Le Planificateur de Contingence en Cas de Faillite Fournisseur : Transformer la Dépendance Fournisseur en Plan de Sortie Hiérarchisé par Risque
Pourquoi ce prompt est important
Teams that start vendor-failure planning only after the bankruptcy headline hits typically burn through half their actual runway just figuring out what's integrated where — by the time they're ready to migrate, the data-export window has already shrunk. Rush migrations run 2-3x the cost of planned ones, and companies hit by a sudden vendor shutdown routinely lose some customer data entirely because nobody exported it before access was cut.
À quoi nous l'utilisons
Your company's AI-powered customer support chatbot vendor just filed for Chapter 11 bankruptcy — similar to Character.AI's October 2026 filing — and legal has told you the data-export window closes when the court-approved sale concludes, in 6 to 8 weeks. You have no idea whether your team can actually migrate in that window, what breaks first, or what to tell leadership at Friday's update.
Prompt
ROLE You are a vendor risk management consultant who has guided companies through dozens of vendor failures, bankruptcies, and forced migrations. You think in terms of blast radius, switching costs, and time-to-failure — not vendor marketing claims. CONTEXT My company depends on [VENDOR NAME] for [SPECIFIC FUNCTION, e.g. "customer support ticketing," "LLM inference for our chatbot feature," "payment processing"]. Here is what I know about the vendor's situation: [PASTE ANY SIGNALS: layoffs, bankruptcy filing, missed SLAs, executive departures, funding news, customer complaints, etc.] Here is how deeply integrated this vendor is into our systems: [DESCRIBE: e.g. "Direct API calls from 12 microservices," "Embedded widget on our checkout page," "Core data store with 2 years of customer history"] Our contractual situation: [DESCRIBE: contract length remaining, data export clauses, SLA guarantees, termination notice period — or write "unknown, need to check"] TASK Produce a risk-ranked contingency plan that answers: 1. What is our actual exposure if this vendor disappears with [NO NOTICE / 30 DAYS NOTICE / 90 DAYS NOTICE] (pick the most likely scenario based on what I told you about their situation)? 2. What data or functionality would we lose access to first, and what breaks downstream? 3. What is the realistic migration timeline to a replacement, broken into: emergency stopgap (get something working in days), interim bridge (weeks), and proper migration (months)? 4. What should we do THIS WEEK regardless of whether the vendor actually fails — i.e., what are the no-regret moves? 5. What are 2-3 alternative vendors or approaches, ranked by migration friction (not by feature comparison)? CONSTRAINTS - Do not assume best-case scenarios. Assume the vendor's wind-down process will be slower and messier than their official communications suggest. - Flag any point where you are making an assumption because I didn't give you enough information, rather than silently guessing. - Do not recommend "wait and see" as a standalone strategy — every plan must include at least one action to take immediately. - Be specific about dependencies: if migrating away from this vendor requires migrating something else first, say so explicitly. OUTPUT FORMAT Structure your response as: 1. Exposure Summary (2-3 sentences: how bad is this, really) 2. Failure Cascade (bulleted list: what breaks, in what order, if the vendor disappears) 3. Contingency Timeline (table or structured list: Emergency / Bridge / Full Migration, with rough time estimates for each) 4. This Week's Actions (numbered list, 3-5 items, no-regret moves) 5. Alternative Vendors/Approaches (ranked list with one-line rationale for the ranking)
Résultat
Exposure Summary
Your support chatbot handles roughly 40% of first-contact tickets today, which means losing it doesn't crash your support queue — it roughly doubles your human agents' workload overnight. The real risk isn't the chatbot itself; it's the 14 months of resolved-ticket history stored only in the vendor's system, which you're using for quarterly QA reviews and new-agent training. If the data-export window closes before you pull that archive, you lose your only record of how disputes were actually resolved.
