AI agent for startups once the process starts repeating
If you are still hunting for product market fit and the process changes every two weeks, do not automate. You would freeze something that has to be taken apart a month later. This page is for a startup past its first customers, where support has started to repeat and the founder is still the bottleneck: onboarding, tickets and the numbers for investors all run through one head.
- 14 daysto the first working automation, counted from the end of the workshop
- EUR 3,500 netper week of workshop, a published price you know before the call
- 5 to 10 secondshow long an AI agent takes to answer from your knowledge base, instead of a question to the founder on Slack
The problem
Why is the founder the bottleneck in their own company?
Start with the case where you should stop reading. A startup still hunting for product market fit changes its process every two weeks. Automation freezes something that is not settled yet, and a month later it has to be pulled apart and written again. Come back when customer support starts to repeat. We say the same thing on the free consultation, before we issue any invoice.
The rest of this page is about the moment that comes later. The customers are there, support looks much the same every time, the team has grown to a dozen or more people, and the company grows faster than it hires. That is when you notice that onboarding does run step by step, except the steps live in the founder's head. Every question about the process lands in Slack with the same person, and a new customer waits for access because that person is in a meeting.
Then there are the numbers that disagree between systems. Sales sit in the CRM, payments in the payment provider, product usage in the database, costs in a spreadsheet. The annual accounts have a statutory deadline and a named owner, so nobody misses that one. Sales, activation and churn have neither a deadline nor an owner, so they get produced the evening before a meeting, from memory.
- Onboarding a new customer stalls while the founder is travelling
- The same figure comes out differently from the CRM and from the payment system
- The investor update is assembled by hand before every meeting
- The tooling bill grows every month and nobody checks which tools you use
Why now
How do you tell that a process is ready to automate?
By repetition, not by headcount. We look at how many times in the last 30 days somebody ran the same sequence, and how many of those times it ran the same way. A process that happened a dozen or more times in a month with the same steps every time is ready. A process that looks different for every customer is not, and we say so plainly.
The second change is in the tools. Google Workspace and Microsoft 365, Slack, Notion, HubSpot and Pipedrive, Stripe and the payment systems all ship documented interfaces today, so data can be read and written without an export to a spreadsheet. We put the logic layer in n8n or Make, and collect the data in Google Sheets or in BigQuery, depending on how much of it there is. Wiring that together is counted in working days.
The third is the AI agent. A rule will move data from one system to another, but it will not tell a new hire how a customer on an annual plan cancels. BOND, our company AI agent, answers from your knowledge base in 5 to 10 seconds and points at the document it used. Asking the founder on Slack stops being the fastest route to an answer.
What changes
What stops going through the founder?
Customer onboarding runs on its own
A signed contract or a stage change in the CRM fires the whole sequence: the account, the access, the welcome email, a task for the account owner, a meeting in the calendar. Nobody has to remind anybody.
Steps fired from one CRM event, the same order for every customer
One set of numbers, every morning
Sales, activation and churn in one place, counted from the same definitions as a month ago. The conversation starts with a conclusion instead of a question about where the figure came from.
One source instead of four dashboards, ready before the day starts
The team asks the agent, not the founder
Questions about the process, the price list, contract terms and the odd edge case have an answer in your knowledge base. The AI agent gives it along with the document it came from, so it can be checked.
An answer in 5 to 10 seconds, at 22:00 the same as at midday
Tickets in a queue, not in direct messages
A request from email, chat, the form or a direct message to the founder lands in one place, with a priority set on arrival. You can see how many are waiting and with whom.
Every ticket has a number and an owner, including the one sent to the founder
The investor update is built from data
The numbers come out of the CRM, the payment system and the product database, using definitions agreed once. Preparing for the meeting stops being an evening of work the day before.
The same definition of every metric every month, so the comparison means something
You can see what you pay for
Subscriptions lined up against the number of active seats and actual usage. The tool you have paid for since spring that nobody logs into comes to the surface.
A list of subscriptions with active seat counts, refreshed monthly
Proof
Do you use this yourselves?
Yes, and it is the only proof we can show without asking a client for permission to publish their numbers. This is not a startup and we are not pretending it is: it is a services company of around 30 people where the AI agent went live in March 2026, built with the same method described above. We show it because the size of the team and the way it works are the closest thing we have to a startup past its first customers. Below is half a year of measurement, together with what those numbers do NOT prove.
Built for ourselves. These are not client deployments but implementations inside the founders' own companies. We say so plainly, because proof you cannot verify is not proof.
JustBottheir agentJustIdea Agency, a performance marketing agency of around 30 people
- 23
- people from the company work with the agent
- 38
- systems wired in permanently
- 112
- processes run without a human
- 475
- thousand lines of the agent’s code
The agent is wired into Google Ads, Meta, the CRM, the task system, the SEO tool, Analytics, Search Console and the data warehouse. First commit 1 March 2026, 23 people use it today.
8 of 38 systems it works in
Google Ads
Meta Ads- HubSpot
- ActiveCollab
- Ahrefs
GA4
Search Console
BigQuery
- Before
- In the twelve months before launch the agency ran on a margin of minus 4.42 percent. Across a twenty month window there was not a single run of three positive months.
