Marshal
Builds and runs AI agents for small-business sales, support and finance workflows, with human approval before actions.
Company basics
- HQ
- United States
- Founded
- 2024
- Delivery model
- Managed (they run it for you)
- Pricing model
- Subscription
- Starting price
- From $199
Scope
- Regions
- United States
- Languages
- English (site only)
About Marshal
Marshal is a managed agent operations service, operated by a company called Machine-like, that designs, deploys and runs custom AI agents for small businesses. The company states its target is founder-led and operator-led businesses in fields such as software, professional services, legal and education, and it covers work in sales, customer support, finance, marketing and operations, so that clients do not have to build or maintain the agents themselves.
Agents connect to a client's existing tools, such as HubSpot, Gmail, Slack, Zendesk, Jira and Airtable, without data migration or replacing systems. According to the site, each action falls into one of three modes: autonomous execution, draft with approval, or escalation to a human before proceeding. Each completed task produces a receipt that logs the actions taken, and Marshal states that it monitors agents, handles exceptions and maintains them during the service period.
The company states that customer data is not used to train AI models, that access runs through credentials the client controls and can revoke, and that data is encrypted in transit and at rest. It offers a free trial that begins once the agent is working on real tasks, and states that customers can cancel at any time.
The company states it was founded in 2024 and is headquartered in New York. It began as Growth Marshal before shifting to agent operations.
What they do
- End-to-end delivery (you don't operate the tools)Source · checked 2026-09-29Marshal designs, deploys, monitors and tunes the agents and handles exceptions; customer approves rules and reviews results.
- Human review includedSource · checked 2026-09-29Human-in-the-loop gates for sensitive actions; drafts need customer approval before sending.
- Works inside your existing systemsSource · checked 2026-09-29Pricing page says it runs on the customer's existing stack; integrations page lists CRM, email, helpdesk tools.
Pricing
- Pricing model
- Subscription
- Published starting price
- From $199
- Pricing page
- View pricing
Evidence of outcomes
BOutcomes reported by the company or its investors.Named customers with quoted leaders and metrics (fitDEGREE 82% of tickets triaged by agents, Reinstein Law 57% more qualified cases), self-reported: https://marshal.ing/customers
A: third-party verifiable outcomes · B: outcomes reported by the company or its investors · C: no publicly verifiable outcomes yet. Demos, forecasts and slogans do not count.
FAQ
How does a client get started with Marshal?
A client creates an account, connects work systems, and Machine-like engineers configure the agent for a specific workload. A free trial begins once the agent runs on real work. (Source: https://machinelike.ai/pricing)
Does Marshal work with existing tools?
Yes. The site states that Marshal integrates with established platforms such as HubSpot, Gmail, Slack and Salesforce through standard integrations, and core systems remain in place. (Source: https://machinelike.ai/pricing)
What happens when the agent is unsure?
The agent avoids guessing. Work outside defined parameters moves to an exception queue or is routed to a human for review. (Source: https://machinelike.ai/pricing)
How is data handled?
Systems stay in place; agents access them through credentials the client controls and can revoke. Data is encrypted in transit and at rest with audit logging, and is not used to train models. (Source: https://machinelike.ai/pricing)
Our read
Editorial view based on public information — not an endorsement.
Capabilities
| Check | Status | Source |
|---|---|---|
| End-to-end delivery (you don't operate the tools) | Confirmed | Source · checked 2026-09-29Marshal designs, deploys, monitors and tunes the agents and handles exceptions; customer approves rules and reviews results. |
| Human review included | Confirmed | Source · checked 2026-09-29Human-in-the-loop gates for sensitive actions; drafts need customer approval before sending. |
| Signed off by licensed professionals (CPA, attorney…) | Not stated | — |
| Works inside your existing systems | Confirmed | Source · checked 2026-09-29Pricing page says it runs on the customer's existing stack; integrations page lists CRM, email, helpdesk tools. |
| Private / on-prem deployment available | Not stated | — |
| API or data export | Not stated | — |
| Multilingual delivery | Not stated | — |
Commitments
| Check | Status | Source |
|---|---|---|
| Public delivery SLA | Not stated | — |
| Outcome-based pricing | Not stated | — |
| Refund / satisfaction guarantee | Not offered | Source · checked 2026-09-29Terms disclaim performance and outcome guarantees; fees non-refundable except as stated in writing. A 14-day free trial exists. |
| Customer data not used for training | Confirmed | Source · checked 2026-09-29Privacy and pricing pages say customer data is not used to train general models unless the customer authorizes in writing. |
| SOC 2 | Not stated | — |
| ISO 27001 | Not stated | — |
| HIPAA | Not stated | — |
| GDPR | Not stated | — |
| Named party responsible for outcomes | Not stated | — |
Funding
Not disclosed
How we check
Every check below links to its source and shows when we checked it. "Not stated" means we found no public statement — it does not mean "no".
Other routes
Use a tool yourself
Get it implemented
Sources
- Company website: https://marshal.ing/
- End-to-end delivery (you don't operate the tools): https://marshal.ing/managed-agent-operations
- Human review included: https://marshal.ing/platform/agent-governance
- Works inside your existing systems: https://marshal.ing/pricing
- Refund / satisfaction guarantee: https://marshal.ing/terms
- Customer data not used for training: https://marshal.ing/privacy
Change history
- 2026-09-29 · Listed on Toolsfine