Relay

Summary

Every time an enterprise changes a core system — its ERP, CRM, HR platform, insurance core, compliance stack — the hardest, most expensive, most failure-prone part isn't the new software. It's moving years of data and workflows out of the old system. That migration is a months-long, hand-built project, and it is the single biggest reason enterprise software changes stall or fail: 70% of transformations fail, and the root cause is repeatedly the data and adoption layer, not the technology.¹

Two forces are turning this from a chronic cost into an acute bottleneck. First, a wave of AI-native systems of record — funded heavily by YC and a16z — every one of which must rip out a legacy incumbent to land a customer.⁹ Second, those vendors still do migration the old way, burning scarce forward-deployed engineers on each deal.

Relay is agent infrastructure that automates the migration itself — source system to destination, data and workflows, source-agnostic, under human control, fully audited. It pays off on both ends: the AI-native vendor selling in, and the enterprise adopting. And migration is only the wedge — the vendor-neutral map Relay builds from every migration positions it to become the long-term stack advisor for the agentic enterprise.

The problem — migration is the tax on every software change

The failure and cost data is strikingly consistent:

• 70% of digital transformations fail¹ — and failed transformation efforts cost businesses an estimated ~$2.3 trillion per year.² The repeatedly-cited root cause is not the technology; it is the data and the human adoption layer.
• 83% of data-migration projects fail outright or exceed their budgets and timelines.³
• Enterprise ERP implementations fail 55–75% of the time — a rate dominated by large-enterprise projects run through third-party consulting firms.⁴

Failure clusters on the data because moving records, relationships, custom objects, and business logic out of an entrenched incumbent is bespoke work every single time. So it is outsourced to human labor — an enormous market whose entire job is to hand-carry data from one system to another, and which still fails most of the time (see Market).

Why now — the AI-native systems-of-record wave makes migration the gate

Enterprise software is being rebuilt AI-native, and the new entrants are replacing the incumbents outright, not sitting on top of them. In the last two YC/a16z cohorts alone: AI-native ERP ripping out NetSuite/SAP, AI-native HR/payroll versus Workday/ADP, AI-native lending cores, RCM, CRM, insurance, and compliance systems. a16z has published the thesis directly — "Workday's Last Workday?"⁹

Every one of these hits the same wall: to win an enterprise customer, they must migrate that customer off the incumbent. That migration is now the gate on their growth — vendors typically deploy 1–4 forward-deployed engineers for ~12 months per enterprise deal (~$300K–$1.8M of cost) before a customer is live,¹¹ making FDE bandwidth, not sales, the binding constraint on enterprise ARR.

The category is real enough that a16z is funding the migration layer directly: Tessera Labs raised a $60M Series A (a16z-led) to automate ERP/SAP migrations with a multi-agent platform.¹⁰ The migration substrate is a category, and it is being contested now.

What we're building

Relay is an agentic migration substrate. It maps a source system and moves its data, relationships, and business logic into the destination — source-agnostic, so the same engine works across whatever a customer is leaving and whatever they're adopting. It handles not just records but the structure and logic around them, at a scale and consistency hand-built projects can't match.

Two properties matter to a buyer regardless of the internals:

• It stays under human control. Every run moves through approval gates — a person signs off on what changes; nothing happens as an opaque black box.
• It's fully auditable. Every step produces a complete, verifiable trail — which matters most exactly where the data is most sensitive and most regulated.

The effect is to turn a bespoke, months-long, mostly-manual migration into a repeatable, governed, largely-automated one.

Both ends — how it helps the vendor and the enterprise

Relay sits between the two sides of every migration, and pays off for both.

The vendor end — the AI-native system of record selling in

• Scale enterprise ARR without scaling FDE headcount — the bottleneck on how many enterprise deals can run in parallel is removed.
• Compress onboarding from ~12 months to weeks — shrinking time-to-live and time-to-revenue.
• De-risk the land — the scariest part of an enterprise's switching decision (moving its data) becomes a demonstrable, solved step.

The enterprise end — the company adopting the new software

• A faster, cheaper, lower-risk switch — years of trapped data move with automated validation, not a hand-built, error-prone project.
• Governance for regulated data — human-approval gates and a full audit trail, where the data is most sensitive.
• People freed for the layer that actually decides success — with the plumbing automated, the team spends its energy on adoption and change management, the human layer where 70% of transformations fail today.¹ Relay doesn't replace that layer; it clears the way to it.

The symmetry is the point: the same automated migration that lets a vendor take more enterprise deals is what lets the enterprise say yes without fear. Relay makes the market on both sides of its own transaction.

The long game — migration is the wedge, the stack advisor is the moat

Migration is how Relay gets into the flow; it is not the end state. Every migration teaches Relay the exact shape of an enterprise's stack — which systems it runs, how its data and workflows are structured, and where it is moving. Because Relay normalizes every system it touches into a common, vendor-neutral form, it accumulates something no system integrator or point tool has: a cross-vendor, cross-enterprise map of how real companies are actually built and how they are changing.

That map compounds:

• Each migration makes the next one faster and safer — a data flywheel: more migrations → a richer cross-stack corpus → better migrations.
• The same map that powers migration powers advice: what to consolidate, what's redundant, what to migrate to next, which vendor actually fits a given stack, and what a move will cost and break before it starts.

This is the return to Relay's original idea — the stack advisor — but earned from real leverage rather than opinion. The one-time migration becomes an ongoing advisory relationship: the system-of-record intelligence layer enterprises (and the vendors selling into them) trust for every future stack decision. That accumulated corpus is the moat — it can't be bought, only earned one migration at a time.

Market

The labor market that exists today just to move data between systems — and that Relay automates:

• System integration services — $553.3B (2025) → $763.8B (2030)⁵
• Legacy modernization — $29.4B (2026) → $66.2B (2031), 17.6% CAGR⁶
• Data migration software — $14.7B (2026) → $48.3B (2035)⁷
• Cloud migration services — $16.9B (2024) → $70.3B (2030), 27.8% CAGR⁸

A half-trillion-dollar-a-year services market exists largely to hand-carry data from one system to another — and it still fails most of the time. That is the tax Relay removes, on top of a fast-growing base of AI-native systems of record that each need it to land.

The bet

Enterprise software is being rebuilt AI-native, one system of record at a time. Each rebuild is gated by a migration that today costs a half-trillion dollars a year in human labor and fails most of the time. Whoever makes that migration fast, cheap, and safe unlocks the whole transition — for the vendors building the new systems and the enterprises adopting them. That migration is the wedge; the durable prize is becoming the stack advisor for the agentic enterprise. That is what Relay is for.