# Turn your workflow into an AI agent

> Source: https://ibl.ai/process

Tell us how the work gets done today, in your own words. We turn it into an agent that runs on the systems you already use, and you keep every bit of control.

*For department leads and administrators.*

- About 10 minutes
- No technical knowledge needed
- One workflow per form

[Describe your workflow](https://ibl.ai/process#form) · [Book a 30-minute call](https://cal.com/iblai/30min)

**How ibl.ai turns a department workflow into an agent your team owns** (0:55). A department can hand one workflow to an agent without a data project, a procurement cycle or a line of code from its own staff. ibl.ai is the Agentic AI Operating System institutions and companies run themselves, and this is the process behind it: you describe the workflow in your own words, our team replies within one business day, a 30-minute call settles what the agent may and may not do, and our engineers build it on a safe copy of your data through connectors scoped to each person's role. Your team writes the test set that every version is scored against, then runs the agent in its own cloud, on any model, with the data reached in place behind its own firewall, code it holds and every action logged.

Video: https://ibl.ai/videos/ibl-ai-process-workflow-to-agent.mp4

**ibl.ai turns a department's everyday workflow into an AI agent from a plain-English description: you tell us what kicks the work off, where the information lives and what done looks like, and our engineers connect the agent to those systems in place. The agent runs on your own infrastructure or ours, on any model, and with ibl.ai you own all the code and the data.**

On this page: [The seven stages](https://ibl.ai/process#the-journey) · [Your sector](https://ibl.ai/process#in-your-sector) · [What you will be asked](https://ibl.ai/process#what-you-will-be-asked) · [Describe your workflow](https://ibl.ai/process#form) · [Questions your team will have](https://ibl.ai/process#leader-questions) · [Where the data goes](https://ibl.ai/process#how-we-reach-your-information)

## How does ibl.ai learn your workflow and build with your team?

*How we work with your team.*

Seven stages, from the form below to the next workflow. The people who do the work today take part in every one of them.

1. **You describe one workflow.** Pick the one your customers, patients, students or staff would feel first, and tell us how it runs today in your own words. [Describe your workflow](https://ibl.ai/process#form)
2. **We read it and reply within one business day.** Our team reads every submission and comes back with two or three follow-up questions or a short call invitation.
3. **A 30-minute call with your team.** We walk through the workflow with you and whoever owns the systems, agree what the agent should and should not do, write down the number that will show it worked, and name the person on your side who will run it.
4. **We build on a safe copy of your data.** The first version reads your systems in place through connectors, in a test copy wherever one exists, and the person who does the work today tries it before it touches anything live. [Agentic Ontology connectors](https://ibl.ai/product/agentic-ontology)
5. **Your team sets the bar.** You write down questions with the answers your team agrees are correct. Every version of the agent is scored against them, and one below the bar does not go live. [How agent evals work](https://ibl.ai/updates/agent-evals-benchmarks-llm-judges-csv-export)
6. **You run it.** The agent runs in your environment, and with ibl.ai you own all the code and the data. Skills are plain-English files your team edits without waiting on us, and we train the person who runs it. [The eight-step agentic development process](https://ibl.ai/architecture#agentic-development)
7. **The next workflow.** The connectors, the test set and the skills carry over, so the second workflow starts further along than the first, and your team can lead it.

## Which workflows make good first agents?

The best first agent is a workflow that is frequent, well understood by the person who does it, and bounded by a clear rule for when a human must step in.

- **Transcript and enrollment verification requests** (Registrar): Frequent, rule-bound, and the record already lives in the SIS.
- **New-hire onboarding paperwork** (Human resources): The same ten steps for every hire, across HRIS, email and the equipment desk.
- **Invoice matching and vendor questions** (Finance): Three-way matching is a lookup, and the exceptions are what people should see.
- **Help-desk ticket triage** (IT): Categorize, look up the account, answer the known half, route the rest.
- **Grant and compliance reporting** (Research or operations): Pulls the same figures from the same systems every quarter.
- **Intake scheduling and reminders** (Clinical or client services): Calendar, record and message in one loop, with a person approving changes.

