1. Home
  2. Services
  3. AI automation services

AI automation services for the work between your inbox and your ERP

CadenceWorks is an AI automation agency for back-office work. We take a job that starts with an email or a PDF, such as a request for a quote (an RFQ) or a purchase order, and have software do the typing. It ends as a draft record in your ERP, and a person on your team usually approves it. An ERP is the software you run orders, stock and accounts in.

RFQs and RFPs, both sidesOrders into the ERPDaily bookkeepingSalesforce and HubSpot
Automation run, customer orders
Order PO-77812
Draft
po_77812.pdf from the orders inbox Read
Customer and ship-to found in the ERP Matched
One line priced below the price list Flagged
Sales order saved as a draft for the order desk

Illustration with sample data.

What an AI automation agency does

It finds work your team repeats by hand and builds software to do it. AI reads the parts fixed rules cannot, such as emails and PDFs. We do this for one kind of work: documents that end as a record in your ERP or CRM.

It suits a team that types orders, quotes or bills into NetSuite, SAP, Odoo, Microsoft Dynamics, ERPNext, Xero, QuickBooks, Sage or a custom ERP. Our team has taken all of these live. You give us sample documents and your rules. The usual result is the job running in your system, with each record saved as a draft until a person approves it. What each build includes is agreed per project.

Which AI service is for you

Our AI services differ by what you walk away with. Pick the card that matches what you want at the end.

This page: AI automation services

You get one back-office job done by software, from the email or PDF that starts it to a draft record in your ERP.

AI consulting services

You get a written plan and no software. Pick it when you are not sure which job to give to AI, or whether your vendor's AI already does it.

AI integration services

You get a connection. An AI tool you choose gets its own limited login to your ERP or your CRM, where you keep customers and deals. A person approves what it writes.

AI agent development

You get a custom AI agent, a program that picks its own next step, for one job. We build one when a workflow tool or the vendor's own AI cannot do it.

Also see AI development services and agentic AI development, or every service we offer on the services page.

An AI automation agency or a tool like Zapier

A tool is software your team sets up. An agency is people who build the job for you, and they may use the same tool inside it.

Zapier or n8n: a tool you run

Zapier sells no-code automation across 9,000+ apps. Its free plan runs two-step Zaps, a trigger and one action, and checks for new data every 15 minutes. Zapier also sells AI agents and chatbots, and n8n's hosted plans include AI agent steps. With either tool, your own team builds each flow and fixes it when it breaks.

An agency: people who build it

An agency scopes the job, builds it, tests it on your documents and keeps it running. The older the system, the longer the build. Layer3 Labs puts two cloud apps joined in Zapier at a 4 to 8 hour build, and an older ERP or custom database reached through its API at 40 to 80 hours. An API, short for application programming interface, is the door a system opens for other software.

Scroll the table sideways to see every column.

RouteWhat it costsWhere it fitsWhere it stops
ZapierFree plan with 100 tasks a month. Paid plans start at $19.99 a month.Moving data between cloud apps on a trigger, such as a new form entry creating a CRM contact.Each successful action counts as a task, so a high document volume adds up.
n8nCommunity edition: free to host yourself, but you pay for the server. Hosted plans start at 20 € a month, billed yearly, for 2,500 workflow runs.The same kind of workflow on your own server, with code steps where needed.On the free edition, someone on your team has to host it, update it and watch it run.
A custom build
Our main work
US agencies: $5,000 to $15,000 for one workflow, per Layer3 Labs.Jobs that start with an email or a PDF and need your rules applied before a record is saved in the ERP.Costs more up front than a workflow tool.

Our view: if a two-step Zap does the job, use Zapier and skip the agency. Call one when the job starts with a document, needs your rules on every line and ends in your ERP. Prices as published on the day we read them, 15 November 2026.

The two main types of AI automation

We sort them by who picks the next step: your rules or the AI. Both sit on top of plain automation, which has no AI in it.

The base: plain automation

Fixed steps on a trigger. Zapier's own example: a new lead comes in, and a Slack message goes to a sales rep. No AI reads anything.

Type one: an AI workflow

The steps stay fixed, but one of them uses AI, such as reading a PDF order. Anthropic calls these systems “orchestrated through predefined code paths.”

Type two: an AI agent

The AI picks its own next step and which tool to call. Anthropic says agents “dynamically direct their own processes and tool usage.”

Which one we usually pick

The simplest kind that does the job. Anthropic warns that agents bring “higher costs, and the potential for compounding errors.”

Back-office jobs we automate

Our team has built all four. Most start with a document and end with a draft record in the ERP or CRM. Clients are not named on this site, and we give no numbers from their projects.

