A
AIOS Wiki
read-only · public mirror
Open AIOS
Wikiai-supportresearchai-support/research/07-phase-5-scenario-vertical-squeeze.md

SCENARIO: VERTICAL SQUEEZE

Hand-authored·18 min read·13 sections·Last edited May 12 by initial import·View history

Bear Case — The Wedge Compresses From Both Sides

1. Scenario Name & 1-Sentence Thesis

"Vertical Squeeze" — AZ's crypto-vertical wedge gets compressed simultaneously from above (Decagon+Lorikeet+Coinbase ship a regulated-commerce pack) and below (Anthropic open-sources the canonical support-agent loop), while the lighthouse customer (Dapper) deteriorates faster than the product can travel beyond it, leaving AZ-spinout stuck at ~$4M ARR with no expansion vector and an acquihire as the dignified exit.

2. Narrative (looking back from May 2029)

June 2026 — AZ-spinout (operating under the working name "Sentinel") signs Dapper at $320K ACV with lighthouse rights and closes a $5M seed at $22M post from a vertical-SaaS-curious seed firm and Coinbase Ventures (ironic, in retrospect). Headcount: 11. The deck is the Priya Bose deck almost word-for-word — VPC, HITL, MiCA-ready, "AI support that survives a Reuters headline." MiCA full enforcement hits July 1; Sentinel has Travel Rule + OFAC modules shipped two weeks before. Three letters of intent from mid-tier exchanges (Bitstamp, Gemini Europe desk, one undisclosed). Mood inside AZ: vindicated.

Q3 2026 — Sentinel signs Bitstamp ($180K) and a Paxos pilot ($90K, 90-day). SOC 2 Type II report delivered late September (Q3 deadline missed by six weeks — first leading indicator). ARR: $590K run-rate. The "State of Crypto Support" report ships in October — gets cited in The Block, Bits & Bips guest spot — and three inbound demos materialize from it. Meanwhile, Decagon ships "Decagon Verticals" in late October as a beta — fintech and healthtech packs first, crypto teased for Q1.

November 2026 — Decagon acquires Lorikeet for $200M (50% stock, 50% earnout-on-fintech-ARR). Public reframe: "Decagon for Regulated Commerce." Coinbase signs as design partner two weeks later — announced at Money 20/20 — for a $4M annual deal that Decagon books as flagship and probably loses money on. Sentinel's three open enterprise pipelines stall within ten days of the announcement; two procurement teams ask for a Decagon-Sentinel bake-off; the Bullish-tier deal that was 80% closed goes silent.

Q1 2027 — Dapper does its fourth layoff round (28%). Justin survives but Mark Kingston is reassigned to a Flow Foundation devrel-adjacent role, and the new VP CS is a cost-cutter from a Toronto fintech. Dapper renewal conversation gets pushed from June to September and the ACV is openly "under review." Sentinel ARR at end of Q1: $1.9M (Dapper, Bitstamp, Paxos converted to $140K, two small wallet operators at ~$80K each, plus a Polymarket pilot).

Q2 2027 — the second hammer falls. Anthropic ships claude-support-agent at the May 2027 dev day — MIT-licensed reference architecture, canonical agent loop, prompt caching tuned for support traces, Zendesk + Intercom + Salesforce + Freshdesk connectors out of the box, "human-in-the-loop middleware" as a first-class primitive. Three of Sentinel's five "we're differentiated by HITL UX" claims become commodity table-stakes overnight. Hacker News front page two days. Within six weeks, Zendesk ships "Zendesk Native AI Agent" built on top of it; Intercom doubles down on Fin and cuts per-resolution to $0.49.

Q3 2027 — Sierra announces ISO 42001 + HIPAA Enterprise + SOC 2 Type II all in one press release, plus a hire from Chainalysis as "VP of Regulated Verticals." A Sierra crypto case study (an undisclosed top-10 exchange) drops in September. The application-layer differentiation claim is now contested on every dimension Sentinel had — compliance, vertical, HITL UX. Pricing in the category compresses ~25% in two quarters.

