×
Cleared for Takeoff: Jeppesen's Phased AI Support Rollout
Aviation leaves no room for error. Maven AGI brings enterprise-grade AI to the Jeppesen ForeFlight support team — embedded directly in Zendesk, grounded in your knowledge, and built to make every interaction faster and more accurate. We start with ForeFlight, expand to Jeppesen, and then bring the same platform to the eCommerce team. From there it extends to every team through our Salesforce integration — one platform, unique systems and knowledge, a unified experience.
Ultimately, we plan to drive more efficiency in support with millions saved by reducing the cost per ticket by ~30%.
Executive Summary
Jeppesen ForeFlight's Tier 1 Customer Support team supports a global pilot community with high-stakes, technical, accuracy-mandatory questions. This proposal frames a two-phase rollout: Phase I starts with ForeFlight email and Copilot in Zendesk; Phase II expansion covers both eCommerce and Jeppesen across Salesforce web, email, phone, chat, and voice. The same platform, pricing, and deployment model scales across all 339,471 annual interactions.
6.1x ROI · 1.7 month payback
$30.00 → $21.65 across 339,471 annual interactions
At the current 40.0% autonomous resolution rate, the three-phase rollout projects $3,298,230 in total gross savings against a Maven investment of $465,221, yielding a 6.1x ROI and $2,833,009 net annual benefit. The all-in cost per ticket drops from $30.00 today to $21.65 after Maven — a 27.8% reduction across 339,471 annual interactions.
This implementation will have the following impact on the organization:
Sub-2-minute resolution on Maven-handled tickets + intent/complexity routing so simple work stops competing with complex cases in the same queue.
Graph of Record kills the Center tab-switching workflow; grounded knowledge replaces ad hoc lookups. Routing lets staffing be planned by intent category.
Copilot-assisted onboarding compresses ramp time and widens the hiring pool, so Tier 1 hires don't need deep aviation expertise going in.
Ultimately, this partnership means increasing the efficiency of the support team without changing the existing technology footprint, knowledge base, and SOPs. It is an intelligence layer to reduce total cost of ownership.
Discovery
What we heard from Jeppesen ForeFlight. Learnings from meetings with the support team — this is the state of play today and where the leverage is.
Phase I + II + III interactions/year
Blended fully loaded cost today
339,471 × $30.00 = $10.18M/year
Every inbound email — a 30-second password reset or a 2-hour weight-and-balance calc for a Gulfstream — lands in the same queue and gets the same agent attention until a human sorts it out. No automated routing today by complexity or customer value.
No systematic way to flag high-ARR or VIP accounts (e.g., a $2.7M ARR customer) for priority handling in Zendesk. A NetJets chief pilot and a GA hobbyist asking the same question get differentiated treatment only when an individual agent recognizes the name — judgment, not system logic.
Agents tab-switch out of Zendesk into Center to pull account details — copying an email, searching, clicking through tabs to see plan status, purchase history, and usage. A prior sync attempt pulled day-old data, which was useless for subscription questions where customers write in immediately after a change.
~80–85% of email volume is unauthenticated — customers have been trained for years to email team@foreflight.com. That limits auto-personalization and blocks pulling account context automatically at the point of contact.
The pilot's guide is a 1,200-page PDF. Keyword search doesn't reliably surface answers, so agents fall back to ad hoc tools — Grok, ChatGPT, Google — adding time and inconsistency to every non-trivial ticket.
Recurring examples with no real decision-making: blank "sent from my iPhone" emails, data-sync toggle checks (including a military-account edge case), and "thanks, that solved it" replies that still require a human to close. Steven estimated these categories alone represent meaningful daily agent time across ~14,200 monthly emails.
Amplitude isn't connected to the support workflow, so there's no way to identify low-engagement customers (e.g., someone who's never used daily weather or logbook) and proactively nudge them into stickier usage — directly tied to renewal risk (flight-plan filers renew at ~98% vs. ~50% for non-filers).
When a customer emails team@foreflight.com or writes into Jeppesen ForeFlight, there's no reliable way to know which of the ~80 Jeppesen ForeFlight products they're asking about without an agent reading the message and pattern-matching. That guess drives routing, knowledge lookup, and SLA — and it happens by hand on every ticket.
Jeppesen ForeFlight's own Support SME screened 48,088 Human/Unclassified tickets from February 2026–July 2026 and independently identified nine recurring, addressable themes — 20.2% of volume, ~9,708 tickets. The slowest-resolving tickets concentrate in generic or unclassified subject lines, not aviation-complex issues — confirming the triage/routing gap described elsewhere in this section.
