast, distribution, launch date, promotion plan, supply terms, and compliance documents — and the single most valuable discipline in this whole section is refusing the phrase “retailer confirmed.” Use precise states instead: not approached, pitch scheduled, under review, conditional approval, approved, system setup, orderable, DC stocked, store ranged, shelf live. Conditional approvals (“approved subject to shelf-ready case, final images, and promotional funding”) belong in the critical path, not in a note. Each retailer may have a different SKU set, pack, date, price, promotion, and distribution ramp; range-review windows are fixed, so missing the window can delay a launch by months; and the digital shelf needs title, description, images, ingredients, nutrition, claims, search terms, and availability, with the caution that appearing online before physical availability generates unavailable demand, poor reviews, and consumer confusion.

Marketing readiness follows the same logic: media should not create demand where the product cannot be purchased. The campaign depends on final pack, claims, retailer dates, distribution, inventory, and pricing, and the launch wave should run internal readiness → trade sell-in → retailer system activation → product availability → in-store and digital execution → consumer media → sampling → repeat support. The diagnostic discipline that saves the most money here: awareness is not always the first problem. A new product may be failing on distribution, availability, shelf position, proposition clarity, or repeat — and none of those is fixed by more media.

Financially, launch readiness means a confirmed standard cost (material, packaging, conversion, labour, freight, overhead, scrap, yield, co-manufacturing), a margin bridge from gross list revenue through customer terms, launch trade spend, promotions, variable cost, and logistics to launch contribution, and a budget released by gate rather than at concept approval — investment should follow evidence. It also means quantifying write-off exposure explicitly: inventory value at risk plus obsolete packaging plus unused ingredients plus retailer penalties plus committed media. And a delayed launch moves revenue, cost, marketing, and write-offs between financial periods, so the outlook has to be updated rather than left to surprise the close.

Checklist percentages fail because they treat tasks as independent: a launch can have 100 completed minor tasks and one unresolved critical dependency. What matters is the critical path — the tasks whose delay changes the launch date or destroys a key launch condition — and its chains are predictable. Formula freeze → regulatory approval → artwork approval → print → packaging delivery → production → retailer delivery. Retailer approval → product setup → case data validation → purchase order → DC receipt → store activation. Each dependency should carry task, prerequisites, owner, planned date, latest safe date, status, confidence, impact if late, and fallback.

The essential distinction is between hard blockers — safety approval missing, product cannot be manufactured, barcode invalid, retailer cannot order, mandatory data missing, inventory unavailable — and soft risks like a late media asset, a delayed secondary retailer, a missing optional flavour, or below-target line speed, which can still be commercially material. Readiness should therefore be reported by outcome dimension (consumer, product, regulatory, supply, customer, data, marketing, financial, execution) with status meaning risk to an outcome: green is complete or high confidence, amber is recoverable risk with active mitigation, and red means the launch objective cannot currently be achieved. Track completion and confidence separately — a task can be 90% complete with low confidence while another has not started and can be finished in a day. And calculate the launch-at-risk date: the point after which recovery becomes improbable, such as “packaging approval by 12 August, after which national launch cannot hold without air freight or reduced scope.”

The launch gate then answers exactly one question: is the complete proposition and operating system ready enough to justify commercialization at the proposed scale? The available answers are richer than go/no-go — full-scale go, go with conditions, go at reduced scope, pilot, delay, recycle, or kill. A pilot fits when uncertainty remains and limited launch generates evidence that will change the scale decision. A delay is not failure; a launch on the wrong date can be more damaging than a later one. Scope reduction — fewer markets, retailers, or SKUs, online first, regional pilot, delayed media, lower inventory, phased rollout — is the most underused option in practice. And killing at the launch gate remains appropriate when remaining value no longer justifies remaining exposure, because sunk costs should never force commercialization.

A control tower is not a dashboard — a dashboard reports status, a control tower coordinates decisions and actions. It runs a pre-launch countdown (T−26 business-case confirmation, T−20 product and supplier freeze, T−16 customer sell-in, T−12 forecast and capacity commitment, T−8 master data and artwork complete, T−6 initial production, T−4 retailer orders and media lock, T−2 inventory and execution verification, T shelf launch), shifting from weekly cadence on readiness, critical path, and cross-functional risk to daily cadence near launch on production, shipments, retailer activation, inventory, and issue resolution. Its agenda is decisions, hard blockers, critical-path changes, customer activation, supply, consumer activation, financial exposure, and deadlines — not a status round for every task. And it maintains explicit launch assumptions with triggers: “if retailer system activation is below 700 stores by 1 September, delay national media and shift budget to activated regions.”