Q4 2027 — Disney pulls Disney Pinnacle (unrelated — a corporate restructuring around streaming + theme parks). Top Shot revenue down 41% YoY. Dapper's renewal closes at $180K (down from $320K). Sentinel ARR: $3.2M, NRR 71% on the cohort that renewed, win rate against Decagon dropped from ~35% to ~14% in two quarters. The "Hacker-vs-Victim Disambiguation Layer" gets cloned by Decagon's vertical pack in November (less elegant, but "good enough" per three lost deal post-mortems).

Q1 2028 — Sentinel raises a $4M bridge at flat from existing investors after a failed Series A process (talked to 14 firms, two term sheets at materially down rounds, both pulled). Headcount frozen at 17. Polymarket churns (in-housed). A Circle-tier prospect that had been the bridge thesis chooses Decagon. Founder fatigue visible.

Q3 2028 — Sentinel pivots messaging to "compliance-grade audit layer for AI support" — stops trying to be the agent, tries to become the compliance overlay on top of Decagon/Sierra/Anthropic. Two pilots; the strategic logic is sound but the GTM motion is now selling to platforms, not buyers. ARR plateaus at $4.1M for three quarters.

Q1 2029 — Decagon acquires Sentinel for $24M (60% cash, 40% retention; the comp lead and two engineers get retention packages; the rest get severance). Press release frames it as "deepening Decagon's regulated commerce moat." Realistically: it's a vertical-team acquihire with the customer book as a sweetener. AZ's strategic stake (held in the spinout cap table at 14% post-bridge dilution) returns ~$2.1M cash to Axiom Zen — roughly 0.7x the cash AZ put in, plus the foregone AZ salary-equivalents over three years. The compliance IP and skills-extraction code get folded into Decagon's vertical pack within nine months.

May 2029 — The Sentinel brand is gone. Decagon's Q4 2029 earnings deck name-checks "regulated commerce" as a $40M ARR sub-segment. Justin from Dapper is at a stealth gaming startup. Mark Kingston is on a podcast about why AI support is "fine for the median ticket."

3. Winners & Losers

Winners

  • Decagon — bought the vertical narrative for $200M (Lorikeet) + $24M (Sentinel) + maybe $40M total in vertical-pack engineering. Now plausibly the regulated-commerce category leader heading into a 2030 IPO window.
  • Coinbase — got a first-party-quality support stack at design-partner pricing, uses it as a moat against smaller exchanges who pay 3-5x more for inferior tooling.
  • Anthropic — claude-support-agent becomes the de facto application-layer reference, drives Claude usage in support workloads, and quietly compresses 30+ application-layer competitors.
  • Sierra — keeps hunting elephants, doesn't care about the squeeze, finishes 2028 at $400M+ ARR with a healthy IPO path.
  • Lorikeet founders — clean $200M outcome, retention earnouts intact.
  • Forethought-style acquihire shops — the playbook keeps working; Zendesk and NICE keep buying $20-50M ARR companies for $200-400M.

Losers

  • AZ-spinout (Sentinel) — acquihire'd at $24M after a $5M seed and a $4M bridge. Founders return below preferred liquidation in some scenarios.
  • Axiom Zen parent — strategic stake recovers <1x cash; loses the optionality narrative; "we built a venture-scale company" thesis fails publicly. Internal credibility hit on future spinout pitches.
  • Dapper Labs — keeps shrinking; Sentinel's tooling, ironically, was its best operational leverage and now belongs to Decagon (who Dapper can't afford).
  • Mid-tier crypto exchanges that signed early — get a forced migration to Decagon when Sentinel sunsets.
  • Pylon, Maven AGI, Crescendo, Wonderful — all get squeezed in the same compression; 2-3 of these don't survive 2028.
  • The "horizontal AI agent" thesis broadly — 60-70% of 2024-2026 application-layer AI startups are dead or acquihire'd by 2029.