73.1% of the screened ticket population resolves in under 3 hours — strong evidence that a large share of volume is low-complexity, repeatable work well within Maven's autonomous resolution range. The remaining 26.9% is where Copilot-assisted handling and human expertise add the most value.
The Jeppesen Tier 1 transition (Salesforce → Zendesk) goes live mid-to-late September. It's not fast enough to be a hard trigger event, but it's real leverage: this team is absorbing new products cold, so AI backstop plus Copilot-assisted onboarding is a genuinely strong wedge, not just a nice-to-have.
Every inbound email — a 30-second password reset or a 2-hour weight-and-balance calc for a Gulfstream — lands in the same queue and gets the same agent attention until a human sorts it out.
The transition and the mandate make this time-bound
The Jeppesen Tier 1 transition and the Thoma Bravo mandate make this a time-bound conversation — AI adoption is happening one way or another.
Jeppesen Tier 1 is joining the ForeFlight Zendesk team
The Jeppesen Tier 1 (Salesforce) team is being absorbed by the ForeFlight Zendesk team. Training is in progress now, with go-live planned mid-to-end of next month.
AI help lands right after the transition
Maven won't be in place before the Zendesk team absorbs the volume, but it becomes valuable immediately after: new products, new systems, and a 1,200-page manual to master.
Thoma Bravo's AI adoption mandate
AI adoption is a company-wide North Star, moving fast. This is bigger than a support-team initiative — it's the direction the whole organization is being asked to take.
Maven AGI Overview
Maven AGI is an enterprise AI platform built specifically for customer support. It's made up of two products that can be deployed independently or together: Maven Copilot, which assists human agents inside the ticketing systems they already use, and Agent Maven, which resolves customer issues autonomously across chat, email, voice, web, SMS, Slack, and in-app channels. Both are powered by the same underlying reasoning engine, and both connect to the systems and knowledge sources the team already has in place.
Maven Copilot: helps agents work faster
Maven Copilot lives natively inside Zendesk, Salesforce, HubSpot, and other ticketing platforms. It drafts suggested replies, summarizes long tickets, surfaces the right knowledge article at the right moment, and pulls in context from connected systems — all without taking the agent out of their existing workflow. The agent stays fully in control of every customer interaction.
Agent Maven: resolves customer issues autonomously
Agent Maven is a customer-facing AI agent that resolves issues end-to-end on the categories of questions the team has approved. It operates across chat, email, voice, web, SMS, Slack, and in-app — wherever the team chooses to deploy it. For everything outside its approved scope, it escalates cleanly to a human, with full conversation context attached so the agent never starts from zero.
Actions: not just answers
Maven doesn't just answer questions — it can take real actions in the team's existing systems. Through native integrations and the Maven App Marketplace, Maven can do things like look up account status, reset a password, check an order, update a record, or trigger a workflow in any connected backend. What Maven can do is determined by which Actions the team configures and approves.
Maven works alongside existing teams, not in place of them. The starting point is making the team more effective at the work they already do — and expanding from there at whatever pace the team chooses.
Why Maven AGI
The technology that makes 'one front door' safe to actually ship — not just safe to talk about.
Real connectors — not scripts or fake integrations. No middleware, no work from the Jeppesen side to build or maintain them. 100+ pre-built connectors read and write across Salesforce, Zendesk, Snowflake, Jira, Slack, and more.
Maven doesn't guess. Every response is grounded in your Salesforce records, KB, and contracts — and escalates when unsure.
Sub-2-minute average resolution on tickets Maven handles. Consistent SLA regardless of the day's ticket mix.
Agent Studio lets ops teams build and tune agents without engineering. Full SDK for custom application building when you need it. Easy to extend, easy to operate.
You get a team, not a login. Category-by-category rollout, live tuning, and dedicated implementation support.
Knowledge, agents, analytics, and integrations on a single platform — no stitching, no separate vendors, one Graph of Record.
Continuous Learning
Every interaction makes Maven smarter — and Copilot is the learning layer. Every agent-edit, accepted draft, and escalated ticket feeds back into the knowledge graph so the AI improves on the categories Jeppesen ForeFlight cares about most.
Agents and customers interact across channels.
Maven Inbox surfaces recurring intents and knowledge gaps.
Suggested KB articles generated from real conversations.
Updated knowledge improves future answers automatically.
A continuous loop — every conversation improves the next.
- The team always stays in control. Suggested content is reviewable, editable, and approved before it goes live.
- No new tools to learn. Knowledge improvements flow into the systems agents already use.
- Improvements compound. Every conversation makes the next one better.
