Logistics / Supply chain · Deep dive

BackOps

San Francisco AI startup from Amazon Shipping alum Sean McCarthy building 'Relay,' an AI-native operating system that reads unstructured messages, logs into vendor portals, and makes phone calls to run shipper and 3PL back-office workflows — files 100% of eligible carrier claims automatically, cuts customer response time 93%, and raised a $26M Series A led by Theory Ventures on March 12, 2026.

emerging

The question that decides it: BackOps sells a horizontal 'agent OS' that stitches WMS, ERP, carrier portals, email and phones into one automated back-office layer for shippers and 3PLs. With Augment ($110M), HappyRobot ($200M at a $1.2B valuation, August 2026), Vooma, Loop ($160M), Fleetworks and the TMS incumbents (Blue Yonder, Descartes, e2open, Manhattan, Kinaxis) all shipping native copilots at the same freight and warehouse back-office spend Gartner sees growing from under $2B in 2025 to $53B by 2030, does BackOps' horizontal positioning become the connective platform layer buyers standardize on — or does it get squeezed between vertical AI-agent winners in freight brokerage and voice, and TMS-native alternatives already sitting on the data?

My take ★★ Very interesting ★ Interesting − Maybe ✗ Not interesting

Snapshot

BackOps is a two-year-old San Francisco startup building an AI-native operating system for the supply-chain back office: the emails, phone calls, portal logins and reconciliations sitting between a WMS, TMS, ERP and carrier website when a package is lost, damaged, delayed or invoiced wrong. Flagship product Relay deploys AI agents that read unstructured messages, log into portals such as FedEx and UPS, place and receive phone calls, and post updates into the customer’s systems of record. The company says it files 100% of eligible carrier claims automatically, cuts customer-response time by 93%, and resolves 84% of issues fully autonomously (Series A release, March 12, 2026). It raised a $26M Series A on March 12, 2026 led by Theory Ventures with Gradient, Construct Capital and 10VC following on — up from a $6M Construct-led seed (June 24, 2025) and a $2M Gradient-led pre-seed (October 2024), taking total funding to roughly $34M in 20 months. It is one of the most-funded shipper/3PL “agent OS” bets Augment, HappyRobot, Vooma and Loop are attacking simultaneously.

Founding story

Sean McCarthy was an early hire on Amazon Shipping and — per Product Market Fit and Category Visionaries podcasts (2025) — the role kept him inside customer warehouses roughly four days a week. The bottleneck he saw was not receiving, picking or trucking but the human back office: admin staff on the phone with carriers filing damage claims, tracing missing packages, replying to end-consumer emails, and rekeying between WMS, TMS, CRM and email. Turnover was brutal and the layer was invisible to software vendors who thought they had already automated the warehouse.

McCarthy quit Amazon in July 2024 and, in six weeks, cold-emailed his way through 85 warehouse-owner interviews. He closed the $2M pre-seed led by Gradient in October 2024 largely off those calls. Co-founder Henry Ou brought the technical half: applied ML leadership at Apple and ByteDance and a prior ERP-integration business. The founding bet: the interesting problems in supply-chain automation are not routing or forecasting; they are the exceptions, phone calls, portals and pdf attachments — a world that rewards patience with edge cases over model horsepower.

How it works

A BackOps deployment starts with a point of communication: an inbound email, Slack message, SMS, phone call, or WMS webhook. Relay ingests the trigger and, using an LLM-driven agent runtime, identifies the shipment, customer, SLA, contract terms and next action. It then executes across the surfaces a human admin would touch — logging into carrier portals (FedEx, UPS, DHL and various LTL carriers per BackOps/Construct materials) to file claims, scraping tracking events into the WMS, drafting in-tone customer emails, creating re-ship orders in the ERP, or placing/receiving phone calls and transcribing them. Series A materials say 84% of issues resolve without human intervention.

What distinguishes BackOps from a generic RPA vendor is the Workflow Recorder. Operators walk through a real workflow on screen once — filing a Damage-in-Transit claim on FedEx, resolving an OS&D discrepancy — and the system captures every click, form field, portal URL and copy-paste, normalizing it into a workflow Relay can run at scale. The recorder attacks the biggest failure mode of enterprise automation: workflows that live only in the head of a senior operator. Institutional knowledge becomes training data.

Product and business overview

The product is two layers. Relay is the runtime — the agent engine that takes a trigger, decides the action, and runs it across email, phone, portals and systems of record. AI Process Center is the design surface where Workflow Recorder captures processes and promotes them to Relay. Around those sit integrations into WMS, ERP, CRM and comms tools and pre-built skills: file-a-carrier-claim, chase-a-POD, reconcile-an-invoice-line. Named use cases (Gradient/Construct/Theory posts, March 12, 2026) are claims filing, missing-package resolution, customer-service triage, order-status updates, and vendor coordination. The target buyer is a shipper, 3PL or warehouse operator with meaningful direct-to-carrier volume — operators for whom the back office is a headcount line and turnover problem.