RFQs and RFPs, on both sides

A customer emails a request for a quote (an RFQ) or for a proposal (an RFP). The automation reads each line, matches it to your item list and drafts the quote for the sales desk. On the buying side, RFQs go out to suppliers and their replies come back. Our team built this as a system of several AI agents.

Orders into the ERP

A customer's purchase order arrives as a PDF. The automation finds the customer and the items and saves a draft sales order. The same pattern drafts purchase orders to your suppliers.

Daily bookkeeping

The day's bills and receipts become draft ledger entries, ready for someone in finance to check before anything is posted.

Work inside Salesforce and HubSpot

Our team has built AI agents in both CRMs. In HubSpot, those agents run on Claude, the AI model made by Anthropic.

More on the first and last: RFQ and RFP automation, AI agents for Salesforce and AI agents for HubSpot.

Signs a job is ready for an AI automation agency

Four checks we run on a job before anyone talks about a build. A job that fails two of them may suit a plain workflow tool better, or no tool at all.

  • The same kind of document arrives every week. Orders, quote requests, bills or supplier replies. One-off documents do not repay a build. Count them: 200 orders a week at 6 minutes each is 20 hours of typing.
  • The rule for each line can be written down. Which price list, which warehouse, which approver. If only one person knows the rule, write it down first.
  • A person already checks the result. That person can become the approver of the drafts, in the screens they use today.
  • The system has a way in. An API, an MCP server or a database login. MCP, short for Model Context Protocol, is a standard way for an AI tool to use a system's functions. The table below shows the limits for five systems.

How much work an automation can push through each system

Each system caps how many calls outside software may make to it, and every document an automation handles takes a few calls. These are the limits each vendor publishes, with the date we read them.

Scroll the table sideways to see every column.

SystemThe limit an automation works withinRead
QuickBooks Online500 requests a minute for each company, and 10 a second for each company and app. Apps on Intuit's Builder tier get 500,000 CorePlus calls per workspace a month. Calls past that are throttled with error 429.16 Nov 2026
HubSpotFor apps built for one account: 100 requests every 10 seconds on Free and Starter, 190 on Professional and Enterprise. The daily limit, 250,000 to 1,000,000 by tier, is shared by every app in the account.16 Nov 2026
SalesforceAn Enterprise org gets 100,000 calls a day plus 1,000 for each Salesforce user license. A production org allows 25 calls at once that run 20 seconds or longer.16 Nov 2026
NetSuiteEvery integration shares one limit on requests at once. For contracts from June 2020 the base is 5 on Standard, 15 on Premium and 20 on Enterprise and Ultimate, plus 10 for each SuiteCloud Plus license.15 Nov 2026
Sage 200A REST API with OAuth 2.0 sign-in that reads and changes customers, suppliers, products, sales and purchase orders and nominal accounts. On Sage 200 Professional, the customer or its support provider turns the API on and sets who may use it.16 Nov 2026

A worked sum: if one order takes 6 API calls, 200 orders a week is 1,200 calls a week. That sits far inside every daily limit above. The limits to watch are the short ones, calls at once or calls a minute, so count the busiest hour before a build.

How a person stays in control

Four rules worth setting for any automation that writes to an ERP.

Rule

Drafts before posts

Set the automation to save a draft and stop. A person then approves each quote, order or ledger entry.

Rule

A login of its own

Give it a role limited to the records its job needs. An administrator login reaches far more than one job needs.

Rule

Flags, not guesses

Send a line that does not match your item list or price list to a person, in place of the closest match.

Rule

A record of each change

Where the ERP keeps a change log, each change shows with a time. Oracle says NetSuite's System Notes “can't be edited by any user, script or app.”

What AI automation costs

Published ranges, with the date each was read. They describe the market, not our prices.

One automated workflow costs $5,000 to $15,000 to build and takes two to four weeks. A project with three to six connected workflows costs $15,000 to $50,000 and takes six to twelve weeks.

Layer3 Labs, updated 2 Jul 2026, read 15 Nov 2026

Retainers for ongoing automation work run $3,000 to $20,000 a month.

Layer3 Labs, updated 2 Jul 2026, read 15 Nov 2026

Layer3 Labs sells AI automation itself, so read its ranges as one firm's view of the market. Ask for AI usage and hosting as separate monthly lines. Six firms' published price lists sit side by side in what an AI agent costs to build.

Tell us which documents pile up

Name the system you run and the documents your team types in by hand. The first call is free.

Tell us the job

Questions about AI automation

What does an AI automation agency do?

It finds work your team repeats by hand and builds software to do it. AI reads the parts that fixed rules cannot, such as emails and PDFs. Layer3 Labs, an agency itself, says agencies charge in four ways: per project, per month, per workflow or by the hour. We take on back-office jobs that end with a record in your ERP or CRM, and a person usually approves each record.

How is an AI automation agency different from Zapier?