- Copilot accelerates learning. Agent edits and accepted drafts on Copilot-resolved tickets become the next training signal, so autoresponse gets smarter faster without manual tuning cycles.
Trust and control
Jeppesen ForeFlight's support requires precision and control — so Maven was built for that.
Questions about account deletion, billing disputes, or any regulatory matter always escalate to Zendesk automatically.
Maven only answers when it's certain. Otherwise, it routes to an agent.
Every AI response shows which knowledge was used and why.
Every interaction is logged, reviewed, and explainable.
Why it works for Jeppesen ForeFlight
Each capability removes a specific way the current support operation makes work harder — and replaces it with something rule-based, not judgment-dependent. The same capabilities scale across the phased rollout: ForeFlight first, then the eCommerce team, then Jeppesen — one platform, unique systems and knowledge, a unified experience.
Intent- and complexity-based routing
Segments commit each conversation to a purpose-built path before triage reaches a human. Password reset ≠ weight-and-balance calc — routed differently, deterministically. SLA stops depending on the day's ticket mix.
Tiered-account segments
The $2.7M ARR account gets flagged and routed by rule, not by whether the agent on shift recognizes the name. Revenue-at-risk protected by system logic, not tenured memory.
Segment isolation across Jeppesen ForeFlight
One Graph of Record, one reasoning engine — but segment-specific knowledge and general knowledge never bleed into each other's answers. The technical answer to 'one front door' without cross-segment contamination.
Graph of Record replaces tab-switching
Plan, purchase history, and usage from Center — scoped to the conversation and available in the same interface. This is the same mechanism behind the 25–40% agent productivity proof point.
Unified knowledge, scoped by segment
The 1,200-page pilot's guide, help center, and Jeppesen ForeFlight docs live in one graph. Segments decide what's eligible per conversation — unification without new risk.
Copilot-safe internal vs. customer knowledge
Internal runbooks visible when Maven assists an agent, invisible when it talks directly to a customer. The same investment funds Copilot and autonomous resolution — no separate build.
Product identification across 80+ Jeppesen ForeFlight SKUs
Simplest path: Maven injects a first message asking which product the customer is inquiring about — and if that's already on the customer record in Salesforce, we skip the question and use it. For the long tail of 80 Jeppesen ForeFlight products, Intelligent Fields classify the product at every turn of the conversation and confirm before escalating. One capability handles the whole catalog — no per-product build.
Implementation
Three phases: CoPilot + email validation first, then Center integration and actions, then expansion scoping once ForeFlight proves value.
Weeks 1–8. Kickoff and success metrics; core knowledge ingestion across Help Center, product guides/PDFs, Confluence, guardrail guides, and curated sample tickets with per-product tagging and validation; CoPilot in Zendesk (Marketplace app) as the validation engine; Maven email auto-response enabled category-by-category as CoPilot validates; Center API discovery; architecture/security review kickoff; product-disambiguation groundwork (glossary + intelligent fields).
Weeks 10–16. Center read connector (ForeFlight data) per Phase 1 discovery; account and entitlement lookups; first write action (nav-data counter reset) behind accuracy gates with confirmation notifications; conversations unified by user identity across surfaces; additional email categories enabled; customer-facing chat validated in parallel behind login; Amplitude usage-data availability explored alongside chat.
Weeks 18+. Scoping memos only — no delivery commitment. After ForeFlight proves value across CoPilot, email, and Center actions, Maven expands into Jeppesen-branded teams — starting with Zendesk-using groups, then Jeppesen's Salesforce environment via native integration, unifying both ticketing ecosystems under one shared source of truth. Candidates also include additional read/write actions, proactive feature-adoption, voice self-service, and in-app ForeFlight chat. Any promotion to build requires a separate SOW amendment.
POC Validation
Proof-of-concept results on real Jeppesen ForeFlight Tier 1 tickets — see the conversation, the demo, and the accuracy we achieved.
POC results on real Jeppesen ForeFlight Tier 1 tickets
With just base knowledge and limited tuning, Maven was able to answer 85 out of 100 real Tier 1 tickets with high accuracy, sourced from live Jeppesen ForeFlight volume.
Tier 1 questions answered with high accuracy based on limited knowledge ingest.
- •Test set: top 100 Tier 1 questions from real ticket volume
- •Approach: iterative tuning rounds — test, identify misses, tune knowledge and Charters, retest
- •Validation: regression run after each pass to protect previously passing answers
The live test set used for each tuning pass: 100 real Tier 1 questions from Jeppesen ForeFlight ticket volume, re-run after every knowledge or Charter change to catch regressions on previously passing answers.
