BOND: the business AI agent that answers from your company data
BOND connects to the systems your company already runs: CRM, invoicing, spreadsheets, the data warehouse. You ask in a plain sentence and the answer comes back in 5 to 10 seconds, assembled from current data rather than from the model's general knowledge. Every integration has its own permission scope, and every query is written to an audit log.
- 5 to 10 secondshow long someone waits for an answer about company data, asked in a plain sentence
- 4 to 6 weeksfrom workshop to an agent in production, roughly two months from the first call
- 0company data used to train models. Dedicated instance, zero data retention mode, every query logged for audit
The problem
How much time does your team spend looking for data the company already has?
The question how many open deals do we have this quarter sounds like a one-minute job. In practice someone logs into the CRM, sets a filter, exports to a spreadsheet, adds figures from the invoicing system and comes back with a number in the afternoon. The meeting that needed the number finished before lunch.
The data is in the company, just scattered. The CRM knows what stage a deal is at. The invoicing system knows whether the client paid. A spreadsheet knows when the delivery went out. None of them answers a question that spans all three, so a person assembles the answer by hand, from scratch, every time.
Until recently there was no answer to that other than another report, because every connection between a language model and a company system had to be written from zero. In November 2024 Anthropic published Model Context Protocol, an open standard for connecting models to data and tools. BOND is built on it: every system is a separate module with its own permission scope, not an integration written from the ground up.
- The answer about numbers arrives hours later and the decision is taken without it
- The weekly report eats 3 to 5 hours of the one person who can build it
- The same figure differs between two systems and nobody knows which one is true
- Knowing where to look for what sits in the heads of two people in the company
Why now
How is an AI agent different from a report and from a chatbot?
A dashboard answers the question someone anticipated when they built the filter. A question phrased differently needs a new view, so it joins the queue at the analytics desk. An AI agent answers a question asked for the first time, because it decides for itself which systems to look into and how to assemble the answer from them.
A chatbot on a website answers from whatever was pasted into it beforehand. BOND has no answer base of its own. At the moment of the question it reaches into your systems and answers from what it finds there, naming the source. Where the data is missing, or where it has no access, it says so instead of guessing.
The third difference is action. The agent does not stop at an answer: it will add a note to a contact, move a deal to another stage, post a summary to a team channel. You set the write scope before the rollout, separately for each integration, and you can leave everything on read only. If you want a language model inside one or two systems rather than a company-wide agent, start with our AI integration services.
What changes
What does the team get after the rollout?
An answer in 5 to 10 seconds, with nothing to log into
You ask where you already work: in Slack, in Microsoft Teams or in the agent's chat window. You do not need to know which system holds the answer.
5 to 10 seconds from question to answer, in English or in Polish
A number from your data, not from the model's knowledge
BOND answers only from the systems it has access to, and shows where the number came from. Where data is missing it says the data is missing, and that behaviour is set on purpose.
Every answer names its source instead of a number with no origin
A report nobody assembles by hand
The weekly summary, the sales breakdown, the list of clients with no contact for 30 days. The agent pulls it from several systems on a single instruction and sends it wherever you want it.
One instruction instead of five logins and an export to a spreadsheet
Operations in your systems, not just a read-only view
Adding a note to a contact, moving a deal, posting to a team channel. Each integration gets its own permission, read or read and write.
Write scope set per integration, not globally for the agent
Control over who sees what
The permissions of the agent mirror the permissions of the person asking. A salesperson will not see through the agent data they cannot reach in the CRM, and every query stays in the log.
Full query log available for audit, zero data retention mode
The whole team asks about data, not one person
Knowing where to look stops being a bottleneck held by two people. A new hire asks the agent in week one instead of waiting for a colleague to have a free moment.
One question box instead of five systems and five passwords
Proof
Do you use BOND yourselves?
Yes, and it is the only proof we can show without asking a client for permission to publish their figures. Two AI agents running in production inside our founders' own companies, both launched in March 2026 and built by the same method described above. Below is six months of measurement, together with what these 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
An agent 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, today 23 people use it.
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 at a margin of minus 4.42 percent. Across a twenty-month window there was not a single run of three profitable months.
- After
- May, June and July 2026 are three profitable 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 on year, with revenue up 1.5 percent
SurfBottheir agentSurf Inc., a premium clothing brand with an online shop and two physical stores
- 16
- people from the company work with the agent
- 21
- systems wired in permanently
- 89
- processes run without a human
- 323
- thousand lines of the agent’s code
An agent wired into the order system, the ERP, the shop, the data warehouse and the ad platforms. Launched in March 2026, fully running by May. Sixteen people use it, warehouse and store staff included.
8 of 21 systems it works in
PrestaShop- BaseLinker
- Comarch Optima
- BigQuery
- Google Ads
- Meta Ads
- GA4
Zalando
- Before
- From February to August 2025 every unit of ad spend came back as 3.53 in online revenue, and advertising absorbed 28.3 percent of that revenue.
- After
- In the same months of 2026 the ratio is 4.19 and advertising is down to 23.9 percent of revenue. The ad budget is 6.9 percent lower with sales 10.3 percent higher.
