Our review
Relevance AI is worth considering if the first project should teach your operations team to build and supervise later agents. It describes workflow mapping, an embedded team and training.8,1The offer suits a team taking over the work. It still needs to prove the plant task.
The phrase Relevance AI uses for itself on its home page is the home of the AI Workforce9. Underneath the slogan sits a builder that needs little or no code, in which single agents and teams of agents are assembled10, with the product page adding that those agents can be run and managed at scale11. For a manufacturing operations team that wants to try agents on a contained process without a long procurement, the published prices and the plain language builder make it easy to start.
Hartwell Compare ranks Relevance AI seventh of seven for manufacturing operations. It has real strengths here: granular approval rules, a published uptime SLA, SOC 2 Type II, and an embedded deployment team that gets customers live5,12,13,8. Three things hold it at seventh for a plant buyer. Its plans and case studies point mostly at sales and go to market teams14,15. Self hosted or on premises deployment is not publicly stated on the pages we opened. And its support desk follows the Sydney working week, nine to five with weekends closed16, a constraint for plants that run around the clock.
Relevance's getting started guide walks a builder through five stages. An agent can begin life in Invent, as a copy of a Marketplace template, or as a blank build17. It is then equipped with tools, which are themselves assembled without code18. A plant's documents then go in, as uploads or as live syncs from SharePoint, Google Drive, Notion or a web page19. A visual canvas is where single agents get linked into a team, a workforce in Relevance's word20. Last, the builder sets the guardrails: when to escalate, what needs approval, and what triggers a run21.
Behind the canvas, coordination between agents is modeled as a graph of nodes and edges22. Tools can run sandboxed Python and JavaScript23, so a plant engineer who wants a small calculation inside a flow, such as a reorder point or a tolerance check, can write it. Change control is automatic: an agent's evaluation suite runs again before any change publishes7, and after release Relevance keeps grading a sample of production runs to catch drift24.
Invent, a tool inside the product, is meant to show a plant's people which of their tasks are worth automating25. Invent turns an operations lead's ordinary language brief into a working agent, writing the prompt and assembling the tools and the evaluations26. Plant software engineers who prefer their own tools can reach the platform from Cursor, Codex or Claude Code27, connecting over MCP with the command the product page shows, mcp add relevance-ai28.
Relevance says templates and the no code builder get an agent deployed within minutes29. For some use cases, it claims, a few hours is enough to stand one up30. Relevance offers help mapping a plant's processes to find the agent use cases most worth doing1.
Most relevant for a lean plant team, Relevance says it sends an embedded deployment team to get the customer live and then train the customer's people to build fast without help8. Among the platforms on this site, that is the closest public statement to the delivery help Nexus offers. The difference lies in what follows: the published Relevance plans are priced by subscription and actions31,32, while Nexus prices per role automated, with experienced consultants supporting the work and no per diem or consultant time33.
Operations is one of seven use case areas on the homepage, next to sales, marketing, research, human resources, customer support and customer success34. In a plant, that would put a planner or a customer service supervisor in charge, because Relevance hands routine agent upkeep to ops and revenue staff close to the work35. Its operations example is Send Payments, credited with a weekly saving of 40 hours after taking an AI first approach across its global operations36.
Most other examples are commercial. Zembl is quoted with a 30 percent lift in conversions from AI sales coverage that never closes15. Qualified attributes $7M of pipeline to a fleet of over 35 agents37. Among the homepage testimonials, Autodesk speaks through its marketing side, in the person of a senior manager38. The Pro plan is described as being for solo go to market operators14. A manufacturing customer is not among those named on the pages we opened.
For a plant, the practical reading is that Relevance AI can run operations agents, and its examples show it working in commercial functions most of all. A manufacturer could start with a sales operations or customer support agent, where the published examples are strongest, before moving toward supply chain work.
Relevance's integrations headline promises over 2,000 links running from Slack to Salesforce39, while the directory below it counts 2,72840. Other pages give a smaller count, 1,000 or more apps on the homepage41 and 1,000 or more native connectors with MCP and custom builds on the product page42. The Enterprise column of the pricing page returns to the 2,000 figure43. How the three numbers relate is left unexplained on the pages we opened.