See Maven running against Jeppesen ForeFlight's exact scenarios
A walkthrough covering the current-state gaps we heard — triage, VIP segmentation, 80-product classification, entitlement fixes, and low-engagement nudges — end to end in one interface.
Maven pulls product and account context from Salesforce before the user types a second message — no product picker, no clarifying round-trip.
Watch classification update turn-by-turn as the conversation evolves, with a confirmation gate before any escalation or entitlement action.
Maven resolves a subscription mismatch against Center in-flow — no handoff, no ticket, agent stays on the high-value queue.
Amplitude signal ties into the workflow: Maven surfaces a proactive nudge for a user who's never touched daily weather or logbook.
Teams already doing more with Maven
"The question wasn't 'can it deflect tickets,' it was 'can it deliver an interaction our customers will be genuinely glad they had.' AskQ cleared that bar and kept raising it."
"Maven runs agents across Support, Cross-sell, Upsell, and Onboarding for Clio — same reasoning engine, same interface, no separate stack per use case. 80% of inquiries answered autonomously."
"The Copilot pattern that reduced onboarding time by nearly 90% at ClickUp — the same shape we're proposing for Jeppesen ForeFlight."
"In a world oversaturated with AI support solutions, Maven is second to none."
Platform fee, usage, and what counts as a billable resolution
One fee for the whole footprint. Does not scale with volume, teams, or channels added.
Charged when Phase II expansion launches (Jeppesen).
- ✓Unlimited agents, teams & channels
- ✓100+ integrations, no per-connector fee
- ✓Agent Designer + Analytics
- ✓SSO, RBAC, SOC 2 / HIPAA controls
- ✓Named CSM + implementation team
One flat rate — no blended cost structure, no volume tiers. Voice is priced separately at $2.50 per conversation.
Billed per conversation — not per minute.
Platform fee plus $1.00 per resolution for digital surfaces, $2.50 per voice conversation.
Resolution = a customer issue that Maven AI handles end to end without human involvement. The customer’s problem must be genuinely solved — not merely acknowledged, answered with a link, or routed away from a queue.
- The customer’s need is understood and addressed completely.
- Any required action is completed, such as processing a refund, updating an account, changing an order, or answering the question fully.
- No human agent needs to take over the interaction.
- The outcome is verified, ideally through customer confirmation and/or no repeat contact about the same issue. Maven’s resolution guidance uses a 48-hour no-follow-up window; teams may also monitor re-contacts within 72 hours as an added durability check.
Billed at $1.00.
Resolution = a customer issue that Maven AI handles end to end without human involvement. The customer’s problem must be genuinely solved — not merely acknowledged, answered with a link, or routed away from a queue.
- The customer’s need is understood and addressed completely.
- Any required action is completed, such as processing a refund, updating an account, changing an order, or answering the question fully.
- No human agent needs to take over the interaction.
- The outcome is verified, ideally through customer confirmation and/or no repeat contact about the same issue. Maven’s resolution guidance uses a 48-hour no-follow-up window; teams may also monitor re-contacts within 72 hours as an added durability check.
Billed at $1.00.
1 ticket = 1 billable resolution. Every ticket where Copilot generates a response counts as one billable resolution at $1.00.
- •Research tab: counts as its own, independent resolution.
- •Thumbs down: not billed.
Billable unit = one voice conversation the AI answers and handles, charged at $2.50 per conversation regardless of call length. A single conversation is billed once even if it includes multiple intents; calls transferred to an agent are still one conversation, and no separate resolution fee applies.
Rates shown are placeholders pending confirmation with Maven RevOps. Annual contract, paid upfront.
Phased Investment & Return
Two phases — Phase I for ForeFlight email and Copilot validation, and Phase II expansion for both eCommerce and Jeppesen. Cost scales with usage; savings scale with resolution volume. The platform fee and usage are contracted annually and paid upfront, not month-to-month.
Jeppesen Expansion
eCommerce Expansion
Detailed gross savings scenarios
All figures annualized. Modelled on 120,000 annual EMAIL tickets at $30/ticket fully loaded. Automation targets the ~40% Tier 1 commerce bucket; whatever Tier 1 tickets automation doesn't resolve are Copilot-assisted alongside the ~60% Tier 2 aviation-complexity bucket. Every ticket lands in exactly one pool — no double counting.
This Phase 1 ForeFlight ROI work is the basis for projecting savings across Phase II expansion (eCommerce and Jeppesen), using the same assumptions, cost model, and resolution methodology.
Two distinct pools drive the model: a Tier 1 commerce bucket that can be automated, and a Tier 2 aviation-complexity bucket that Copilot handles. Every ticket lands in exactly one pool — no double counting.