Pivoted (mixed outcome)

  • Sentinel's compliance lead lands well — joins Chainalysis as Head of AI Compliance. The role itself becomes a category.
  • Mark Kingston archetype — power-user-as-product-input becomes a recognized pattern; he writes a Substack, gets a Director of Support Ops role at a stablecoin issuer in 2029.

4. Our Position

ARR ceiling: $4.1M (peaked Q3 2028, plateaued). Valuation trajectory: $22M post (seed) → $35M effective at bridge → $24M acquisition. Fate: Acquihire by Decagon, Q1 2029. ~$24M total consideration (60/40 cash/retention). AZ recovery on cap table: ~$2.1M cash on its strategic stake plus ~$1.4M in engineer retention packages flowing to two ex-AZ engineers (who are no longer AZ's anyway). Net to Axiom Zen: ~$2.1M, vs ~$3M of AZ-allocated dev cost over the 2025-2026 period plus $750K of post-spinout services-in-kind. AZ slightly underwater on the venture, materially underwater on opportunity cost. Honest framing: This is a Forethought-shape outcome at 1/8th the price. It's not a wipeout — the team and brand survived, the product code shipped, the customers were real — but it's the worst-case-that-isn't-a-zero. The thesis was right in shape and wrong in timing and scale.

5. Key Metrics Shape

MetricQ4 2026Q4 2027Q4 2028
ARR$1.1M$3.2M$4.0M
Logos498 (one churn)
Pipeline conversion (qualified → close)28%17%9%
Win rate vs Decagon35%14%8%
Win rate vs Sierran/a (didn't compete)22%11%
NRR108%71%64%
Gross logo retention92%78%71%
ACV (new logos)$145K$130K$98K
ACV (Dapper specifically)$320K$180K$140K
CAC payback (months)111928
Burn multiple1.4x2.1x3.4x
Magic number0.70.40.2
Sales cycle (days)78134168
Inbound % of pipeline38%21%14%

The ugliest line is NRR 71% in 2027: it's not just churn, it's downsell. Existing customers renew but at 60-80% of original ACV because Decagon's pack came in at 70% of Sentinel's price with 80% of the functionality.

6. What AZ Tried That Didn't Work

  1. Anchoring on Dapper's economics. Even after Priya warned against it, the first three deals were priced as Dapper-shape ($150-200K) instead of Marcus's $300-400K compliance-grade pricing. Lost the pricing power early; couldn't reclaim it after the Decagon pack landed.
  2. Treating SOC 2 + ISO 42001 as an engineering project, not a sales weapon. SOC 2 Type II shipped six weeks late (Q3 2026 → mid-Q4); ISO 42001 slipped to Q2 2027. The MiCA-deadline forced-buy window closed before the certifications were ready, and 4-6 deals that would have closed on certifications-as-differentiator went to Decagon-with-promises instead.
  3. Believing HITL was a permanent moat. The HITL pattern was right but it was a UX pattern, not an architecture moat. When Anthropic shipped HITL middleware as a primitive in Q2 2027, the "we have a thoughtful HITL flow" claim became "everyone has HITL."
  4. Underweighting the indemnification weapon. Priya's "indemnify above contract value" lever was floated but never operationalized — legal review took two quarters and by the time the policy was ready, Decagon had matched on the top-tier SKU.
  5. Going founder-led on sales too long. The first AE wasn't hired until $2.4M ARR (vs Priya's $1.5M trigger). By the time the AE ramped, the competitive landscape had shifted.
  6. Hoping Dapper would expand into a real reference. Dapper's logo was less valuable in 2027 than in 2026. By 2028 it was a liability — "Dapper is shrinking" came up in two-thirds of late-stage deals as an objection.
  7. Trying to horizontalize too late. The 2028 pivot to "compliance overlay on top of Decagon/Sierra" was the right pivot but six quarters late. The platform players had already absorbed the workflow.
  8. Not raising big enough early. A $5M seed gave 18 months runway in a market that compressed in 12. A $10M seed at $30M would have bought the second wave of certifications and a real GTM motion before Decagon's pack hit.