Business model and pricing

BackOps does not publish pricing (through September 2026). Series A and partner posts describe a SaaS motion with usage layered on top. The economic pitch: 84% autonomous resolution and 93% faster response translate into a headcount comparison against the customer’s back-office team, plus recovered carrier-claim dollars otherwise written off. The 100%-of-eligible-claims-filed metric is the sharpest wedge because it is a P&L line item — every unfiled damage or shortage claim is money the shipper forfeits to the carrier. The unresolved question is whether BackOps monetizes on seats (competing with TMS copilots) or on outcomes (per claim recovered). Rivals are already fragmenting — Pallet per task, HappyRobot per voice call, Loop on data volume — and BackOps has not picked a lane.

Traction over time

Independent traction — ARR, logo counts, retention — is not disclosed through September 2026. Proof points are workflow-level, not commercial-level.

Market analysis

Gartner projected on April 7, 2026 that SCM software with agentic AI grows from under $2B in 2025 spend to $53B by 2030, a 93.5% CAGR, with 60% of enterprises using SCM software adopting agentic features by 2030 (up from 5%), and $11B of annual SCM revenue shifting away from legacy vendors — precisely the reallocation BackOps is trying to catch. Theory’s memo (March 2026) frames the immediate wedge — carrier claims and lost-package resolution — as a $3.5B market growing 13% annually, citing 85M US packages arriving damaged in 2024 (up 30% YoY). Parcel volume keeps growing, carrier reliability has trended sideways, and offshored back offices — Filipino and Indian BPOs at $8-15/hour — are exactly the labor pool AI agents are cheapest to displace.

Competitive intel

Augment ($110M by September 2025) is the closest horizontal analogue, with a broker- and 3PL-facing agent product and a founder (Harish Abbott) with more distribution than any BackOps competitor. HappyRobot raised a $150M Series C at a $1.2B valuation in August 2026 ($200M total; 150+ enterprise customers including DHL and Uber Freight) and owns the freight voice-agent surface — direct overlap if it extends into portals and email, with ~6x BackOps’ capital. Vooma ($16.6M) is narrower, targeting mid-market broker workflows. Fleetworks ($17M Series A, October 2025) attacks the phone-call surface carrier-side. Loop ($160M after a $95M Valor-led Series C in April 2026) started in freight audit and is now the logistics data-and-intelligence layer — complementary today, converging threat if it adds action-taking.

The structural competitor is the TMS/SCM/WMS stack — Blue Yonder, Descartes, e2open, Manhattan, Kinaxis, SAP, Oracle — all sitting on the transaction data BackOps’ agents must reach into, and all shipping native copilots in 2026. Gartner implies most agentic AI spend runs inside SCM software. Silent competitors — in-house builds at the largest shippers, offshore BPOs underbidding on labor — cap the top and bottom of the market.

History and evolution

July 2024: McCarthy leaves Amazon, runs 85 discovery interviews. October 2024: $2M Gradient-led pre-seed. June 24, 2025: $6M Construct-led seed, first paying shippers and 3PLs in production. Late 2025: Workflow Recorder shipped. March 12, 2026: $26M Theory Ventures Series A, headline metrics on record, first in-house recruiter hired. No pivots or failed launches disclosed as of September 2026.

What people say

The case for. Gradient (March 2026) frames BackOps as the intelligence layer warehouse operators deserve, praising the workflow-recording approach that lets non-engineers extend automation coverage. Construct Capital (June 2025, March 2026) argues the value is being the connective layer across systems never designed to talk to each other, citing 84% autonomous resolution as evidence BackOps has crossed the reliability bar most agent products fail. Theory emphasizes McCarthy’s $100B manual-logistics-labor number and the $3.5B carrier-claims wedge.

The complaints. Independent criticism is thin. FreightWaves’ coverage (June 2025, March 2026) is straight reporting rather than review. HackerNews and Reddit skepticism on AI agents for supply chain in 2025-26 lands on three recurring complaints: (1) agent reliability on nondeterministic tasks — the “last 15%” gap that turns 85% autonomous into a bug-ticket generator; (2) portal fragility — carrier and vendor sites change layouts, add captchas or block scrapers, and every change breaks an automation; (3) “AI-washing” of what is fundamentally RPA plus an LLM wrapper. Gartner (April 2026) is that most agentic AI spend accrues inside incumbent SCM software, implying horizontal challengers get squeezed.