This data is based on conversations with Steven Roth.
The live Autonomous Resolution Rate plus four fixed reference scenarios, all applied to the same 120,000 annual email volume. Every Maven-handled ticket is a billable resolution at the flat $1.00 price. Autonomous resolutions on the Tier 1 pool get full $30 deflection credit; every remaining ticket is Copilot-assisted and credited $5.00. The 40.0% lever moves savings and net benefit only — total cost and the per-resolution price stay flat.
% of the 40.0% Tier 1 pool Maven resolves autonomously = 19,200 tickets/year
| Line item | Current — 40.0% | Conservative | Target | Growth | Full integration |
|---|---|---|---|---|---|
| 1. Tier 1 commerce support — autonomous resolution | |||||
Gross automation savings Resolved tickets × $30 fully loaded | $576,000 | $720,000 | $936,000 | $1,152,000 | $1,296,000 |
| 2. Copilot-assisted value (Tier 2 + leftover Tier 1 — non-overlapping with auto-resolved) | |||||
Total Copilot-assisted value Every ticket not autonomously resolved is Copilot-assisted: billed the same $1.00, credited $5.00 per ticket ($30 × 20.0% / 1.2). No double-counting: each ticket sits in exactly one bucket. | $504,000 | $480,000 | $444,000 | $408,000 | $384,000 |
| 3. Maven cost (flat $1.00 per digital resolution + platform fee) | |||||
Total Maven investment Resolution usage (flat $1.00 × all Maven-handled digital resolutions) + platform fee. The slider changes the autonomous/assisted split, not the total resolution count or price per resolution. | $170,000 | $170,000 | $170,000 | $170,000 | $170,000 |
| 4. Total business case | |||||
Net annual benefit Net annual benefit = gross savings − Maven investment. | $910,000 | $1,030,000 | $1,210,000 | $1,390,000 | $1,510,000 |
Headcount avoidance, not headcount reduction
The savings above are framed as capacity the team won't have to add as volume grows and the Jeppesen Tier 1 workload rolls in — not as a plan to cut existing roles. The FTE-equivalent figures represent hires that become unnecessary, not people to be removed. That's the defensible way to position the number to Sean.
Beyond deflection dollars, three benefits Steven surfaced on the call that make the price defensible on their own.
Steven referenced ClickUp's Copilot rollout as the proof point — agent onboarding dropped from 9 months to 1. The same shape is on the table for Jeppesen ForeFlight's agent ramp, directly compressing the largest fixed cost in support.
Called out as a benefit in its own right. Agents pick it up without a training curve, and the same interface serves Copilot today and autonomous resolution tomorrow — no re-platforming between phases.
Copilot means Jeppesen ForeFlight can unlock hiring at a lower skill tier for Tier 1 work — not needing Tier 2-level domain expertise for Tier 1 volume. That's where the actual savings show up. Automation means they won't need to hire at the same rate as volume grows.
Total email volume (120,000) and $30 fully loaded cost/ticket are the two fixed inputs driving every scenario; the lever controls the share Maven resolves. Voice tickets are explicitly excluded from this model.
What this means for workforce management
The $2.75M–$3.35M modeled benefit is avoided headcount growth, not attrition — capacity that scales with incoming Jeppesen Tier 1 volume without a proportional hiring plan. The ticket-screen cross-check (~$232K–$459K) gives a conservative, already-validated floor to build into capacity forecasts.
Sub-2-minute average resolution on Maven-handled tickets drives a major average handle time reduction on the automatable segment. Intent- and complexity-based routing keeps SLA consistent regardless of daily ticket mix — a 30-second reset no longer competes with a 2-hour weight-and-balance calc for the same queue priority.
Graph of Record removes the Center tab-switching workflow, and grounded knowledge replaces ad hoc lookups in ungoverned tools like Grok, ChatGPT, or Google — cutting off-system time spent hunting for answers.
Agent ramp compressed from 9 months to 1 month (ClickUp reference) — the single biggest lever on training cost and new-hire productivity curve. Copilot also enables hiring at a lower skill tier for Tier 1 volume, widening the labor pool and lowering cost-per-hire.
Deterministic routing means volume can be forecasted and staffed by intent category instead of one undifferentiated queue. POC-validated 85% accuracy gives a known, tunable automation rate per intent to build into staffing models.
Tiered-account segmentation removes reliance on individual agent tenure to recognize high-value accounts — VIP handling is rule-based, not dependent on who's on shift, reducing single-point-of-failure risk from undocumented judgment.
Same business case, viewed through a staffing and operations lens.