Hand the repetitive work to AI agents. Free up more than half of your team's time.
We map how the work actually moves through your team, delete the steps that shouldn't exist, and run AI agents inside the tools you already use. The decisions stay with your people.
Led by the founder of an AI company built before ChatGPT and acquired by a San Francisco firm.
A week, before and after. The same job, with the repetitive work handed to an agent.
Pick a team. The first calendar is a typical week today, with every task in it. The second is the same week once an agent takes on the repetitive work. The decisions stay where they were, and the empty slots are work that no longer needs a person.
Your process document is not your process.
The real one has rework loops, people chasing each other for answers, and long stretches where a case just waits. A request can sit in a queue for days and need only minutes of actual work. Automate the documented version and you automate the wrong thing.
The waiting is the process. The work is minutes.
How we find the real process
We look at the work from three sides, because each one shows something the others miss.
-
Interviews: the why
We talk to the people who do the work: who decides, which workarounds they rely on, and the thresholds nobody wrote down.
-
Your systems' logs: the what
Your ERP, CRM and helpdesk record every case, so we can see how many arrive, how long each step takes, where cases wait and which ones loop back.
-
Everything else
Procedures, spreadsheets, email threads and chat hold the exceptions and rules your systems never captured.
Sort every step
In that order. Most teams start with the model.
-
Delete
Steps that exist because of an old workaround, a duplicate check or a report nobody reads. A deleted step costs nothing to run.
-
Plain code
Steps with fixed rules: copy this field, send that reminder, match these two amounts. Ordinary software does them reliably and cheaply.
-
An agent
Steps that need reading and judgement within clear limits, such as an unstructured email, a mismatched invoice or a missing document. The agent prepares the case.
-
A person
Decisions with consequences: approvals, exceptions and negotiations. The agent hands them over with the context attached, and a person makes the call.
Supplier invoices, step by step
How the method sorts a typical accounts-payable process.
- Print each invoice for a paper approval file Delete
- Copy the invoice into a second tracking spreadsheet Delete
- Read the PDF invoices that arrive in the shared inbox An agent
- Extract the supplier, amounts and PO number An agent
- Match the invoice to its PO and goods receipt Plain code
- Work out why a match failed and gather the evidence An agent
- Send approvers a reminder Plain code
- Approve an exception above the tolerance A person
- Build the payment-run file Plain code
- Approve the payment run A person
Illustration of the method, not a client engagement.
How an engagement works
You give us access to the work and the people who do it. We do the mapping, the building and the measuring.
-
Map the real process
Interviews, system logs and documents show how the work really moves, including the loops and the waiting.
-
Set the baseline
Before we build anything, we measure volume, cycle time, waiting time and exceptions, so the result can be compared with today.
-
Sort and build
We delete what shouldn't exist, write plain code for the fixed rules and build agents for the steps that need judgement. Each process has one owner on your side.
-
Measure and expand
We measure the same numbers again. Once the process holds, we move on to the next one.
Built on the systems you already run
You keep your ERP, your CRM and your helpdesk. Agents connect to them and work by your rules.
-
Connected to your stack
Agents work through the APIs of your ERP, CRM, helpdesk and data platforms, with only the permissions you grant.
-
A person makes the call
Agents prepare each case and stop at the decision. Approvals and exceptions go to the person who owns them.
-
A record of every action
Every decision and change is logged, so your team and your auditors can see what happened and why.
More than 20 years in software. Building AI since before ChatGPT.
I lead the agency. Before ChatGPT, I founded an AI company that built models of every kind to automate fraud and security work. A company in San Francisco acquired it, and I went on to work there.
There we built AI, and the processes around it, where it had the most impact. We adopted new advances while they were still at an early stage and reached very high levels of automation.
The agency brings that experience to your operations. We've worked on processes of every kind, and each engagement starts the same way: finding out how the work really runs.
Questions we hear on first calls
What happens on the discovery call?
How long until the first agent is live?
What can the agents actually do?
Do the agents work with the tools we already have?
How much of our team's time will it take?
How do you handle security and data?
How do you know it worked?
How does pricing work?
Free up more than half of your team's time.
Agents take over the repetitive work: re-keying, chasing, matching. Your people keep the decisions. Bring one process to a call and we'll show you how much of it an agent can take.