7. The Pivot Options

The realization moment: Q2 2027, ~6 weeks after the Anthropic dev day. Three internal datapoints converge: (a) two enterprise deals stall citing "we're going to wait for the Anthropic-Zendesk integration"; (b) Decagon's vertical pack public pricing comes in at 65% of Sentinel's; (c) Q1 2027 board deck shows pipeline conversion dropped from 28% to 19%. The CEO knows by July; the board is told in September.

Realistic pivot paths from a $4M-ARR-stalled crypto vertical specialist:

Pivot A — Compliance overlay / audit layer (the one they tried) Sell to Decagon/Sierra/Zendesk as a partner, not compete. Productize the regulator-ready audit bundle as a standalone SKU. Realistic ceiling: $8-12M ARR, but margin profile is awful (you're a tax on the platform). Outcome: $40-60M strategic acquisition by a compliance vendor (Chainalysis/TRM) or by the platform itself. Probability of executing well: 25%.

Pivot B — Insider-threat / support-tooling-security pure-play The Coinbase-breach narrative is durable. Pivot away from "AI support" entirely, become "support tooling security and forensics." Sell to CISOs not VP CS. Comparables: Cyera, Varonis-for-support. Capital-efficient and defensible. Realistic ceiling: $15-30M ARR over 3 years; could become a real company. Probability of executing well: 15% — requires a security-DNA hire most teams don't have.

Pivot C — Open-source the skills layer, monetize hosting Decagon CEO's worst-case: AZ open-sources skills extraction, becomes the Cursor-for-support pattern. By 2028 it's late, but a desperate version is plausible. Realistic ceiling: $5-15M ARR with a venture-style narrative. Probability of executing well: 10% — requires DevTools chops, GTM motion not in the team's DNA.

Pivot D — White-label to crypto exchange platform vendors Sell to HollaEx, Codono, white-label exchange platforms; ship as embedded module to their downstream operators. Distribution-first. Realistic ceiling: $6-10M ARR, low ACVs, high logo count. Probability of executing well: 30% — closest to the existing motion.

Pivot E — Accept the acquihire (what they actually do) Realistic. Honest. Returns capital to investors, gets the team a soft landing, preserves AZ's reputation for "ships product, exits cleanly." Probability chosen: ~50% in this scenario.

The pivot they should attempt at the moment of realization (Q3 2027) is B (security pure-play) — but they don't, because the founding team is product-and-CS-DNA, not security-DNA, and the two senior compliance hires are not CISOs.

8. Leading Indicators (what to watch NOW in May 2026)

  1. Decagon ships a vertical pack feature OR signs a top-10 exchange/stablecoin issuer publicly. Watch for any Decagon press release with the word "vertical," "regulated," "compliance pack," "fintech," or any logo from the crypto-native list. Trip-wire: Decagon publishes a crypto case study before Sentinel signs its third non-Dapper logo.
  2. Anthropic / OpenAI primitives. Watch the Anthropic dev day agenda (typically May/October), specifically anything titled "agent loop," "support reference architecture," "HITL middleware," "prompt caching for customer service traces," or any MIT-licensed support stack. Trip-wire: a first-party Anthropic Zendesk/Intercom connector ships.
  3. AZ misses Q3 2026 SOC 2 Type II deadline by more than 4 weeks. Internal trip-wire. Tied to: "have we hired the compliance engineer (ex-Persona/Chainalysis/Fireblocks) by July 2026?"
  4. Lorikeet raises a Series B at $300M+ post OR is acquired in the next 12 months. Either signals the regulated-vertical thesis is being absorbed by a larger player and the wedge is closing.
  5. Dapper layoff round in 2026 (any further reduction past current state). Each layoff is a tick down on lighthouse value and a tick up on renewal risk. Trip-wire: Dapper headcount drops below 150.
  6. Dapper builds rather than buys. Specifically: Dan Carrero's vibe-coded Retool replacement ships and replaces meaningful workflow. Trip-wire: any internal Dapper memo or Slack about "we're going to extend our internal tooling instead of going with Sentinel for X."
  7. Sierra hires a "regulated verticals" lead OR ships ISO 42001 + HIPAA enterprise in the next 9 months. Trip-wire: Sierra LinkedIn posting for a "VP Regulated Verticals" or "Head of Compliance Products" role.
  8. AZ pipeline conversion drops below 22% on qualified deals in any quarter. Combined with sales cycle stretching past 90 days — that's the squeeze starting.

