Skip to content
All service tracks

AI agents for incoming work

AI agents for incoming work: intake, sorting, CRM, follow-through. First processes in 2–4 weeks. Working systems, not another tool to manage.

Fee
$2,500
Length
1–2wks

What the engagement covers

Scope of work6
  • 01Intake on every channelWhatsApp, email, web forms, the CRM inbox and the phone arrive in one queue with one owner, instead of five places nobody is accountable for. A call is answered, worked out and written up the same way a form is.
  • 02Sorting and routingThe request is read, classified and sent to the person or the system that handles it, with the reason attached.
  • 03Write-back to the CRM you runRecords created and updated in the system already in place. No second database, no migration.
  • 04Follow-throughChasing, reminders and status replies happen without anybody remembering to send them.
  • 05Limits set before anything answersMoney, complaints and commitments you do not control go to a person by configuration, not by how confident the model is.
  • 06HandoverDocumentation, monitoring and the runbook, so the workflow keeps running when we are not here.

What you get out of it

  • First workflows in 2–4 weeks, not months
  • One queue instead of five channels, with one owner
  • One queue across WhatsApp, email, forms, CRM and the phone
  • You own the code and the data — leave whenever you like
Where this sits7
  • Visibility AnalyzerfreeFree
  • Free AI DiagnosticfreeFree
  • Managed Automation$750/mo
  • Visibility Upkeep$900/mo
  • Visibility Sprint$1,500
  • Discovery Sprintyou are here$2,500
  • Implementation Sprintfrom $5,000
How it can end1–2

1–2

weeks of measuring

the arithmetic
  • it adds up · we build
  • it does not · money back

Both exits are real. Only one of them is paid for.

Two ways to get the agent running

You can build this in-house; the parts have existed for years. The difference is how many of them have to be wired together and who fixes it when one moves.

Do it yourself

  • Bring every channel into one queue: mail, WhatsApp, forms and the phone all arrive differently and break differently
  • Write down what the agent may decide and what it may not, as a rule rather than as trust in the model
  • Wire the write-back into the CRM you already run, without a second database and without a migration
  • And keep somebody who repairs it the week a provider changes a webhook format
Discovery Sprint

Hand it to us

  • One to three processes in production in two to four weeks, not a platform and not a pilot in a demo account
  • The boundary to a person is set by rule before the first reply, not after the first complaint
  • Documentation, monitoring and a runbook at the end, so it keeps working without us

Not for you if

Where this is the wrong buy3
  • You need a pilot to show the board. We put processes into production; a demo account costs the same and handles no real request.
  • You want the agent deciding money and complaints on its own. Those go to a person by rule, and that is a condition rather than a setting.
  • There is nowhere for it to write yet. An agent needs a system to put the record in; without one you need a CRM before you need an agent.

Three requests, from where they land to where they stop

The same sequence every time, on the kinds of request most businesses get every day. What changes between them is where the line to a person sits, and that line is decided per process while the work is in front of us.

A form on the site4

Someone fills in the contact form out of hours. A name, a phone number and one sentence about what they want.

  1. 01Reads the request and files it as a new enquiry: nothing has ever arrived from that address before.
  2. 02Opens a card in your CRM carrying the sentence the person wrote and the page they sent it from.
  3. 03Answers within the minute, naming what happens next and when, so nobody waits on silence until morning.
  4. 04Chases once on the second day if nobody has replied, and the chase lands on the same card.
  5. Goes to a personThey ask what it will cost. Anything outside the range you publish goes to whoever owns the queue, with the thread attached.
An email about an order4

A customer writes to ask where their order is. No order number, no reference, and the subject line says nothing.

  1. 01Finds the order by the address the email came from, so nobody has to ask a customer for a number they do not have.
  2. 02Reads the status out of the system that actually holds it, not out of a copy that may be a day behind.
  3. 03Answers with that status and the date the order is expected.
  4. 04Updates the card, so the next person to open it sees that the question was answered and what was said.
  5. Goes to a personThe order is late, or they want money back. Refunds and complaints go to a person by rule, whatever the agent could work out on its own.
A call nobody could pick up4

The phone rings while everyone is with a customer. Today that is a missed call and a number on a screen.

  1. 01Picks up, and takes the reason for calling in the caller's own words.
  2. 02Says when somebody will call back, so the caller is not left guessing whether the message landed anywhere.
  3. 03Opens the same card the form opens, and the queue keeps one shape whatever channel fed it.
  4. 04Sends it to the person who handles that kind of call, and writes down why.
  5. Goes to a personThey want a commitment on a date. A promise you have to keep is not the agent's to make, so it goes to the person who can keep it.

Asked before anyone buys this

How is an AI agent actually developed?
The order is intake, rules, systems, handover. What arrives and where from; what the agent may decide and what it may not; which systems it reads and writes; and the line where a person takes over with the conversation attached. The model is the last thing chosen, not the first.
What does an agent do that a script does not?
A script follows a path and stops when the request leaves it. An agent works out what is being asked, does the part it is allowed to do, and hands over what it should not decide. The difference shows on the requests that do not fit the form.
How long before something is running?
Two to four weeks for the first one to three processes in production. Not a platform, and not a pilot that lives in a demo account — the processes that arrive most often, running against the systems you already use.
What does it cost?
Discovery is $2,500 and decides which processes are worth it; the build starts from $5,000. Running it afterwards is $750 a month if you want us to keep it.
What do you need from us to start?
Access to the channel the work actually arrives in, one person who can explain how it is handled today, and an hour to walk a real case end to end. It runs on the systems you already have: your inbox, your CRM, your phone number.
What happens after the handover?
You own the code and the data, so the workflows keep running whether we stay involved or not. Monitoring and an SLA are a separate monthly subscription; leaving after the handover is a supported outcome, not a penalty.
How do I choose between AI agent development companies?
Ask three things: who owns the code and the accounts when the engagement ends, what happens to the cases the agent cannot decide, and whether you can see one process running rather than a demo. An engagement that ends in a report has automated nothing.
What happens to the cases the agent cannot decide?
They go to the person who owns the queue, which is why the queue gets one owner in the first place. Where that line sits is decided per process during the build, with the work in front of us.
Next step

Put a number on one of these processes

Thirty minutes on one process from this track: which steps cost you most, what we would take on first, and whether the arithmetic closes. The sprint that follows is refunded in full if it does not.

Put a time in the calendar

Thirty minutes, free. The sprint is what the call is about.

Fee
$2,500
Length
1-2 weeks

Refunded in full if we conclude you should not build.

Or start with the free 15-minute diagnostic