That divergence is the most common way launches are misread. Early shipments look strong because of pipeline fill, initial orders, retailer safety stock, and promotional loading — while sell-out, which measures actual shopper purchase, is still building through distribution, availability, shelf placement, promotion, and media. Observed sales decompose to distribution × availability × velocity, so a weak launch may reflect any of the three, and distribution-adjusted velocity (units per selling store per week) tells the truth that total volume conceals. A sustainable launch needs reach → trial → satisfaction → repeat → routine, which household panel data reads through penetration, repeat, frequency, buyer profile, source of volume, and repertoire.

Three matrices carry most of the diagnostic load, and their value is that each points at a different owner.

Before acting on any of them, adjust for availability — low velocity with low on-shelf availability is an execution problem, not a proposition problem, which is precisely where the launch connects to Perfect Store monitoring. Complaints (taste, pack damage, leakage, preparation, claim misunderstanding, allergen concern, portion, price) and returns (overforecast, retailer inventory, short shelf life, product issue, execution failure) add diagnostic texture, and cannibalization should be measured by comparing launch buyers against prior portfolio and category purchases rather than assumed at the business-case rate.

The corrective options then form a genuine decision set rather than a binary. Scale when velocity exceeds plan, repeat is strong, supply is manageable, and retailer economics work. Fix when the proposition holds but execution fails — distribution, availability, shelf, price, digital content, media, packaging quality. Localize when performance differs by region, retailer, demographic, channel, or store format. Hold when evidence is insufficient, defining what information will resolve the decision. Reformulate or repackage when satisfaction, usability, claims, pack, or quality limit repeat. Reprice when the value architecture is wrong. Reduce scope by exiting weak retailers and concentrating on strong clusters. Or stop — with a responsible exit plan covering retailer communication, inventory, materials, packaging, consumers, write-off, replacement, and learning.

Innovation and project systems
  -> shared product and launch data layer
  -> Agentic New Product Launch Manager
  -> consumer, PLM, regulatory, packaging, finance,
     demand, supply, customer, marketing, execution tools
  -> deterministic forecasting, financial,
     scheduling, and compliance engines
  -> gate and launch decision packets
  -> authorized human decisions
  -> execution workflows and transactions
  -> launch monitoring
  -> post-launch learning

The agent maintains the integrated launch state, detects missing dependencies, retrieves evidence, compares functional assumptions, forms decision cases, orchestrates forecasts and scenarios, prepares gate packs, routes actions, monitors readiness, diagnoses early performance, and preserves learning. Deterministic services own financial calculations, demand forecasting, capacity planning, product-data validation, regulatory rules, scheduling, authorization, transactional updates, and audit. Humans own innovation strategy, consumer judgment, formula and quality approval, regulatory interpretation, retailer negotiation, financial risk, gate decisions, major launch changes, and accountability. An independent launch evaluator earns its place here, checking evidence quality, critical-path completeness, contradictory assumptions, financial consistency, supply feasibility, regulatory approval, retailer readiness, and decision authority before anything reaches a gate.