Bonus: watch Forethought retrospectives. Ex-Forethought PMs writing post-mortems on LinkedIn (2026 will produce a wave) are the canary for what compression-from-platforms looks like in this category.

9. Historical Analogue

Primary: Forethought (2017-2026). AI customer-support startup, $65M raised, peaked ~$25M ARR, acquired by Zendesk for $200M+ in March 2026 — Zendesk's largest deal in two decades. Forethought had the right insight (AI in support is real, ServiceNow/Zendesk are slow), the right early customers (Upwork, Instacart, Asana), but got compressed between Zendesk catching up natively and Decagon/Sierra eating the high end. The Sentinel arc rhymes almost exactly, just one tier of capitalization smaller and one vertical narrower.

Secondary: Inflection AI (2022-2024). Built a beautiful product (Pi), got $1.5B in funding, then the platform players (OpenAI, Anthropic) made the application layer commoditizable, and Microsoft executed an acquihire-not-acquisition for $650M in March 2024. The Anthropic-ships-canonical-support-agent moment is the platform-eats-the-app-layer dynamic at a smaller scale.

Tertiary: Pivotal Tracker (2008-2025). Closer to the AZ DNA — built inside a consultancy (Pivotal Labs), spun out, ran for 17 years, never quite escaped its origin culture, shut down April 2025. The services-to-product transition is the ambient drag in this scenario.

The reason this rhymes: vertical wedges work when the platform layer is slow and the horizontal players are distracted. When both move at once — as they do in 2026-2027 — the wedge company has 12-18 months and needs to either get to escape velocity ($15M+ ARR with a clear expansion thesis) or accept a strategic exit.

10. Probability Estimate

p(scenario) = 30-38%, point estimate 33%.

Justification:

  • p(Decagon ships a vertical pack OR acquires Lorikeet by mid-2027) ≈ 55%. They have the capital, the strategic logic is obvious, and Coinbase Ventures-as-design-partner is a board-level conversation already.
  • p(Anthropic ships canonical support agent reference architecture by end of 2027) ≈ 45%. Anthropic has been moving toward primitives; whether they ship a specific support-agent loop vs. just primitives is the uncertainty. The Decagon CEO perspective explicitly recommended against this — but Anthropic doesn't optimize for Decagon.
  • Conditional on both: p(Sentinel stalls at $3-5M ARR | both events) ≈ 70%, because the squeeze geometry is forced.
  • p(Dapper materially deteriorates by 2027) ≈ 65% — base rate from current trajectory.

So roughly 0.55 × 0.45 × 0.70 × (Dapper conditional) ≈ 25-30%, plus path-correlated scenarios where only one of the two events happens but the wedge still closes (e.g., Sierra moves into compliance verticals first, or Zendesk acquires Decagon and the platform subsumption happens differently) ≈ +5-8%. 33% point estimate.

This is the most likely bear scenario but not the modal scenario overall — the modal scenario is muddier (a $10-15M ARR business that raises a Series A, doesn't quite hit escape velocity, and either acquihires at $80-120M or grinds along as a profitable mid-size). Vertical Squeeze is the cleanly bad version of the muddy outcome.