## What does a first workflow look like in your sector?

*What this looks like in your sector.*

Pick your sector for one worked example: the flow from trigger to done, the step a person on your side keeps, and the agent your team ends up owning. Open the card for how the work runs today, the systems it touches and the number you would name.

### Higher Education: Advising holds and registration questions

A student emails about a hold or a closed section, an advisor looks up the record in Banner, checks the degree plan, and replies; about 60 a week in peak registration, 15 minutes each.

1. **What starts it:** A student writes in about a hold or a course
2. **What the agent does:** Reads the record, the hold and the degree plan in place
3. **A person checks:** An advisor approves any override or exception
4. **Done looks like:** The student gets a cited answer the same day

- **Systems it reads:** Banner (SIS), Canvas (LMS), EAB Navigate, Outlook
- **The agent you own:** [Academic Advisor](https://ibl.ai/solutions/higher-education/agent/academic-advisor-agent)
- **The number you name:** Advising questions answered within one day
- [Agentic OS for Higher Education](https://ibl.ai/product/agentic-os/higher-education) · [Higher Education solutions](https://ibl.ai/solutions/higher-education)

### K-12: IEP meeting preparation

Before each review, a case manager pulls the current plan, progress data and accommodations from three systems and assembles a packet by hand; two to three hours per meeting.

1. **What starts it:** An IEP review date approaches
2. **What the agent does:** Pulls the plan, progress data and accommodations into a draft packet
3. **A person checks:** The case manager reviews and edits the draft
4. **Done looks like:** The packet goes to the team and the family on time

- **Systems it reads:** PowerSchool (SIS), Frontline, Google Classroom, ParentSquare
- **The agent you own:** [Special Education Support](https://ibl.ai/solutions/k-12/agent/special-education-agent)
- **The number you name:** Hours of preparation per IEP meeting
- [Agentic OS for K-12](https://ibl.ai/product/agentic-os/k12) · [K-12 solutions](https://ibl.ai/solutions/k-12)

### Enterprise: IT help-desk ticket triage

Tickets arrive in ServiceNow, a technician categorizes each one, looks up the account, resets what can be reset and routes the rest; half the queue is password and access requests.

1. **What starts it:** A ticket lands in ServiceNow
2. **What the agent does:** Categorizes it, looks up the account in Okta, resolves the known half
3. **A person checks:** A technician approves any access beyond a standard role
4. **Done looks like:** The ticket closes, or reaches a person with the context attached

- **Systems it reads:** ServiceNow, Okta, Slack, Confluence
- **The agent you own:** [IT Support](https://ibl.ai/solutions/enterprise/agent/it-help-desk-agent)
- **The number you name:** Tickets resolved without a technician
- [Agentic OS for Enterprise](https://ibl.ai/product/agentic-os/corporate) · [Enterprise solutions](https://ibl.ai/solutions/enterprise)

### Small Business: Quotes and estimates

A customer asks for a quote by email or phone, the owner looks up past jobs and prices, writes the estimate that evening and sends it the next day.

1. **What starts it:** A customer asks for a quote
2. **What the agent does:** Pulls prices and past jobs, drafts the estimate
3. **A person checks:** The owner approves or edits it
4. **Done looks like:** The quote goes out the same day and is logged

- **Systems it reads:** QuickBooks, Google Calendar, Email
- **The agent you own:** [Quote & Estimate Agent](https://ibl.ai/solutions/small-business/agent/quote-estimate-agent)
- **The number you name:** Hours from inquiry to quote
- [Agentic OS for Small Business](https://ibl.ai/product/agentic-os/small-business) · [Small Business solutions](https://ibl.ai/solutions/small-business)

### Government: Permit and service-request intake

Requests arrive by web form and email, staff check each for completeness and eligibility, pull the case history and reply, citing the rule; a backlog builds every spring.