- After
- May, June and July 2026 were three positive months in a row, the first such run in the whole window. Revenue barely moved, so the change sits on the cost of delivery, not on sales.
Margin up 5.55 percentage points year over year, on revenue up 1.5 percent
Operations run by the agency agent in consecutive months of 2026, from the first commit in March. From 83 operations in March to almost 63,000 in August, with users growing from 4 to 23. This curve is here because a rollout nobody uses looks exactly like a successful one in a report.
Measurement: Agent invocation log in the data warehouse, March to August 2026, counted at the level of a single tool call.
And in your company
What if we are at an earlier stage?
Four sentences come up in startups more often than any others. The first one ends in a no from us, and we think that is how it should be.
We are before product market fit, but we want to get ready.
Then we say: not yet. A process that changes every two weeks has to be dismantled after you automate it, so you would pay twice. The one thing worth doing now is recording events in one place, so that in six months there is something to count from. That is advice, not a project, and we do not charge for it.
There are five of us and everyone does everything.
Headcount does not decide it. Repetition does: whether the same sequence happened a dozen or more times in the last month, and whether it looked the same every time. If onboarding looks different for every customer, we automate the common part or nothing at all.
We do not want to be tied to a vendor for years.
The workflows, the integrations and the documentation are yours and they sit on your accounts in n8n, Make or the cloud you nominate. You can change them yourselves, hand them to somebody else or switch them off, without asking us and without our access.
We have no budget for another tool in the stack.
The workshop costs EUR 3,500 net per week and you know that price before the call. You pay for language models and tools directly to the vendor with no commission from us, and on some rollouts they fit inside plans you already hold. If the workshop concludes that tidying up a spreadsheet is enough, we will say so.
- Teamwork
- Google Workspace, Microsoft 365, Slack, Notion
- Sales and customers
- HubSpot, Pipedrive, the form on your site
- Payments and invoicing
- Stripe, payment systems, Fakturownia
- Tickets and support
- Shared inbox, site chat, forms, Slack
- Data and reporting
- BigQuery, Google Sheets, GA4, SQL databases
- Automation layer
- n8n, Make, custom API integrations
When not to automate yet, and when it is worth it
The thresholds we use to tell that a process has settled. They are not an industry standard, just what we see in companies of this size, and we check them against your own last 30 days.
| Process | Not yet | Worth it now |
|---|---|---|
| Customer onboarding | under 5 a month, each one different | a dozen or more a month, the same steps |
| Key numbers in one place | metric definitions change every month | definitions settled, data in at least two systems |
| AI agent on a knowledge base | the knowledge fits in 3 heads | a team of a dozen or more, docs in Notion or on a drive |
| Ticket queue | a few tickets a week, all to the founder | a dozen or more a week, at least 2 people answering |
| Investor update | once a quarter, numbers from one spreadsheet | monthly, numbers from three systems at once |
| Invoicing and billing | a dozen or so invoices a month, one currency | subscriptions, renewals and cancellations on a fixed cycle |
| Customer email sequences | a list under 200 addresses, sent by hand | segments and triggers from in-product behaviour |
A row on the left of the threshold is not a failure, it is a stage. In the workshop we mark which processes are already on the right, and take the three with the highest return.
How we work
Five steps from the founder's head to a process in a system
Every step ends with something you can judge before you decide on the next one. The first one can end with a recommendation not to go further.
Workshop
One week. We walk the customer journey from signature to first cancellation, and the journey of a question from the team to an answer. We come out with two lists: the processes that already repeat, and the ones we leave alone because they still change.
Picking three automations
From the first list we take the three with the highest return and measure them in founder hours and team hours, not in percentages. We record the state before the start, so there is something to compare after thirty days.
Building the first one
Fourteen days to a working solution. In startups we usually begin with customer onboarding or with one set of numbers, because those take work off a single person fastest.
Handover to the team
The accounts, the workflows and the documentation stay on your side. Someone on the operations side learns to change the steps: a new customer type, a different welcome email, an extra condition in the ticket queue.
Measurement and the next one
After thirty and ninety days we check the indicators agreed in the workshop. Only then do we decide whether we take another process, or whether you run this yourselves from here.
Before and after
How a founder's week changes
Before
- A new customer waits for access because the founder is in a meeting
- A question about the process goes to one person on Slack, on Sunday too
- The investor numbers get made the evening before the meeting
- A new hire learns the process from whoever happens to have a free slot
After
- Onboarding starts the same day, without waiting for one person
- A question about the process is answered from the knowledge base in 5 to 10 seconds
- The numbers are ready every morning and the meeting opens with a conclusion
- A new hire gets an answer with the source document, without interrupting anyone
What does not change is how much the founder knows about the company. What changes is how much of that knowledge they have to say out loud in person before the team can move.
For the founder and the operations lead
What the person ordering this gets
The first thing shows up in the calendar. The hours that go today on repeating the same explanation and clicking through onboarding come back to the product and to customer conversations. That is the one resource you cannot buy faster than you can hire people.