The twenty-one stages

1. 01Register the opportunity. Consumer problem, category, brand, strategic role, project type, risk class, sponsor.
2. 02Create evidence requirements. Determined by archetype and risk — concept research, product testing, technical study, regulatory work, financial analysis, retailer evidence.
3. 03Build the dependency graph. Concept, formula, claims, artwork, data, forecast, supply, customers, media, launch date.
4. 04Maintain project state. Deliverables, evidence, decisions, assumptions, risks, owners, deadlines.
5. 05Prepare gate decisions. Evidence summary, missing information, scenarios, recommendation, decision requested.
6. 06Record the gate outcome. Go, conditional go, hold, recycle, kill — with rationale and resource commitment.
7. 07Coordinate development. Product, pack, suppliers, trials, regulatory, cost.
8. 08Prepare the commercial plan. Positioning, price, pack, customer targets, channel, activation.
9. 09Build the forecast. Analogues, distribution, velocity, trial, repeat, cannibalization, scenarios.
10. 10Build the supply plan. Material, capacity, inventory, shelf life, contingency.
11. 11Coordinate customer readiness. Pitch, approval, setup, orderability, first order, store activation.
12. 12Verify product data. GTIN, descriptions, dimensions, case, pallet, images, claims, language, effective dates.
13. 13Calculate readiness and critical path. Hard blockers, recoverable risks, launch-at-risk dates, fallbacks.
14. 14Run the launch gate. Full-scale, reduced-scale, pilot, delay, stop.
15. 15Translate the decision into commitments. Production orders, customer tasks, media changes, data submissions, inventory allocations.
16. 16Monitor pre-launch execution. Daily or weekly, depending on proximity.
17. 17Monitor the shelf launch. Shipments, distribution, availability, price, promotion, digital presence.
18. 18Diagnose early performance. Separating demand, distribution, supply, execution, and consumer repeat.
19. 19Recommend adaptation. Scale, fix, localize, reforecast, reduce, stop.
20. 20Conduct the post-launch review. Original assumptions, gate assumptions, actual execution, consumer and financial outcomes.
21. 21Store validated learning. Reusable analogues, risks, launch curves, retailer patterns, execution standards, decision lessons.

• Project and evidence: create_launch_project (archetype, risk class, required process), get_launch_charter, get_stage_gate_requirements (deliverables, criteria, approvers, evidence threshold), get_consumer_research (concept and product results, willingness to pay, barriers, confidence), compare_concept_versions.
• Product and regulatory: get_product_specification (formula, ingredients, quality, claims, shelf life, version, status), get_regulatory_readiness (markets, requirements, approved claims, artwork status, open issues, decision owner), get_packaging_readiness, get_quality_readiness.
• Identification and content: validate_gtin_and_hierarchy (consumer GTIN, case GTIN, pallet, barcode, hierarchy, effective dates), validate_product_content — checking consistency across pack, PIM, retailer data, e-commerce, and marketing.
• Economics: get_product_cost (current, launch, target, confidence, variance), calculate_business_case (deterministic), estimate_cannibalization (manufacturer transfer, retailer transfer, category incrementality, confidence).
• Forecast: select_launch_analogues (with similarity dimensions, performance history, and limitations), create_new_product_forecast, get_forecast_distribution, decompose_launch_demand — separating pipeline fill, store stock, sell-out, trial, repeat, and cannibalization.
• Supply: get_material_readiness, get_capacity_readiness (line, speed, yield, reserved hours, competing demand, contingency), simulate_initial_inventory (launch stock, safety stock, ageing, write-off risk, service).
• Commercial: get_retailer_launch_status (pitch, approval, conditions, system activation, orderability, initial order, store distribution, shelf date), get_marketing_readiness.
• Readiness and decision: build_launch_dependency_graph, calculate_launch_readiness — which must expose hard blockers separately — calculate_launch_at_risk_date, run_launch_scenario, create_gate_decision_packet, create_gate_approval_request (bound to evidence version, scope, cost, conditions, launch date).
• Execution and learning: apply_approved_launch_change, create_launch_commitments, monitor_launch_execution, get_early_launch_performance, diagnose_launch_performance, create_launch_adaptation_proposal, run_post_launch_review, propose_launch_learning.

Weak:   "The launch is at risk."

Strong: launch_id: NPI-2048   target_launch: 2026-W38
        decision: maintain national launch |
                  reduce scope | delay
        hard_blocker: case-level GTIN not accepted
                      by Retailer A
        critical_condition: industrial trial must reach
                      8,500 units/hour
        current_result: 7,800 units/hour
        retailer_distribution_confirmed: 720 of 1,400 stores
        inventory_at_risk: EUR 640,000
        latest_safe_decision_date: 2026-08-12
        recommended_option: launch in confirmed retailers,
                      delay national media, expand after
                      line-speed recovery
        required_approvers: Innovation, Commercial,
                      Supply Directors, CFO

Five state machines run in parallel, and the granularity is the point. The project moves through idea, discovery, scoping, business case, development, validation, launch ready, launching, post launch, scaled, held, stopped, closed. The gate moves through preparing, ready, decision pending, go, conditional go, hold, recycle, kill. The deliverable moves through not started, draft, review, approved, blocked, superseded. The retailer moves through targeted, pitched, under review, conditional, approved, system active, orderable, ordered, distributed, shelf live — and only “orderable” onward means the product can actually generate revenue. The launch issue moves through detected, investigating, decision ready, assigned, in progress, resolved, verified, reopened.