1. **What starts it:** A request arrives by form or email
2. **What the agent does:** Checks completeness and eligibility, pulls the case history, drafts a reply citing the rule
3. **A person checks:** Staff approve anything with a fee or a decision
4. **Done looks like:** The requester gets an acknowledgment within a day and a tracked case

- **Systems it reads:** ServiceNow (case management), Records management, Granicus govDelivery, Microsoft Entra ID
- **The agent you own:** [Citizen Services](https://ibl.ai/solutions/government/agent/citizen-services-agent)
- **The number you name:** Requests acknowledged within one business day
- [Agentic OS for Government](https://ibl.ai/product/agentic-os/government) · [Government solutions](https://ibl.ai/solutions/government)

### Legal: New matter intake and conflicts

A prospective client calls, an assistant gathers the parties and the matter, conflicts runs a screen, a partner reviews, and an engagement letter goes out; four to six days end to end.

1. **What starts it:** A prospective client inquiry arrives
2. **What the agent does:** Collects the matter details, runs the conflicts screen, drafts the engagement letter
3. **A person checks:** A partner clears the conflicts result and signs off
4. **Done looks like:** The engagement letter is signed and the matter is opened

- **Systems it reads:** Clio, Intapp Conflicts, NetDocuments, DocuSign
- **The agent you own:** [Client Intake](https://ibl.ai/solutions/legal/agent/client-intake-agent)
- **The number you name:** Days from inquiry to signed engagement
- [Agentic OS for Legal](https://ibl.ai/product/agentic-os/legal) · [Legal solutions](https://ibl.ai/solutions/legal)

### Financial Services: Client onboarding and KYC

A new account application starts a document chase, a sanctions and beneficial-ownership screen, and a compliance review before the account opens; a week or more for a business client.

1. **What starts it:** A new account application comes in
2. **What the agent does:** Collects the documents, runs sanctions and beneficial-ownership screening, assembles the file
3. **A person checks:** A compliance officer clears any alert
4. **Done looks like:** The account opens with a complete, auditable file

- **Systems it reads:** Salesforce Financial Services Cloud, NICE Actimize, LexisNexis WorldCompliance, DocuSign
- **The agent you own:** [KYC/AML Specialist](https://ibl.ai/solutions/financial-services/agent/kyc-aml-agent)
- **The number you name:** Days to open a business account
- [Agentic OS for Financial Services](https://ibl.ai/product/agentic-os/financial-services) · [Financial Services solutions](https://ibl.ai/solutions/financial-services)

### Healthcare: Prior authorization

An order needs payer approval, a coordinator checks coverage, gathers the clinical documentation from the chart, submits through the payer portal and chases the status; PHI stays in the EHR throughout.

1. **What starts it:** An order needs payer approval
2. **What the agent does:** Checks coverage and eligibility, drafts the submission from the chart
3. **A person checks:** The clinician signs the submission
4. **Done looks like:** The request is filed and its status tracked to a decision

- **Systems it reads:** Epic (FHIR), Availity, Payer portals
- **The agent you own:** [Prior Authorization](https://ibl.ai/solutions/medical-healthcare/agent/prior-authorization-agent)
- **The number you name:** Hours from order to submission
- [Agentic OS for Healthcare](https://ibl.ai/product/agentic-os/healthcare) · [Healthcare solutions](https://ibl.ai/solutions/medical-healthcare)

### Nonprofit: Gift acknowledgments

A gift lands in the donor CRM, a development associate looks up the donor's history, writes a thank-you by hand and logs it; in a busy appeal season the letters fall days behind the gifts.

