Automation and Operations
Which processes to automate first: a simple scoring method
Score each process on hours, clarity, data readiness and risk, check payback, then automate the top one or two. Worked example and pass/fail test.
Workflow Automation
AI Browser Automation Agent
Multi-step actions completed on websites with no API, from form fills to submissions, with approval built into the steps that need it.
Some of your most important systems have no API: supplier portals, government sites, older booking tools. So someone on your team logs in and clicks through the same screens every day. This agent operates a real browser to do that work across up to six workflows, filling forms, moving between pages and submitting data. It pauses for your approval wherever you say, and logs every run.
- Clears repetitive portal and form work across up to six workflows
- Pauses for a named approver at the steps you choose
- Stops and reports when a page changes instead of guessing
$9,700
20–30 working days
Optional maintenance $485/mo
What it is
AI Browser Automation Agent operates a real browser to complete multi-step tasks on websites that do not offer an API, logging in, filling forms, navigating between pages, and submitting data, pausing for your approval wherever you set an approval rule. Every run is logged with the time, the steps taken and who approved it.
Why businesses buy it
Copying data into portals that have no API is slow, dull work, and it tends to land on capable people who should be doing something else. Mistakes creep in when attention drifts, and the work stops when that person is away. With this agent, those clicks happen on schedule, every run is logged with a time and an approver, and failed steps are reported rather than skipped. The written scope sets out the workflows, systems and approval points before you order, so you know exactly what will be automated and where a person stays in charge.
Problems this solves
Staff retype the same data into portals that offer no API
The agent operates a real browser to log in, fill forms and submit data on sites without an API. Up to six workflows can be set up for these repetitive tasks.
Manual web form entry leads to typos and missed fields
The agent follows the same steps on every run, using data from your connected systems. Fields are filled consistently, and you can add an approval pause before anything is submitted.
Nobody can say who submitted what on an outside portal
Every run is logged with the time, the steps taken and who approved it. You get an audit trail for each portal task instead of relying on memory.
Automating a portal task feels risky without a human check
Approval rules let you decide where the agent stops and waits for a person. Sensitive submissions can require sign-off, while routine steps carry on without anyone watching.
A good fit if
You have a website-only process, a supplier portal, a government form, a legacy system with no API, that someone on your team repeats by hand.
The task involves more than one step: logging in, navigating, filling fields across more than one page, and submitting.
Some of those steps are safe to complete automatically, and others, like final submission, should wait for a person's approval first.
You can define up to six distinct workflows and want them all running through one agent with a consistent approval layer.
Not the right fit if
You only need one specific, narrow task repeated on a schedule, not a broader capability spanning several workflows. See AI Web Task Agent, which is scoped to that.
You need the site watched for changes, not acted on. See AI Website Monitoring or AI Competitor Website Monitor.
The work is research, not action, gathering and comparing information rather than submitting it. See AI Supplier Research or AI Market Research Agent.
Your process needs more than six workflows or write access to more than two systems. Add a note at checkout for a Change Quote, or use our Custom AI enquiry form for larger builds.
How it works
- You define the workflows and the rules. Up to six browser-based workflows are mapped out during intake, each with its own steps and its own approval rule.
- The agent drives a real browser. It logs in, navigates between pages, and fills fields exactly as a person would, on sites that have no API to connect to directly.
- Approval gates sit where you put them. A step you mark as needing sign-off pauses and waits for the approver you named, rather than submitting on its own.
- Approved steps complete the task. Once approved, or automatically for steps you did not gate, the agent finishes the submission, download or update.
- Every run is logged. What was done, on which site, at what time and under whose approval, is recorded so you can audit any run afterward.
- Exceptions are flagged, not guessed at. If a page layout changes or a step fails, the agent stops and reports it rather than continuing on a broken path.
What you provide within 72 hours
Your portal lists these the moment you order. Delivery starts the working day after everything is in and your access is checked.
- Admin or API access to the connected systems, with sandbox or test records if you have them.
- Your rules and playbook for each workflow, such as approval limits and what counts as an exception.
- Ten real examples of the task being done, such as screen recordings, screenshots or a written walkthrough.
- The name of the approver for any step that needs a person before it goes ahead.
- Admin access and any API keys for connected systems, shared by adding our team as a user or creating a key you control, never by email.
- Your business information (company basics, hours and brand voice), shared during intake.
- One decision-maker who can approve the workflow rules during intake.
- An API key for your own AI model account (for example OpenAI or Google Gemini), created for this build with a spending limit.
The 72-hour rule. Send everything on your checklist, including access, within 72 hours of ordering so your build starts on time.
Timeline
Start
Intake confirmed and up to six workflows, with their approval rules, agreed with you.
Weeks 1–2
Browser navigation, form logic and approval gates are built for each workflow.
Week 3
The connected systems are linked with read and write access and tested against sandbox or live records.
Weeks 3–4
Fifty documented test scenarios run across the workflows, and you receive the results.
Week 4
You review the test runs and we complete two rounds of changes to steps, rules or approval points.
Go-live
The agent goes live, the written handover report runs, and 30 days of support begins.
| Option | Delivery time | Price |
|---|---|---|
| Standard | 20–30 working days | $9,700 |
| Express | 15–25 working days | $11,150 |
| Rush | 10–20 working days | $12,600 |
What you receive at handover
- AI Browser Automation Agent deployed in your environment with your approval rules
- Workflow, prompt and guardrail configuration for up to six web tasks
- Read and write connections to up to two agreed systems
- A test report covering the agreed scenarios
- An admin guide and a written handover report
- 30 days of post-launch support
Expected benefits and how to measure them
- Can free back-office staff from repetitive portal work.