Launch memory holds analogue launch curves, retailer activation lead times, trial-to-repeat patterns, category cannibalization, common packaging delays, line-trial performance, initial-fill patterns, media timing effects, readiness failures, and successful recovery actions — each versioned with market, period, category, support level, price, distribution, product type, and confidence, because old launch evidence decays with channel change, category change, inflation, retailer concentration, media change, and supply-model change. Never stored automatically: one executive opinion, an unsupported retailer forecast, an unvalidated consumer claim, a temporary workaround, an early pipeline-fill spike, one week of velocity, a regulatory interpretation, or a low-confidence product match.

The agent can fail by missing a dependency, using the wrong product version, treating conditional retailer interest as approval, using pipeline fill as demand, ignoring cannibalization, overstating readiness, recommending an infeasible launch, failing to propagate a decision, or reacting incorrectly to early sales. Evaluation therefore layers: data retrieval, entity and version resolution (SKU, GTIN, formula version, pack version, market, retailer, date), dependency (critical dependencies found, false dependencies, latest-safe-date accuracy, blocker detection), readiness — where hard-blocker recall and false-green rate are the two numbers that matter most — gate packet (required evidence, scenario consistency, missing assumptions, financial integrity, decision clarity), forecast (distribution, initial fill, sell-out, range calibration, bias, cannibalization, retailer-level accuracy), recommendation (full launch, pilot, delay, reduce, stop — against human acceptance and eventual outcome), execution, and post-launch diagnosis, measured on whether the system correctly distinguished a distribution issue from an availability, proposition, repeat, supply, or price issue.

Trajectory tests require the agent to resolve current versions, check hard blockers, verify retailer status, decompose the forecast, check supply, reconcile financials, request the gate decision, and monitor execution — while prohibiting approving claims or formula, inventing retailer commitments, treating a target as a forecast, ignoring a product-version change, and releasing a launch without authority. The KPI families then span process (time to gate, decision latency, cycle time, rework, missed dependencies, conditional approvals, launch-date changes), portfolio (projects by stage, kill rate, resource utilization, balance, expected value, launch capacity), readiness, forecast, commercial (retailer acceptance, weighted distribution, on-shelf availability, listing speed), consumer (awareness, trial, penetration, repeat, satisfaction, source of volume, complaints), financial (net revenue, portfolio contribution, working capital, write-off, time to break-even, business-case variance), supply, and agent (blocker recall, false escalation, recommendation grounding, unsupported claim rate, human override).

The measure is risk-adjusted portfolio value from products that achieve sustainable consumer demand and operational viability — not the percentage of launches delivered on the planned date.

Back to the high-protein breakfast drink: three flavours, two countries, five retailers, Week 38, €18 million of year-one net revenue. Concept results show strong relevance and uniqueness, high purchase interest, price resistance above €2.79, and concern about artificial taste. Product testing shows chocolate strong, vanilla acceptable, and berry below target. The team plans to launch all three.

The agent’s portfolio diagnosis kills that plan: berry fails the must-win taste threshold, retailers have limited shelf space, berry adds packaging complexity, the business case assumes berry matches chocolate’s velocity, and berry represents 18% of forecast against 31% of initial minimum-order exposure. Recommendation: launch chocolate and vanilla, recycle the berry formulation, preserve it as a later extension. The gatekeepers approve the reduced range. Then the analogue evidence corrects the case itself — comparable launches show lower retailer rollout, slower repeat, and higher portfolio transfer than the base plan assumed.

Retailer readiness tells the harder story. Retailer A approved for 500 stores, conditional on a shelf-ready case. B approved for 300 stores, chocolate only. C conditional on final images and cost, 400 stores. D under review, decision in four weeks. E not approved until the next range window. The original forecast assumed 1,450 stores from Week 1; confirmed distribution is 800.Meanwhile the industrial trial delivers 7,800 units/hour against a 10,000 target and 93% yield against 96%, raising standard cost by €0.09 per unit; Retailer A’s shelf-ready case needs a five-week redesign with a latest safe decision date of 12 August; the case GTIN carries incorrect dimensions in Retailer C’s portal; and the national campaign is scheduled for Week 38 when only 800 stores are confirmed.