1. **What starts it:** A gift is recorded in the donor CRM
2. **What the agent does:** Reads the donor's giving history, drafts a personal thank-you
3. **A person checks:** A development staff member approves or edits it
4. **Done looks like:** The thank-you goes out and is logged on the donor record

- **Systems it reads:** Raiser's Edge NXT or Salesforce (donor CRM), Mailchimp, Microsoft 365 or Google Workspace
- **The agent you own:** [Donor Stewardship](https://ibl.ai/solutions/nonprofit/agent/donor-stewardship-agent)
- **The number you name:** Days from gift to thank-you
- [Nonprofit solutions](https://ibl.ai/solutions/nonprofit)

## What will you be asked?

Four short sections. Every question has an example beside it, and nothing asks for a system's technical name.

### Section 1: About you

So we know who does the work and who can say yes.

Example answer: Dana Reyes, Registrar's Office, Lakeview University. I manage the people who do it.

### Section 2: The workflow

A step-by-step story, as if you were training a new hire, plus how often it happens and what goes wrong.

Example answer: Transcript requests: 40 a week, about 15 minutes each. A request arrives by email, we look the student up, check holds, generate the PDF and send it.

### Section 3: Where the information lives

Which systems, files and tools the workflow touches, and how people reach them today.

Example answer: Banner for the record, a shared Excel on SharePoint for the log, Outlook for the request itself.

### Section 4: Rules, risk and success

Which steps need a person's approval, whether anything is sensitive, the one number that will tell you the agent worked, and the person on your side who will run it.

Example answer: A manager approves anything with a hold. Student records, so FERPA. The number is requests answered within a day; today it takes about three. Sam Ortiz, our office manager, will run it.

### Before you start

- **Do I need to involve IT first?** No. Describe the workflow as you know it and name the systems in your own words. We ask IT the technical questions later, and the form has a place to tell us who to ask.
- **What if the workflow spans several systems?** That is the normal case. Each system gets its own connector and the agent works across them.
- **Will the agent act on its own?** Only where you say it may. Every step that needs a person's approval stays a human step, and the form asks you to name those steps so the first version is built around them.

## What happens to the information you share?

Your submission is read by the ibl.ai team only and used to scope the agent. Do not paste passwords, personal records or anything you would not email a colleague; names of systems and a description of the work are all we need.

When the agent is built, it runs where you choose, in your cloud, on-premise or fully air-gapped, and with ibl.ai you own all the code and the data: the skills your team writes, the connectors and the agent itself sit in your own environment, not inside a vendor's product.

## Describe your workflow

One workflow per form. If the form does not load below, open it in a new tab. [Open the form in a new tab](https://docs.google.com/forms/d/1bX6JqxkxbFjJyR1de4ZhbuCTM5xnTcXTJwUvCHNg_yk/viewform)

## What will your team want to know before an agent joins it?

*Questions your team will have.*

Straight answers to the questions we hear from every department, so nobody is surprised later.

### Question 1: “Where do we start?”

With one workflow, the one where people would feel the difference first. Run it, learn from it, then decide what comes next.

That is why the form takes one workflow at a time, and why the best first candidates are frequent, rule-bound work.

[Which workflows make good first agents?](https://ibl.ai/process#good-first-agents)

### Question 2: “Can't we just build this ourselves?”

For reports, research and drafting, often yes, and we encourage it. The hard part begins when the work touches a system of record, personal data or payments, or has to run every day without someone watching it.

That is what ibl.ai's engineers build with your team: connectors scoped to each person's role, with every action logged.

[How ibl.ai works with your team](https://ibl.ai/service/ai-transformation)

### Question 3: “How will we know it worked?”

By a number your team names before anything is built: requests answered within a day, hours per onboarding, days to close. The form asks for it, the call writes it down, and we read it together after the first month.

[What will you be asked?](https://ibl.ai/process#what-you-will-be-asked)

### Question 4: “How do we know the AI is right?”

Your team writes a test set: questions with the answers it agrees are correct. Every version of the agent is scored against it, a version below the bar does not ship, and every step you mark for approval stays with a person.