The second is quieter, but it speaks up whenever a question arrives from outside. Sales, activation and churn counted from data, on a definition written down once, stop depending on who assembled the figures and at what hour. Asked where a number comes from, you answer with a method instead of from memory.
- Process knowledge stops leaving the company with one person
- Accounts, workflows and documentation sit on your side, with the right to change them without us
- The first rollout covers three processes, not the whole company at once
- If there is nothing to automate, you hear it in the workshop and not after a rollout
What it costs
The workshop price is published, the rollout price comes after it
The workshop costs EUR 3,500 net per week and we publish that before any call, because a company counting every outgoing has a right to know what it is signing up for. In a startup one week is usually enough. We quote the rollout after the workshop, once we know how many systems we connect to and what state they are in.
You pay for language models and tools, meaning n8n, Make or the cloud, directly to the vendor. Model cost sits in the range of EUR 0.005 to 0.07 per query depending on the model, and when everything runs on your own infrastructure you do not pay for it at all. No commission from us in any of those cases.
Process workshop
EUR 3,500 net
per week, in a startup usually one
A self-contained piece of work. It ends with a process map and a 30 day plan your team executes without us.
- A week of work inside your processes and your tools
- Processes split into the ones that already repeat and the ones still changing
- Three automations with the highest return, measured in hours
- The indicators to measure after 30 and 90 days
- Most popular
Workshop and first rollout
Quoted after the workshop
scope follows from the number of systems and the state of the data
For a startup where support already repeats and the founder still stands in the middle of every process.
- Everything in the workshop
- The first automation built within 14 days
- Connections to the CRM, the payment system, Slack and the product database
- Handover of accounts, workflows and documentation to your team
- Measurement after 30 and 90 days
AI agent for the team
Quoted after the workshop
for teams where questions to the founder come up daily
Everything in the rollout plus BOND, the company AI agent answering from your knowledge base in 5 to 10 seconds.
- Everything in the rollout
- An agent answering from your documentation and data, with the source shown
- A daily view of sales, activation and churn
- Alerts on unanswered tickets and on numbers drifting out of range
- Access for the whole team, in English and in Polish
If the workshop shows the process has not settled yet, we say plainly that you should not automate it and we do not propose a rollout. The workshop still ends with a document you can defend without us: a process map, a list of thresholds to re-check in a quarter, and a 30 day plan your own team can execute.
Frequently asked questions
When should a startup NOT automate?
When the process changes every two weeks because the product is still looking for its market. Automation then freezes steps that will be different in a month, and it has to be dismantled. We test this with a simple question: how many times in the last 30 days did the same sequence repeat without changes. If the answer is under a dozen or so, we tell you to wait.
What is an AI agent for a startup?
An agent that reads your documentation, your ticket history and the data in your CRM and payment system, then answers in a plain sentence in 5 to 10 seconds and shows the source it used. Ours is called BOND and a rollout takes around two months. It removes the need to ask the founder about things that are already written down somewhere.
How long does the first rollout take?
One week of workshop and fourteen days to the first working automation, counted from the end of it. We measure the effect after thirty and ninety days, on indicators agreed before the start. We run at most two rollouts at a time, so the start date can be several weeks out.
We have no IT team, who maintains this afterwards?
The person who knows the process, not a programmer. We take the technical part, and after handover changing a step in onboarding or a condition in the ticket queue looks like editing a form. The accounts and the documentation sit on your side, so you can also hand the whole thing to somebody else.
Do we have to change the tools we already use?
No. We read the data through interfaces and leave Slack, Notion, HubSpot and Pipedrive where they are. Swapping a tool in a startup costs more than the rollout itself, because it takes the team weeks to learn. If something is genuinely missing we raise it in the workshop and show what it would change.
How do you check whether a process already repeats?
We count occurrences in the last 30 days and compare the steps between them. What interests us is how many times the process happened, how many times it ran identically, and how many times somebody had to ask the founder to move it forward. Three numbers are enough to settle it, and they need no access to your systems.
What do tools and language models cost outside your invoice?
You pay for them directly to the vendor with no commission from us, and on some rollouts they fit inside plans you already pay for. Model cost usually runs between EUR 0.005 and 0.07 per query. When everything sits on your own infrastructure there is no extra cost, and in the workshop we give a range for the chosen option so the spend is no surprise.
Do you help with the investor update?
On the technical side yes: we collect sales, activation and churn from the CRM, the payment system and the product database, with one definition of each metric written down once. That is what makes consecutive months comparable. We do not advise on what to put in the update, and we have nothing to say about how it will be received.
Let us check whether there is anything to automate yet
The free consultation takes 30 minutes. We walk through one of your processes, usually customer onboarding, and tell you whether it repeats enough to hand to a system. If it is too early, you hear that straight away and we talk again in a quarter. You write directly to Paweł, with no reception desk and no ticket.
We run at most two rollouts at a time, so that someone is actually sitting on each of them.
Book a free consultationPS. If somebody on your team waited this week for the founder to reply before they could move on a customer, that is exactly the cost. A process that already repeats can be handed to a system in fourteen days. A process that still changes is better left alone, and we will tell you that too.