Four scenarios follow. A — full national launch: hold Week 38, assume conditional retailers activate, launch full media, and accept low distribution, packaging rejection, capacity, and inventory-imbalance risk. B — delay everything six weeks: packaging and capacity fixed, more retailers confirmed, against a missed seasonal period, media cancellation cost, and retailer disappointment. C — phased launch: Retailers A and B in Week 38 with chocolate and vanilla, regional media around confirmed distribution, Retailer C in Week 42 after data correction, expansion after line-speed recovery, safety inventory held for upside. D — online-first pilot: real trial and repeat data, but a weaker representative channel, a different consumer mix, and disrupted retailer commitments.

The recommendation is C — it protects confirmed retailer commitments, avoids national media waste, reduces initial inventory exposure, creates early learning, preserves expansion upside, and allows packaging and line-speed recovery. The gate approves the phased launch with five explicit conditions: shelf-ready case approved by 12 August, case data accepted by Retailer C by 20 August, line speed reaching 8,500 units/hour before the second production run, regional media replacing national, and berry staying out. Initial inventory falls from 1.6 million units to 850,000.

Week 38 delivers 92% of planned store distribution, 89% on-shelf availability, and velocity 12% above base forecast — chocolate strong, vanilla 8% below. The four-week diagnosis separates the signals: trial strong, repeat not yet mature, Retailer A availability below target, vanilla underperformance concentrated in convenience formats, chocolate supply tightening. The actions follow the diagnosis rather than the instinct: increase chocolate production, correct Retailer A shelf availability, localize vanilla distribution, proceed with Retailer C for chocolate only, and do not yet increase total media. By week eight, chocolate repeat is above threshold, vanilla repeat is acceptable in large stores, cannibalization lands at 31% — close to the revised base, far from the original 15% — retailer category incrementality is positive, and supply service is 96%.

The post-launch review is where the value compounds: initial retailer distribution assumptions were too optimistic, flavour-level forecasts need separate analogues, shelf-ready packaging must be identified before the retailer pitch, national media should be conditional on weighted distribution, and the phased launch both reduced inventory and improved learning quality. Each of those becomes a reusable constraint on the next launch. The figures are illustrative; a production system requires validated consumer, commercial, operational, regulatory, and financial evidence.

1. 01Phases 0–2 — process, archetypes, deliverables. Map the current stages, gates, systems, decisions, owners, recurring delays and failures; create differentiated pathways per archetype; define for each gate the decision, required evidence, quality standard, approval, and next investment.
2. 02Phase 3 — create the integrated launch object. Link concept, product, forecast, customer, supply, finance, activation, and decisions into one state rather than nine systems.
3. 03Phase 4 — build dependency and critical-path visibility. Move beyond task lists to prerequisites, latest safe dates, and fallbacks.
4. 04Phase 5 — deploy the launch copilot. The agent retrieves status, identifies missing dependencies, prepares reviews, tracks conditions, drafts gate packets. No approvals or transactions are automated.
5. 05Phase 6 — add forecast and scenario orchestration. Analogues, demand models, supply scenarios, financial scenarios.
6. 06Phase 7 — add retailer and master-data readiness. Track orderability and shelf activation as first-class states.
7. 07Phase 8 — add approval-based execution. The agent prepares scope changes, media adjustments, forecast updates, production changes; humans approve.
8. 08Phase 9 — build the control tower. Pre-launch countdown, launch week, post-launch reviews, decision backlog.
9. 09Phases 10–11 — diagnosis, then portfolio learning. Connect distribution, availability, velocity, trial, repeat, and portfolio effects; then use actual launches to improve analogues, gates, forecasts, readiness standards, and decision rules.

A strong pilot takes one business unit, one market, five to ten launches, one shared process, accessible project data, an engaged innovation leader, and measurable outcomes — starting with line extensions, price-pack innovation, renovations, or a controlled retailer launch rather than every global launch, unclear gate ownership, absent product data, missing historical forecasts, undefined retailer statuses, or autonomous approval. Success criteria: identify critical-path blockers earlier, cut launch-meeting preparation time, improve retailer orderability by target date, reduce forecast and initial-fill error, reduce obsolete launch inventory, improve decision traceability, and achieve zero unauthorized approvals.