Zapier is a tool you set up yourself. An agency is people who scope the job, build it, test it and keep it running, and they may use Zapier inside it. Zapier's free plan runs two-step Zaps with 100 tasks a month. Layer3 Labs puts two cloud apps joined in Zapier at a 4 to 8 hour build, and an older ERP or custom database reached through its API at 40 to 80 hours. Our view: if a two-step Zap does the job, use Zapier.

How is AI related to automation?

Automation runs fixed steps: when a field says one thing, do another. AI handles the input that varies. It reads an email or a PDF, finds the customer, the items and the amounts, and hands clean data to the fixed steps. In the builds we usually do, the AI reads and your rules decide.

What are the main types of AI automation?

There are two main types, sorted by who picks the next step. An AI workflow keeps fixed steps but uses AI for one of them, such as reading a PDF. An AI agent picks its own next step and which tool to use. Both sit on top of plain automation, which runs fixed steps on a trigger with no AI at all. Anthropic warns that agents bring “higher costs, and the potential for compounding errors,” so the usual pick is the simplest kind that does the job.

How much does AI automation cost?

Layer3 Labs, a US AI agency, puts one automated workflow at $5,000 to $15,000 and a project with three to six connected workflows at $15,000 to $50,000. Retainers for ongoing work run $3,000 to $20,000 a month (guide updated 2 July 2026). On top of the build come the AI model's usage and the server it runs on. Our AI agent cost article compares six published price lists.

How long does an AI automation project take?

Layer3 Labs puts one workflow at two to four weeks, three to six connected workflows at six to twelve weeks, and a whole function, such as finance, at three to six months (guide updated 2 July 2026). Its ranges assume data that is clean enough and standard tools. Layer3 says cleaning your data before the build starts can cut weeks off.

Can I do AI automation for free?

For small volumes, yes. Zapier's free plan includes 100 tasks a month on two-step Zaps, and each completed action uses a task, some more than one. n8n's Community edition is free to host on your own server, though you pay for the server and run it yourself. Oracle says its NetSuite AI Connector Service is not a paid feature, but the AI tool you connect to it may need a paid plan.

Are AI agents worth it?

They are when the hours they save are worth more than the build cost plus the monthly running cost. So count before you build. Take one job, count the documents per week and the minutes each one takes, and price that time. For example, 200 orders a week at 6 minutes each is 20 hours of typing a week.

How many documents a day can an automation handle?

Start from the system's API limit, because every document takes a few calls. QuickBooks Online allows 500 requests a minute for each company, and 10 a second for each company and app. For apps built for one account, HubSpot's daily limit is 250,000 requests on Free and Starter, 625,000 on Professional and 1,000,000 on Enterprise, shared by every app in the account. A Salesforce Enterprise org gets 100,000 calls a day plus 1,000 for each Salesforce user license. Count the calls one document needs before anyone builds.

Do you use Zapier, Make or n8n?

Where they fit. They move data between cloud apps on a trigger, such as a new form entry creating a CRM contact. When a job needs a document read, your rules applied and a record written into an ERP, that part is usually written as code, and a workflow tool can still start it.

How do we choose an AI automation agency?

Ask for a written scope before you sign. It should list each workflow, each system, the tests and what counts as done. Layer3 Labs names a large fixed quote with no written scope as the clearest red flag. Then ask who will build it, what happens when the AI reads a document wrong, and what it costs each month to run.

What happens when the AI reads a document wrong?

In the usual design, the record waits as a draft, so a person can catch it before anything posts. A line that does not match your item list or price list can be set to go to a person in place of the closest match. Where the ERP keeps a change log, each change shows who made it and when.

Will AI automation replace our staff?

It takes over the typing. In the builds we usually do, a person on your team still approves each record the automation saves, such as the sales desk for a quote or finance for a ledger entry. To see the hours involved, multiply the documents a week by the minutes each one takes, before anyone builds.

We're a UK business on Sage. Where does Making Tax Digital fit in?

In your accounting software, which is where an automation's drafts land. GOV.UK says all VAT-registered businesses should now be signed up for Making Tax Digital for VAT. It says some VAT records must be kept digitally, in compatible software or software that connects to HMRC systems. An automation that drafts bills in Sage, Xero or QuickBooks writes into that same software, so the record stays in one place. Which records your business must keep is a question for your accountant.

Which systems can you automate?

ERPs and CRMs. Our team has taken ERPs live on SAP, Oracle NetSuite, Odoo, Microsoft Dynamics, ERPNext, Xero, QuickBooks, Sage and custom ERPs, and has built AI agents in Salesforce and HubSpot. The route in depends on the system. NetSuite uses REST web services, SAP Business One the Service Layer and Odoo its external API or MCP server. Sage 200 has a REST API with OAuth 2.0 sign-in. A custom ERP with no API gets a database login of its own.