[How agent evals work](https://ibl.ai/updates/agent-evals-benchmarks-llm-judges-csv-export)

### Question 5: “Who owns it and keeps it running?”

You do. On ibl.ai you own all the code and the data: the agent runs in your cloud or on-premise, and its skills are plain-English files your team can read and change.

We name the person on your side who will run it before we start, and we train them before handover.

If it works, your team decides who improves it and what to do with the time it gives back. Analytics show usage and cost per agent, and the next workflow starts from the connectors and skills already built.

[The eight-step agentic development process](https://ibl.ai/architecture#agentic-development)

### Question 6: “Is our data safe?”

It stays in your environment: your cloud, on-premise or fully air-gapped. Each agent can redact, mask or block personal information before the model sees a message, and Incognito keeps a conversation out of history and memory.

Ask us where the data goes, who can see it and what the model provider keeps, and we answer all three in writing.

[Incognito and PII filtering](https://ibl.ai/updates/incognito-pii-filtering-agent-privacy)

### Question 7: “What happens to our people?”

They lead. The person who does the work today describes it, tries the first version and approves what goes live. Skills are written in plain English, so the people who know the process edit them, and a hands-on workshop teaches your team to build the next agent.

[How ibl.ai works with your team](https://ibl.ai/service/ai-transformation)

### Question 8: “Which model should we use?”

Whichever fits today; you will change it. ibl.ai is model-agnostic, so an agent picks from dozens of providers and a switch keeps its prompts, data and connectors. What your team keeps is the workflow, the test set and the data.

[Any model, one pick away](https://ibl.ai/updates/llm-catalog-45-providers-any-model)

## How can ibl.ai reach the information your workflow uses?

Only the data the first workflow touches is involved, read in place through a connector scoped to each person's role, so no warehouse project comes first. You do not need to know how we reach a system, but if you do, say so: it saves a meeting.

### Where does the data go?

Nowhere new. The agent reads each system in place, inside your environment, and a person approves the steps you mark.

**Your systems** (Your environment): SIS, EHR, CRM, ERP, files: left where they are → **Connectors** (Agentic Ontology (MCP)): One per system, scoped to each person's role → **The agent** (Your cloud, on-premise or air-gapped): Runs in its own sandbox, on the model you choose → **Your team** (Every action logged): Approves the steps you mark; edits the plain-English skills

**The test set:** Your team's questions and correct answers score every version before it goes live

### For your IT team: the five ways we reach a system

| Route | In plain words | What to tell us | A sign it exists |
|---|---|---|---|
| API | The system offers a door for software to read and write through, with your permission. | The vendor's name and whether anyone has used its API before. | A vendor page called Developers or Integrations; an app already connected to it. |
| Database query | Your IT team can ask the system's database directly, usually read-only. | Who on the IT team runs reports or queries today. | Reports that arrive from IT as spreadsheets nobody clicked through a screen to make. |
| Scheduled export or data dump | The system drops a file on a schedule: nightly, weekly, at term end. | What the file is, where it lands and how often. | A CSV or Excel that appears in a shared folder every morning. |
| Browser automation | There is no door, so the agent logs in and clicks the same screens a person would. | Which screens, and whether a second sign-in step is involved. | A vendor portal with a login and no export button. |
| Staging or test environment | A safe copy of the system to practice in before anything touches live data. | Whether one exists and who can grant access to it. | An address with test, staging or sandbox in it that IT uses before upgrades. |

Whichever route applies, the data is read in place through a connector scoped to each person's role. Nothing is copied into a warehouse first. [See the eight-step agentic development process](https://ibl.ai/architecture#agentic-development) and the [Ontology connectors](https://ibl.ai/product/agentic-ontology) that do this work.

## Prefer to talk it through?

Book a 30-minute call and describe the workflow out loud; we take the notes.

- [Book a 30-minute call](https://cal.com/iblai/30min)
- [Contact us](https://ibl.ai/contact)