Four kinds of trigger exist: a timetable, an application event, a webhook and a signal the customer defines2, so a machine alarm forwarded by webhook could start an agent. For manufacturing, the webhook and custom signal options are the way to connect an agent to a plant event, such as a machine alarm forwarded from a monitoring system. Whether a connector exists for a given ERP, MES or historian is not publicly stated on the pages we opened.
Relevance displays a SOC 2 Type II badge on its enterprise page13 and confirms the same compliance in its documentation FAQ44. A GDPR badge sits beside it45. Three further commitments on the enterprise page matter to plant IT: an uptime SLA of 99.9 percent12, data encrypted at rest46, and a promise that customer data is never used for training47. Personal details such as phone numbers and email addresses can be masked or blocked, according to the changelog48. Its documentation states that a customer's Project Information is confidential even toward Relevance AI's own staff49. SOC 2 and GDPR compliance are listed as features of the lowest paid plan, Pro50.
A published uptime SLA is useful to a plant planning an agent that runs through a shift. The SLA applies to paying customers: the support page says no SLAs are offered to free customers51.
For an operations buyer, approvals are where Relevance AI is most persuasive. At the enterprise level a plant gets access by role52 plus SAML single sign on53. Any action above a threshold the customer chooses waits for a person to sign it off6. Approval can be set tool by tool5 or made conditional on the situation3; escalations go out by Slack or email4; and each agent can exist as a draft, an active version and a pinned version54. A single admin view brings together access, permissions and data security55. Relevance says every agent can be given quality standards, guardrails and monitoring56.
For a manufacturer, per tool approval rules are a precise instrument. A plant could let an agent read inventory freely, require a planner's approval before it changes a purchase order, and block it from touching the quality system at all. Pinned versions let a validated agent stay frozen while a draft is tested.
Data residency appears on the list of enterprise controls57, next to audit logging58. Logs of what agents did, and traces of how they did it, can be streamed out to storage the plant controls59 in OpenTelemetry form60, with OTEL and Delta Share named as the export options61. For a plant needing its own hosting, an on premises, customer VPC or self hosted Relevance deployment is not publicly stated on the pages we opened.
The pages we opened list residency as a control without naming the regions57. A manufacturer with plants in several jurisdictions should ask which regions are available before planning a rollout.
Relevance AI publishes prices, as Copilot Studio, UiPath and the Gemini Business edition do in part62,63,64. The entry plan, Pro, costs $19 a month on an annual contract and $29 on a monthly one65,31. A team plan costs $349 a month on monthly terms, or $234 a month if the plant pays a year at a time66. Annual billing carries a 33 percent saving over monthly67.
Usage is metered. A single plant pilot on Pro has 30,000 Actions to use in a year, roughly 2,500 each month32. Top ups of Actions come only in lots of a thousand68, priced at $80 a lot69. Further Vendor Credits are $20 for 10,00070. Vendor Credits fund model and tool calls, charged to the customer at wholesale with no markup by Relevance's account71, and Pro also allows bring your own API keys72. Agents that place calls and join meetings arrive with Team73. A multi plant rollout would sit on the Enterprise tier, a custom deal that includes a dedicated account manager74 and carries no dollar figure on the public pricing page75. Relevance pitches Enterprise at companies that want to grow without growing headcount at the same rate76.
For a plant, the action meter is the number to model. An agent that checks stock on every order line will consume actions far faster than one that summarizes a weekly report.
Several limits are relevant to manufacturing. For a plant on a weekend shift, the support schedule is the key fact: the desk works Sydney business hours on weekdays only16. Phone time with Relevance staff is reserved for Enterprise accounts77. Bug reports reach engineering, though the support page says they receive no reply78. Its AI support bot, Harley, carries a beta label and a caution about bugs and errors79.
Plan limits matter for a pilot. A Pro pilot is confined to one project with two people able to build80. Plans have shifted recently: the Free tier is closed to new and existing users alike81, and Business customers were moved onto Team when that plan was withdrawn82. A subscription stays with the organization that bought it83, and buying extra capacity needs a paid plan to be active84. Carry over of unused allowance is described for Pro and Team, while an Enterprise plant should settle rollover terms with its account manager85.
