Custom AI agent for one process
Pick one process. An AI agent runs it, one week from the kickoff workshop.
You pick the process. We map it, measure what it costs to run today, and put an agent behind it that reads the same input your team reads. Live five working days from the workshop, running at least 75% below the cost we measured together, or you get your money back.
One week from the kickoff workshop. Five working days, counted from the day the process is mapped and signed off. The clock does not start at the first email.
The baseline is measured before anything is built. What the process costs to run today is written down in the workshop and goes into the contract.
The 75% Guarantee. Late, off the written requirements, or short of a 75% cost reduction, and you get your money back.
The process that eats a morning, every morning
Something arrives. An email, an order, a PDF, a form. Somebody reads it, works out what it means, and types the answer into a second system that already holds most of the information.
The work is simple and it never ends. It has to happen before the real work starts, and it takes somebody who understands the business. So it lands on the person who understands the business best, and that person becomes the reason the process cannot go any faster.
A second pair of hands doubles the throughput and doubles the cost. The volume keeps growing anyway, because the business is growing, which was the whole point.
Why one week is a real number
Nothing is trained. A general model is placed behind one mapped process and given your rules, your examples and access to the two or three systems the process touches. That removes the data-collection phase and the training budget, which is the difference between one week and one quarter.
The model reads and decides. The code that writes into your systems is ordinary deterministic software, so it fails loudly and leaves a log. The agent is tested against cases from the last few months where the right answer is already known.
It gets a short list of things it is allowed to do and refuses everything else. Anything below the agreed confidence line goes to a person.
What you get
One process, one fixed scope, agreed in writing before the build starts.
A process mapping workshop, two to three hours. With the people who actually run the process. It ends with the map and the numbers: how long each step takes, how often it goes wrong, what it costs to run. Those numbers become the baseline for the guarantee.
The build sprint. The agent itself, built against the written requirements and tested on your real cases from the last few months.
A presentation and handover session, one to two hours. How to hand the agent work, how to read what it decided and why, and when to take a case back.
Agent platform access in your name. The accounts, the prompts and the rules belong to you. Another supplier can take this over without asking us for anything.
Written documentation. What the agent does, what it is allowed to touch, what it escalates, and what to do on the day it stops.
What the week looks like
Day 0
The process mapping workshop. Two to three hours. The process is mapped, the baseline is measured, the requirements are signed. The clock starts here.
Days 1 to 3
The build. The agent reads real cases and is corrected in front of you.
Day 4
Testing against cases where the right answer is already known, and the corrections that come out of it.
Day 5
Handover. Your team runs it on live work, with the documentation and the accounts in your name.
Five working days, counted from the kickoff workshop, the day the process is mapped and the requirements are signed off. If access to a system the agent needs is outstanding, the clock pauses until it is granted, and you are told on the day it happens.
The 75% Guarantee
Live inside one week of the kickoff workshop, doing what the written requirements say, and running at least 75% below the baseline measured in that workshop. Miss any of the three and you get your money back. This is a promise with a refund behind it. It is a commitment about your process, and it reports no result for anybody else.
The baseline is measured in the workshop, before the build. The hours the process takes, at the rate you state, plus the rework the errors cause. That number goes into the contract. Neither side can move it afterwards.
The same method measures both sides. After go-live the process is measured the same way, including what the agent itself costs to run. A saving that exists only because the method changed does not count.
Defects against the requirements are fixed at no cost. Nobody can guarantee software with no defects. What is guaranteed is that anything failing to match the signed requirements is corrected without a further invoice.
What is needed from you. The workshop attended by somebody who can make a decision, access to the systems the agent touches, real cases to test against, and thirty days of live running before the measurement.
How you claim. Email info@peakcodeconsulting.ch within thirty days of the measurement, naming the number that was missed. You get an answer within ten working days.
These terms go into the contract before the build starts, in these words. The 75% is the commitment, and the refund is what happens if it is missed.
Who this is for
This is for you if
- One process eats hours every week, and it is the same process every week.
- The input arrives as text a person has to read: email, orders, forms, documents.
- You can point at the two or three systems the answer has to land in.
- You can supply real cases from the last few months to test against.
This is not for you if
- The process runs differently every time and nobody can describe it in a workshop.
- The input is already structured and always looks the same. A rules-based automation is cheaper and we will say so.
- A wrong decision cannot be undone, such as a payment or a filing with an authority.
- You want the agent without the mapping. The guarantee is settled against the baseline the mapping produces.
The questions we get asked
- What does it cost?
- It is quoted in writing after the first call, against the process you bring. There is no list price, because pricing an agent before seeing the process is guessing. The fee is fixed before the build starts.
- What happens when the agent gets something wrong?
- Every case is logged with the reason it decided what it did. Anything below the agreed confidence line is handed to a person. Testing runs on cases where the right answer is already known, so the wrong answers surface in the build week rather than in front of a customer.
- Where does our data go?
- The model and the hosting region are agreed before the build, Swiss or EU included. The agent reaches only the systems you grant it. Every connection is listed in the documentation you get at handover.
- How is this different from automating the process with a workflow engine?
- A workflow engine runs steps that are already defined and needs the input to look the same every time. An agent copes with text a person wrote. If the workshop shows the process is stable and structured, we say so and build the simpler thing.
- What happens after the week?
- Either the next process, or a care plan that keeps the agent running as your systems change, or nothing at all. The handover is written so that further work stays a choice.
Tell us which process is eating the hours
The first call is free.
Describe the process in a sentence or two. We come back within 48 hours on whether an agent fits it, what the workshop would have to cover, and what the baseline would be measured against.
If a simpler automation would do the same job for less, the reply says so.