Skip to content
Yonge solutions home

AI service

Custom AI agents that work inside the systems you already use

A custom AI agent is a program that takes one recurring job off the staff, such as answering web inquiries or updating the CRM after a call. We build it around the workflow as it runs today and connect it to the systems you already use; anything that sends a message, spends money or changes a record waits for a person to approve it. You get a working agent, tested on real past cases, that logs every action it takes.

Book a consultationRead the FAQ

Deliverables

What you get

  • 01

    Workflow map

    A written description of the job: where requests come in, what the agent may do on its own, what waits for approval, and when it hands the case to a person.

  • 02

    Connections to the systems you use

    Integrations with the tools the workflow touches: CRM, email, calendars, shared drives, spreadsheets, ERP or POS, the website, and messaging apps such as WhatsApp where the platform offers an API. We connect through your own accounts, so the data stays where it already lives wherever the tools allow.

  • 03

    Approval steps

    Anything that sends a message to a customer, spends money or changes a record goes into a queue for a person to approve. You decide later which actions, if any, can run without it.

  • 04

    Evaluation on real examples

    Before launch the agent works through a set of real past cases, and we compare its output with what staff actually did. You see where it went wrong and what we changed.

  • 05

    Action log

    Every step is recorded: what came in, what the agent decided, which system it called and who approved it. When a result looks wrong, you can trace how it happened.

  • 06

    Handover and monitoring

    Documentation, admin access and a walkthrough for the people who will own the agent, followed by a monitoring period in which we read the logs with you and adjust instructions and rules.

How it runs

The steps, in order

  1. 1

    Pick one workflow

    We talk to the people who do the work now and choose a job with clear inputs and a clear definition of done. Starting with one job gets something useful running before the scope grows.

  2. 2

    Collect real examples

    We gather past cases, such as emails, forms and documents, together with the outcome the team reached. They become the test set, with anything that should not leave the company's systems removed first.

  3. 3

    Build and connect

    We choose a model for each step, write the agent's instructions and tools, and connect it to the systems it needs, starting with read-only access or a test account.

  4. 4

    Test, then launch with approvals on

    The agent runs against the test set until you accept its results. It goes live with every consequential action waiting for a person.

  5. 5

    Hand over and monitor

    Your team gets the documentation and admin access. During the agreed monitoring period we review the log with you and fix the cases it gets wrong.

What we build with

  • Claude
  • OpenAI
  • Gemini
  • Google Vertex AI
  • Google Cloud Vision
  • Python
  • Node.js
  • PostgreSQL
  • MongoDB
  • AWS

Case studies

Where this service did the work

Commercial real estate

Example scenario

An inquiry agent that answers listing questions within minutes

A commercial brokerage in the GTA

First reply to a portal inquiry
< 5 min
First reply to a portal inquiry
Tours booked from portal leads
1.5×
Tours booked from portal leads

The brokerage lists office and industrial units on several portals, and every inquiry lands in one shared inbox. Brokers reply between showings, often the next day, and most first replies repeat what is already on the listing sheet.

  • Read portal lead emails and website form submissions from the shared inbox the team already uses.
  • Answer only from each active listing sheet: size, net rent, additional rent, zoning and possession date.
  • Ask about size, use and move-in date, then offer tour times from the listing broker's calendar.
  • Log every conversation and tour in the CRM, and pass rent negotiation and anything not on the sheet to the broker.
Custom AI agents

First 90 days

Financial services

Example scenario

An intake agent that collects documents before the first call

A mortgage brokerage in South Florida

Less time to a complete document file
3 days
Less time to a complete document file
Fewer status calls to the office
35%
Fewer status calls to the office

Each file needs pay stubs, W-2s, tax returns, bank statements and the purchase contract, and an assistant chases them by email one at a time. Borrowers also call to ask where their file stands, and someone opens the file to read out what it already says.

  • Ask the questions on the brokerage's intake checklist, in wording its compliance officer approved, and book a call with a licensed loan officer.
  • Send each borrower a list of what is missing, check that every upload is the right document for the right tax year, and file it in the document portal.
  • Send reminders on the schedule the file owner sets, and flag anything unclear to a person.
  • Answer status questions only from the milestones staff mark in the CRM.
Custom AI agents

First 6 months

Industries

Where it fits best

Commercial real estate

Portal inquiries answered from your listing sheets within minutes, with the tour booked on the broker's calendar.

Financial services

Intake and document chasing inside your compliance rules, so licensed staff spend their day on files that are ready.

Education

Trial-class questions answered in English and Chinese on WeChat and your website, with the trial booked into your timetable.

Auto dealers

Every listing-site lead answered from live inventory within minutes, with the test drive on a salesperson's calendar.

Dental and medical clinics

Booking, reminders and recall through your practice software, with patient data kept in the systems you have approved.

Building materials

Spec and stock questions answered from your own data, and contractor lists turned into draft quotes your sales desk approves.

FAQ

Custom AI agents: common questions

How much does a custom AI agent cost?

The build cost depends on how many systems the agent connects to, how usable their APIs are, and how much testing the workflow needs before it is safe to run. Model usage is a separate running cost, paid to the AI provider, and it grows with volume. We quote after the workflow mapping, once the systems and the number of cases are known.

Which model do you use: Claude, GPT or Gemini?

We choose per step, not per project. Reading a long contract and sorting a high volume of short emails are different jobs, and the models differ in accuracy and price on each. We test candidates on real examples from the job, and a step can be moved to another model later if prices or quality change.

What happens when the agent gets something wrong?

It will, sometimes. That is why anything that sends, spends or changes a record needs a person's approval, and why every action is logged. When a mistake turns up, we find it in the log, add the case to the test set, adjust the instructions or rules, and re-run the tests.

Where does our data go, and is it used to train AI models?

The agent reads from and writes to your own systems, and data stays in your accounts wherever the tools allow. Text sent to a model goes to that provider's API; Anthropic, OpenAI and Google state that their paid business APIs do not train on your inputs by default. We check the current terms with you before anything is connected, and we leave out fields the workflow does not need.

Who owns the agent once it is built?

Your data stays in your systems and remains yours. The code, prompts and configuration written for the project are handed over with documentation and admin access, so another developer could maintain them. Ownership is written into the contract before work starts.

How long does it take to build an agent?

It depends on the number of systems involved, how quickly we get access and example data, and how many rounds of testing the workflow needs. One workflow with one or two integrations moves much faster than an agent that spans several departments. We give a schedule after the mapping step.

Other AI services

Related services

AEO, GEO & SEO

Technical SEO plus the work that gets a site cited in ChatGPT, Perplexity, Gemini and Google AI Overviews, measured on real buyer questions.

Marketing automation agents

A content pipeline that drafts platform-native posts in the brand's own voice, scores them, and publishes once a person approves.

AI training for teams

Staff learn to use AI tools on their real work, with a written policy on what data goes where and a separate track for managers.

Tell us about the workflow that costs you the most time

A first call is about the business you run, not a pitch. We will tell you what an agent could handle, what it should not, and what it would take to build.

Book a consultation