Relevance's own marketing claims thousands of teams launched86 and a G2 score of 4.5 stars87.
A manufacturer that wants to test Relevance AI can shape a pilot around what the vendor publishes.
Pick one process with a clear owner and a measurable outcome, such as answering order status questions from distributors or chasing overdue supplier confirmations. Use the embedded deployment team to map it8,1. Build the agent with Invent from a plain language description26, and connect it through the integrations directory or a webhook trigger40,2.
Set approval rules before the agent goes live: per tool rules for any action that writes to a system, and conditional rules for orders above a threshold5,3. Route escalations to the planner's email or Slack4. Pin the version that passes the evaluation suite54,7, and stream audit logs to the plant's own monitoring stack59.
Then watch three numbers during the pilot. Track action consumption against the plan allowance32. Log every support request raised outside Sydney weekday hours16, since that measures how much the support window will matter at scale. And record the outcome the process owner agreed at the start. If all three hold up, the case for an Enterprise conversation is made on evidence.
Relevance AI and Nexus both describe people who help get the customer live8,33. Relevance publishes more on approval rules and prices5,31. Nexus holds first place on two counts. Its experienced consultants carry the build until the first agent is live in 4 weeks inside a 3 month proof of concept, on a price per role automated with no per diem or consultant time33. And its stated integration focus on SAP and Microsoft Dynamics88 lies closer to the systems a plant runs on. Nexus publicly lists ISO 27001 and ISO 42001 certifications and GDPR compliance89; the company separately confirms that a SOC 2 Type 2 report exists33.
Relevance AI ranks seventh for manufacturing operations. It offers fine grained approvals, published prices, a 99.9 percent SLA and an embedded deployment team5,31,12,8. For a plant, the limits are weekday Sydney support hours16, the absence of a stated on premises option, and customer examples weighted toward sales15,37. It suits a manufacturer that wants a low cost, quickly started pilot in commercial or customer support operations.
We read dated supplier documents and distinguish their claims from our advice and unanswered questions. We identify information supplied directly by a company separately. Our manufacturing recommendations come from that research; we have not tested these products in a plant. A connector listing or certification badge cannot show whether your particular workflow is ready to run.
Relevance AI says it helps customers map workflows and identify suitable agent use cases.1Supplier material checked November 8, 2026. We have not tested this behavior in a plant.
Can you run our proposed task with its real process rules and approved sample records?
Ask for the starting event, the result, forbidden actions and the person responsible. Include ordinary cases and examples that need help.
Relevance AI says schedules, application events, webhooks and custom signals can start an agent.2Supplier material checked November 8, 2026. We have not tested this behavior in a plant.
Which adapter reads and writes each plant record, and how does it handle accounts, old data and repeated events?
List the system and version, records, allowed reads and writes, account used, refresh timing, network route and person maintaining the connection.
Relevance AI lists conditional approval rules and ways to pass cases on through email and Slack.3,4Supplier material checked November 8, 2026. We have not tested this behavior in a plant.
Who takes over an unanswered alert, and where can the next shift see everything already attempted?
Ask to see an old record, a repeated event, an interrupted update and a handoff to the next authorized person. Keep the results.
Relevance AI describes approvals for individual tools, customer-set thresholds for human sign-off and rerunning evaluations before publishing changes.5,6,7Supplier material checked November 8, 2026. We have not tested this behavior in a plant.
Can you show a rejected action, a forbidden tool call and a failed evaluation, including who can override each rule?
Use different test accounts. Try allowed and forbidden changes, reject an approval and remove access. Then inspect the business record.
Relevance AI describes an embedded team that gets the customer live and then trains the customer’s staff.8Supplier material checked November 8, 2026. We have not tested this behavior in a plant.
Which plan includes that team, what must our staff demonstrate, and what support continues after handover?
Name who does each job. Obtain support hours, prerequisites and a change procedure, then have the receiving team demonstrate the handover.