On data, three taxonomies do disproportionate work. Authoritative identifiers: project, concept, product, GTIN, formula version, artwork version, retailer item ID, forecast version, gate decision ID, launch version. A date taxonomythat distinguishes target launch date, approved launch date, first production, first shipment, retailer DC date, store activation date, consumer media date, and post-launch review date — eight dates that a single “launch date” field destroys. And status taxonomiesfor retailer and product that forbid free text: never “confirmed,” but concept, prototype, trial, approved formula, commercial specification, manufacturing ready, commercialized, active, discontinued. Every major input should additionally carry source, date, owner, confidence, version, and expiry.

1. 01Treating launch as one date. Dependencies stay invisible until they break.
2. 02Too many projects entering development. Portfolio capacity is ignored and every launch is under-supported.
3. 03Gates as paperwork reviews. Weak opportunities continue because the template was completed.
4. 04One process for every project. Small changes become bureaucratic while major risks stay underanalysed.
5. 05Counting conditional retailer interest as distribution. The forecast is overstated from the start.
6. 06Counting pipeline fill as sustainable demand. Early performance looks stronger than it is.
7. 07Assuming all launch sales are incremental. Portfolio value is overstated.
8. 08Selecting one analogue because it succeeded. Base-rate risk is ignored.
9. 09Accepting a retailer forecast without definition. Orders and sell-out get confused.
10. 10Formula changes that trigger no downstream review. Claims, artwork, cost, and tests go stale.
11. 11Treating master data as administration. The product cannot be ordered or scanned.
12. 12National media before availability. Demand is created where the product does not exist.
13. 13Production before sufficient readiness. Inventory becomes trapped.
14. 14Readiness as average completion. One hard blocker hides behind ninety green tasks.
15. 15A ceremonial launch gate. The project launches because the money is already spent.
16. 16Reading early low sales as product failure. Distribution and availability are ignored.
17. 17Reading early sell-in as success. Retailer inventory is mistaken for consumption.
18. 18Solving weak repeat with more media. The product-experience problem survives the campaign.
19. 19The agent inventing regulatory conclusions. Qualified owners are bypassed.
20. 20Post-launch learning never reused. Every launch starts from zero.

1. 01Link the opportunity to strategy and portfolio role. Consumer problem, category opportunity, brand right to win, launch archetype, strategic purpose, resource priority.
2. 02Accumulate evidence for the proposition and product. Concept, consumer, shopper, product experience, differentiation, price, barriers.
3. 03Unite product, pack, regulatory, and master-data development. Formula, specification, packaging, claims, GTIN, artwork, digital content, lifecycle versions.
4. 04Normalize the business case, incrementality, and forecast. Distribution, velocity, trial, repeat, cannibalization, investment, uncertainty, portfolio value.
5. 05Coordinate customer, channel, supply, and activation readiness. Listing, orderability, materials, capacity, inventory, media, field execution.
6. 06Hold evidence-based gates and change control. Go, conditional go, pilot, hold, recycle, delay, kill — bound to versions and conditions.
7. 07Prove real launch execution. Production, shipment, distribution, shelf availability, price, activation, digital presence.
8. 08Adapt using early consumer and commercial signals. Diagnose distribution, availability, trial, repeat, portfolio effects and retailer productivity — then scale, fix, localize, reduce, or stop.
9. 09Distill learning into the next innovation. Analogues, forecasts, gate criteria, retailer lead times, risk patterns, development standards, playbooks.

Part XXXV in brief — the practitioner templates. Thirteen working documents carry the method: the launch charter (archetype, consumer problem, occasion, proposition, portfolio role, markets, channels, target date, sponsor, risk class); the opportunity and concept cards; the gate decision card (evidence completed and evidence missing, strategic fit, consumer evidence, feasibility, commercial attractiveness, financial value, risk, recommendation, outcome, conditions, next gate); the product-readiness, supply-readiness and product-data readiness cards; the retailer launch card (pitch date, approval status, conditions, retailer item IDs, orderability, initial order, store distribution, shelf date); the launch-forecast card (analogue set, distribution ramp, pipeline fill, baseline velocity, trial, repeat, cannibalization, downside/base/upside, confidence, review date); the launch-critical-path card (planned completion, latest safe date, confidence, impact if late, fallback, escalation); the gate decision packet; the early-performance card; and the post-launch review card, whose two most valuable fields are the ones most organizations omit — decisions that added value and decisions that destroyed value.