11. The Three Mistakes That Make This Scenario Real

  1. Failing to hire the Compliance Engineer (ex-Persona/Chainalysis/Fireblocks) by July 2026. Marcus Tan's "boring truth" — without this hire, SOC 2 + ISO 42001 + MiCA tooling slip, the forced-buy window closes, and the certification-as-sales-weapon advantage evaporates before Decagon's pack ships. This single hire is the highest-leverage 2026 decision.

  2. Anchoring 2026 pricing to Dapper's $200K-shape rather than the compliance-grade $300-400K. Once the first three contracts get signed at Dapper-economics, pricing compression is locked in. Sentinel never gets the chance to be the "regulated-grade premium" vendor — it's the "cheaper than Decagon" vendor, which is the worst position when Decagon cuts price.

  3. Not separating the cap table, brand, and CEO authority cleanly enough. AZ-the-agency culture leaks into operating cadence; the founding CEO is a hybrid AZ-and-Sentinel role; decisions go through AZ partners. By the time the squeeze comes, the company has neither the focus of a venture-backed startup nor the patience of a services firm. Marcus Tan's single question — "are you willing to spin this out with separate cap table, separate brand, and a CEO who is allowed to fire AZ as a vendor on day one" — is answered "no, kind of, sort of," and that ambiguity costs 6 months of velocity in 2026 when speed mattered most.

12. The Asset Salvage Value

What AZ owns at the end of this scenario, residual:

  • Team: ~6 senior engineers and 1 compliance lead with 3 years of crypto-vertical AI support experience. Highest-value asset. In 2029 these are the most experienced "AI support for regulated verticals" engineers in the market. Decagon retains 3-4; the rest are placeable at Chainalysis, TRM, Fireblocks, Circle, or as a founding team for a next-cycle company. Conservatively $4-8M of human capital value.

  • IP & code: HITL middleware, skills/corrections engine, regulator-ready audit bundle, hacker-vs-victim disambiguation classifier, 15-20 working integrations (Etherscan, Solscan, Flow Scan, Persona, Chainalysis, TRM, Fireblocks, etc.). Decagon absorbs most of this; some is open-sourceable. Standalone value once absorbed: ~$0 to AZ. Pre-absorption: ~$5-10M.

  • Brand (Sentinel): Modest. Three years of category presence, a published State of Crypto Support report (years 1-3), some podcast equity. Brand likely sunset in acquisition. Salvage value: low, $200-500K to a non-competitor.

  • Customer relationships: 8-9 active accounts, mostly mid-tier crypto. Decagon values these at maybe $8-12M as a customer-book sweetener. Strategic value to a non-Decagon acquirer: similar.

  • Cap table optionality: AZ holds ~14% of Sentinel post-bridge. At $24M acquisition: ~$2.1M cash. Carry to AZ partners: minimal. Tax-loss harvesting: meaningful for AZ's own ledger.

  • Reputational asset (the underrated one): AZ now has a documented playbook for "spin a vertical AI product out of an agency, get to $4M ARR, exit cleanly." That's a credible second-spinout pitch even after a modest outcome. The AZ-Dapper 2018 spinout + AZ-Sentinel 2026 spinout becomes a two-data-point pattern. Whether this attracts the right next-cycle founder is the real long-tail asset.

  • Lessons asset (hardest to value, easy to underweight): The Decagon CEO and Marcus Tan perspectives, replayed against the actual outcome, become AZ's internal playbook for the next venture studio bet. If AZ internalizes "compliance hire by month 4, separate cap table, $10M not $5M seed, exit-cleanly mindset from day one," the next spinout is materially better-positioned. This is the only way the Vertical Squeeze scenario produces a positive long-term return: AZ learns the shape of the squeeze early enough to avoid it next time.

Net salvage to AZ: ~$2-3M cash + 4-6 senior alumni placed across the crypto-AI ecosystem + a sharper second-spinout playbook. Not a wipeout. Not a win. A tuition payment for a still-credible venture studio.

— End scenario document —