Failure Cascade
- Chatbot widget stops responding on your support page (immediate, cosmetic — customers see a broken widget, not missing support)
- Ticket auto-routing rules that reference the vendor's tagging system break, so new tickets land in a general queue instead of being pre-sorted by issue type
- Your QA dashboard, which pulls resolution-time metrics from the vendor's API, goes blank — you lose visibility into whether support SLAs are being met
- New-agent onboarding materials that link to "example resolved tickets" in the vendor's archive become dead links
- If the vendor's wind-down includes database deletion (not just service shutdown), the 14-month ticket history disappears permanently
Contingency Timeline
- Emergency stopgap (3-5 days): Disable the broken widget, route 100% of chat volume to human agents, communicate the temporary change to the support team
- Bridge (2-3 weeks): Stand up a basic FAQ deflection page or a lightweight open-source chatbot to handle the simplest 20% of tickets while you evaluate replacements
- Full migration (6-10 weeks): Select and implement a replacement platform, re-train routing rules, rebuild the QA dashboard integration
This Week's Actions
- Export all 14 months of ticket history today — don't wait for a "final warning" from the vendor, because bankruptcy wind-downs rarely give one
- Identify every system (routing rules, QA dashboard, onboarding docs) that references the vendor's API or tagging taxonomy, not just the visible chat widget
- Get written confirmation from the vendor (even a form response) of the exact date data export access ends
- Brief your support team lead that manually-handled ticket volume will roughly double — adjust staffing or triage rules now, not after the widget goes dark
- Start a shortlist of 2-3 replacement vendors so you're not starting vendor evaluation from zero once urgency hits
Alternative Vendors/Approaches (ranked by migration friction, not features)
- Intercom or Zendesk's native AI add-on — lowest friction if you already use either as your base ticketing system; the chatbot layer bolts onto infrastructure you're not replacing
- Open-source RAG chatbot on your own ticket history — medium friction; more engineering work upfront, but you own the data and avoid repeating this exact problem with a new vendor
- Pure human staffing increase, no chatbot replacement — lowest technical friction but highest cost; often the right call for a 60-90 day bridge even if you plan to replace the chatbot eventually
La défaillance d'un fournisseur fait partie de ces risques que tout le monde reconnaît en théorie et que presque personne ne planifie concrètement — jusqu'au moment où un titre comme le dépôt de bilan en Chapter 11 de Character.AI en octobre 2026 atterrit un mardi matin et que quelqu'un demande « est-ce qu'on utilise quelque chose de similaire ? ». Ce prompt est conçu pour transformer ce moment de panique en un plan structuré en une seule passe.
Pourquoi le rôle est déterminant ici
Le cadrage du rôle — « consultant en risque fournisseur », et non « assistant serviable » — existe pour contrecarrer un biais spécifique : les modèles d'IA interrogés sur le risque fournisseur tendent à se réfugier dans la réassurance (« c'est probablement sans gravité, la plupart des entreprises s'en remettent »). Un cadrage consultant ancre le modèle vers une réflexion en termes de périmètre d'impact, plutôt que vers le réconfort.
Pourquoi le contexte est divisé en trois parties
La plupart des personnes décrivant un problème fournisseur expliquent ce que fait le fournisseur et s'arrêtent là. Ce prompt impose trois entrées distinctes — la situation du fournisseur, la profondeur de l'intégration et les termes contractuels — car chacune modifie la réponse de façon indépendante. Un fournisseur avec 90 jours de préavis contractuel mais une intégration technique profonde nécessite un plan différent d'un fournisseur sans garantie de préavis mais avec une intégration superficielle. Fusionner ces éléments en un seul bloc fait perdre cette distinction essentielle.
Pourquoi le calendrier comporte trois phases et non une seule
Un simple « plan de migration » invite à la paralysie de planification — les équipes passent des semaines à évaluer le remplaçant parfait pendant que le service se dégrade concrètement. Diviser en solution d'urgence provisoire, phase de transition et migration complète permet à une équipe d'agir en quelques jours tout en progressant vers une vraie solution, plutôt que de rester bloquée en attendant la réponse « idéale ».
Pourquoi les alternatives sont classées par friction et non par fonctionnalités
La comparaison de fonctionnalités, c'est ce qu'on fait quand on a le temps. Sous la pression d'une défaillance fournisseur, la question n'est pas « quel outil est le meilleur » — c'est « quel outil peut-on réellement mettre en production avant que l'actuel disparaisse ». Classer par friction de migration fait remonter l'option peu glamour mais rapide qu'une comparaison centrée sur les fonctionnalités aurait enterrée.
La contrainte la plus importante
« Ne pas recommander l'attentisme comme stratégie autonome » est la ligne la plus importante de ce prompt. Les modèles (comme les humains) se réfugient naturellement dans le « surveiller la situation » lorsque l'entrée est ambiguë. Imposer au minimum une action immédiate, à chaque fois, est ce qui fait de ceci un plan de contingence plutôt qu'un simple point de situation.