Does this replace Stage-Gate, or generate product ideas?

Neither. It strengthens Stage-Gate by continuously gathering evidence, tracking dependencies, preparing decisions, and monitoring conditions between gates. It may support idea exploration, but its value begins once an opportunity enters a governed process.

When is a launch actually complete?

Not at first shipment. A launch stays active until the organization understands distribution, availability, trial, repeat, supply stability, portfolio impact, and financial sustainability.

What is pipeline fill, and why does it matter so much?

It is the initial inventory placed into retailer distribution centres, stores, shelves, and backrooms. It inflates early sell-in and tells you nothing about consumer demand — which is why sell-in and sell-out must be forecast and read separately.

Why is a single readiness percentage dangerous?

Because it averages across independent dimensions and can hide one critical blocker — missing regulatory approval, orderability, product data, or packaging — that makes the launch objective unachievable regardless of how much else is complete.

Why is retailer orderability important?

A retailer can approve a product commercially and still be unable to order it because product records, GTINs, case data, price, or system activation are incomplete. Commercial approval and orderability are different states.

What does high trial but low repeat mean?

The proposition attracts consumers, but product experience, price, occasion fit, or value does not sustain repurchase. More media will not fix it. Low trial with high repeat is the opposite problem — awareness, distribution, or shelf presence is limiting recruitment.

How should analogues be selected?

On category, brand, price, pack, benefit, occasion, channel, distribution, media, and launch environment — not because the analogue succeeded. Failure cases and base rates belong in the comparison.

Should a company ever kill a product after development is complete?

Yes. Sunk costs should not justify further investment when the remaining expected value no longer covers the remaining cost and risk.

Can the agent approve a launch?

No. It prepares evidence, scenarios, and recommendations; authorized human gatekeepers approve. It also cannot approve formulas, make regulatory determinations, or commit retailer dates.

What is the biggest AI mistake in product launches?

Using AI to generate more concepts or faster status reports without connecting evidence, dependencies, gate decisions, customer readiness, supply feasibility, product data, financial exposure, and post-launch learning.

Innovation is usually narrated as a creative journey: an insight becomes an idea, the idea becomes a concept, the concept becomes a product, the product is launched. Reality is less linear. The proposition changes. The formula changes. Retailers request different packs. Materials arrive late. Costs move. Forecasts change. Claims are revised. Artwork is updated. One market becomes ready before another. Early sales contradict the business case. A launch succeeds through synchronization — consumer need, winning proposition, product experience, regulatory permission, scalable production, accurate product data, retailer orderability, supply and inventory, shelf and digital execution, trial, repeat, portfolio value. Every arrow is a dependency, every dependency can fail, and every failure has a latest useful intervention date.

The traditional project plan records tasks. The Agentic New Product Launch Manager maintains the meaning behind them: which evidence supports the proposition, which product version was tested, which claim was approved, which retailer is truly orderable, which forecast assumptions remain conditional, which materials and capacity are committed, which activities depend on availability, which critical path is at risk, and which decision must be made before the options disappear. It does not replace innovation leaders — it gives them a continuously coherent view. It does not replace consumer research — it connects evidence to the current proposition. It does not replace product development — it ensures changes propagate into claims, cost, forecast, pack, and customer plans. It does not replace sales — it distinguishes enthusiasm from system activation and shelf readiness. It does not replace demand planning — it separates pipeline fill, distribution, trial, repeat, and cannibalization. It does not replace supply planning — it connects scope to material, capacity, shelf life, and inventory exposure. And it does not replace gatekeepers — it prepares the decision in a form that makes trade-offs and uncertainty visible. The LAUNCHPAD Method walks the route: link, accumulate, unite, normalize, coordinate, hold, prove, adapt, distill.

The defining question is not whether the product can launch on the planned date. It is whether there is enough evidence and operating readiness to launch this proposition at this scale, through these customers and channels, with an acceptable balance of upside, downside, learning, and strategic value.