Outlook: the open question

What would have to be true for BackOps to win: the Workflow Recorder must produce a network effect the TMS incumbents cannot copy, and the horizontal agent OS must remain a cheaper, faster deployment than what Blue Yonder, Manhattan, SAP or Oracle can ship natively. The strong case is that BackOps has picked the right wedge (carrier claims and exception handling — P&L visible), the right buyer (shippers and 3PLs whose back office is a headcount line and turnover problem), and the right technical bet (workflows recorded from real operators). The 100%-of-eligible-claims-filed and 93%-faster-response numbers, if they hold under third-party audit, are what makes a CFO reallocate budget quickly.

The weak case: horizontal “agent OS” is where three or four well-funded startups already sit and every TMS/SCM incumbent is now shipping. Augment has more distribution, HappyRobot ~6x the capital and owns phone calls, Loop is bigger and moving from data into action, and the incumbents own the data. The falsifiable version: does BackOps’ Workflow Recorder plus its carrier-claims wedge produce measurable share gains in shipper and 3PL logos over the next 18 months against Augment on the horizontal side and Blue Yonder/Manhattan copilots on the incumbent side — or does the market fragment so fast that OS positioning is worth less than a narrower, deeper wedge?

How to attack it

The most attackable seam is monetization. BackOps prices (opaquely, as of September 2026) as SaaS with usage layered on. Every named workflow — claims filed, packages recovered, invoices reconciled — is a P&L line item, and every shipper has a written-off number for how much money the carrier keeps because the back office never got around to filing. A well-funded attacker should build a claims-recovery-as-a-service agent priced purely on a percentage of dollars recovered. No seat license, no usage meter, no procurement cycle: the shipper signs a one-page contract, the agent files every eligible claim across FedEx, UPS, DHL and LTL carriers, and the vendor takes 20-30% of the money that comes back. That structure beats BackOps on procurement friction and risk, and — critically — on what the TMS incumbents can copy: Blue Yonder and Manhattan cannot ship a percentage-of-recovered model without cannibalizing license revenue.

The second seam is portal fragility. Every agent company scraping carrier and vendor portals is one captcha refresh from a broken customer. An attacker could ship a portal-integration marketplace where certified carrier partners provide sanctioned API surfaces — turning scraping into a partnership problem and locking in carriers before BackOps or Augment can. The third seam is talent capture: BackOps has ~22-30 employees (PitchBook, mid-2026) and is only now hiring its first in-house recruiter; a funded competitor could burn a small slice of its capital on ex-BackOps founding engineers at 2-3x liquidity. Finally, the horizontal-OS category is exposed to a vertical specialist going 10x deeper on one industry (medical distribution, cold-chain, hazmat) where portal complexity and regulatory scar tissue reward focus.

Adjacent-segment play

The most obvious adjacency is not another logistics use case; it is carrier claims recovery for the shipper’s cargo insurer. Freight and cargo underwriters already pay adjusters and TPAs to process cargo damage claims, and the same agent stack — reading emails, logging into portals, filing claims, chasing PODs — is the exact workflow TPAs like Sedgwick and Crawford run manually today for cargo lines. Reserv is proving there is an AI-native-TPA business (P&C, ~$100M ARR by May 2026); a cargo variant Relay could power is a natural adjacent buyer, sold to the underwriter rather than the shipper.

A second adjacency is construction and MEP subcontractors, whose back office runs the same shape of workflow (invoice reconciliation, warranty claim filing, RFI chasing) across GC portals and email. Third is specialty-pharmacy and medical distribution, where DEA-scheduled reconciliation and 340B claims are labor-heavy and portal-driven. Down-market, the platform could be repackaged as a prosumer agent for Shopify merchants shipping 50-500 packages a day — priced monthly. What does not generalize is anything requiring human physical presence — warehouse-floor operations, driver tasks, quality inspection — a large chunk of “supply chain” spend where an agent OS has no traction.

Sources and further reading

• BackOps Raises $26M Series A to Build the AI-Native Operating System for Global Supply Chains — BusinessWire, March 12, 2026
• BackOps raises $26M to automate global supply chains with AI — FreightWaves, March 12, 2026
• BackOps AI Raises $6M Seed Round to Redefine Logistics Automation — PR Newswire, June 24, 2025
• Supply Chain Automation Startup BackOps.ai Secures $2M in Pre-Seed Funding led by Gradient — PRWeb, October 2024
• An Agent Skills Platform for the Real World: Our Investment in BackOps — Theory Ventures, March 2026
• One Billion Lost Packages — Tomasz Tunguz, Theory Ventures, March 2026
• BackOps: The Intelligence Layer Warehouse Operators Deserve — Gradient Ventures, March 2026
• Gartner Forecasts SCM Software with Agentic AI Will Grow to $53 Billion by 2030 — Gartner, April 7, 2026
• HappyRobot Series C mints FreightTech’s newest unicorn — Yahoo Finance, August 2026
• Loop Raises $95M Series C to Scale Its AI Platform Across the Supply Chain — Yahoo Finance, April 2026
• Sean McCarthy, Co-Founder & CEO of BackOps AI — Category Visionaries podcast — 2025