- Helps keep a clear log of every web submission and approval.
- Makes it easier to apply the same checks on every run.
- Staff hours on portal tasks
- Log weekly hours on the chosen tasks for a month before launch. Compare with a month after.
- Submission errors
- Count rejected or corrected submissions each month from portal confirmations or your own records. Compare before and after.
- Runs completed without help
- Use the agent's run log to compare completed runs with runs that needed a person to step in.
Before you start: List the web tasks you plan to automate, how often each runs and how long it takes. Note any recent errors or rejected submissions.
These are things to track, not promised results. Outcomes depend on your data, your team and your market.
Running costs
Running this month to month costs roughly $50 to $250, covering AI model usage and hosting, on top of whatever plan you already pay for the connected systems themselves. Volume of runs and how many steps each workflow has both move that figure.
The build price is one-time. Usage during build, testing and afterward bills to your own accounts, so every provider relationship stays in your name.
Ask about running costsCustom requirements
If you need more than six workflows, write access to more than two systems, or a workflow that spans more than one browser session per run, add a note at checkout. You get a written Change Quote. Your original order keeps its normal timeline either way. Larger or bespoke projects go through the Custom AI request form instead.
Request a custom versionKeep it running
For this build we recommend monthly maintenance, with a monthly allowance of change hours for adjusting steps or approval rules as the sites you automate change their layout. It starts once the 30 days of included support ends.
$485/mo (5% of the build price a month, includes 4 change hours)
Optional. Starts after the post-launch support period. Cancel any month.
How maintenance worksHow this compares
| Do it yourself with a SaaS tool | Hire a custom agency | Upmark Labs | |
|---|---|---|---|
| Price | A browser-automation platform charges monthly per bot or per run, on top of your own build time | An agency quotes automation work after a discovery call | Ours is one fixed price, shown on this page before you buy |
| Approval | A no-code automation tool usually runs straight through unless you build the gate yourself | An agency's approval handling depends on what was scoped | Ours has approval rules built into the workflow from day one |
| Scope | A platform charges more as you add workflows or steps | An agency's workflow count is whatever was agreed | Ours covers up to six workflows in the fixed price |
| Testing | A platform leaves testing to you | An agency's testing depends on the project | Ours runs 50 documented test scenarios before you see it live |
Questions about this product
What kind of tasks can AI Browser Automation Agent handle?
Multi-step tasks on websites without an API, logging in, filling forms across more than one page, navigating a portal, and submitting data. Up to six distinct workflows fit in this build, each with its own steps and its own approval rule, so a supplier portal task and a government form task can both run through the same agent.
Is AI Browser Automation Agent different from AI Web Task Agent?
Yes. AI Web Task Agent is scoped to one well-defined task, done reliably on a schedule, with variations of that one task. This build covers up to six distinct workflows across different processes, such as supplier portal updates, government form filings and order checks. Choose this one when several separate browser jobs need automating, and the Web Task Agent when there is one clear job.
How does the approval step work in AI Browser Automation Agent?
You mark specific steps, such as a final submission or a payment action, as needing sign-off. The agent completes everything up to that point, then pauses and waits for the approver you named during intake before finishing the step. Steps you do not gate complete automatically without a pause.
What happens if a website changes its layout on AI Browser Automation Agent?
The agent stops and reports the failure rather than guessing at a broken page. Layout changes are the most common reason a browser workflow needs a fix, and small adjustments like this fit inside the monthly change hours that come with maintenance, which you can cancel any month.
How much does AI Browser Automation Agent cost to run each month?
Roughly $50 to $250 a month for AI model usage and hosting, on top of whatever you already pay for the systems it connects to. You pay these directly in your own accounts. The build price is a single one-time payment that covers setup, testing and launch, with no monthly fee from us.
Can AI Browser Automation Agent write data into my systems, not just read it?
Yes, up to two connected systems with read and write access are included in this build. Which systems, and what the agent is allowed to write, are agreed during intake and tested against sandbox or test records before going live.
Can I add a seventh workflow to AI Browser Automation Agent later?
Yes, through a Change Quote priced before the work starts. Adjustments to the six workflows already built, such as new fields or a changed approval rule, fit inside the monthly change hours that come with maintenance instead of a full Change Quote.
What if a workflow does not work right when AI Browser Automation Agent launches?
You get two rounds of changes after the 50 documented test scenarios run, before launch. After delivery you have five working days to report anything that does not match the agreed workflows, fixed as part of the original price, not as a Change Quote.
Guides on this topic
All guides Get the browser tasks off your team's desk, with approval built in.
Order AI Browser Automation Agent and up to six browser workflows run on schedule, each with the approval rules you set and a full log of every run. The scope is written down before you pay.
Works well with
Workflow Automation
Zapier or Make to n8n Migration
Moves up to 10 Zapier or Make workflows to n8n Cloud or your own server, runs old and new side by side…
Workflow Automation
Automation Care and Build Plan
Each month: one new workflow of up to 12 steps across 3 systems, or 12 hours of changes, plus monitoring of up…
Workflow Automation
AI Report Automation for Nonprofit
Assembles recurring reports from multiple sources on a schedule, adapted for Nonprofit teams that prepare donor, board and funder reports with limited…