Return on ad spend from 3.53 to 4.19, 18.5 percent higher, on a budget 6.9 percent lower
Operations run by the agency agent in successive months of 2026, from the first commit in March. From 83 operations in March to almost 63 thousand in August, with users growing from 4 to 23. This curve is here because a rollout nobody uses looks identical in a report to a successful one.
Measurement: Agent call log in the data warehouse, March to August 2026, counted at the level of a single tool invocation.
And in your company
What if our systems are in a different state?
Four sentences we hear most often in the consultation, and what we answer to each of them.
Our data sits in five systems and in spreadsheets.
That is the normal starting point, and a spreadsheet is a source the agent connects to as well. The workshop exists to set the order: first the two systems that hold the answers to the most frequent questions, then the rest. Connecting everything at once stretches the rollout and does not make the first answer arrive sooner.
We do not want company data leaving the building.
Company data does not reach any external training service. BOND runs on a dedicated instance in zero data retention mode, and every integration has its own permissions, agreed before the rollout. All queries are logged and available for audit, so you can see what people asked and where the answer came from.
We already have a language model the team uses.
That is a good start, because people no longer need convincing about the tool itself. The difference is access to data: a model without integrations answers from whatever you paste in, so a question about order status needs an export first. BOND reaches into the systems itself, at the moment of the question.
The team will not use it.
They will if it removes work nobody wanted to do, and they will not if they read it as aimed at them. That is why the pilot runs with a group of 5 to 10 people from different departments, and usage is measured in the query log. A rollout nobody uses shows up in that log in the second week.
- CRM
- HubSpot, Pipedrive, Salesforce, LiveSpace
- Invoicing and accounting
- Fakturownia, sevDesk, Xero, Comarch Optima
- Sales and stock
- Shopify, WooCommerce, PrestaShop, BaseLinker, Allegro
- Team tools
- Google Workspace, Microsoft 365, Slack, Notion, ClickUp, Jira, Google Drive
- Data and reporting
- BigQuery, Google Sheets, GA4, PostgreSQL and other SQL databases
- Automation layer
- Make, n8n, Zapier, custom API integrations
Questions from the team and the systems BOND answers from
Examples from rollouts we are running. The permission column is the important one: you decide where the agent only reads and where it is allowed to write.
| Question or instruction | Source of the answer | Integration permission |
|---|---|---|
| How many open deals do we have this quarter? | CRM: HubSpot, Pipedrive or Salesforce | read only |
| Has client X paid their invoice? | Fakturownia, Xero or Comarch Optima | read only |
| Which clients have had no contact for 30 days? | CRM plus activity history | read only |
| Which campaigns return the most? | BigQuery, GA4, the cost spreadsheet | read only |
| Where is the contract with company Y and when does it end? | Google Drive, Notion, document archive | read only |
| Add a note to the contact and assign a task | CRM, task system | read and write |
| Post the daily summary to the sales channel | Slack or Microsoft Teams | write in one channel |
The rows are a starting point. A first rollout usually covers two or three systems and a dozen or so of the most frequent questions; the rest goes in when it turns out to be needed.
How we do it
Five steps from a question in Slack to an agent in production
Every step ends with something you can judge before you decide on the next one.
Process workshop
We write down the questions that come up in the company every day and check which systems hold the answers. It ends with a list of use cases ranked by the time they cost today, and a map of permissions.
Integrations and permissions
We connect systems as separate modules: CRM, invoicing, spreadsheets, the data warehouse. For each one we set the scope, read or read and write, and you decide what the agent is allowed to look at.
Company context
The agent learns your vocabulary: product names, department shorthand, metric definitions, what an active client means in your company. Without this step it answers correctly but not in your terms.
Pilot with a group
A group of 5 to 10 people from different departments asks about real work for 2 weeks. We collect the questions the agent answered badly or did not answer at all, and that is the fix list before production.
Production and measurement
Access for the whole team, error monitoring, query log. After 30 and 90 days we look at what people ask most often and what they still search for by hand, and we extend the scope accordingly.
Before and after
How the week changes in the company
Before
- Monday starts by assembling a report from several systems
- A question about a number goes over Slack to whoever has CRM access
- A new hire spends a month asking colleagues where to find what
- A question phrased differently than usual needs a new view in the report
After
- The summary is already waiting and only needs a comment
- Everyone asks for themselves and gets an answer in 5 to 10 seconds, within their own permissions
- A new hire asks the agent in week one and gets an answer with its source
- A question asked for the first time gets an answer straight away, with no queue at the analyst
The first answers on real data show up in the pilot, 2 weeks after the integrations are connected. The full picture after 30 days in production, a second measurement at 90.
For the person signing this off
What the board gets out of it
You stop being the point every question about numbers passes through. Instead of waiting for a report from whoever happens to have system access, you ask yourself and get an answer with its source named. In a board meeting or in front of an investor that is the difference between a number you checked and a number you remembered.
The second thing is project risk. A technology rollout that fails to stick costs the credibility of the person who pushed it through, and closes the subject inside the company for two years. So we spread the risk across stages: the workshop ends with a document, the pilot with a usage count rather than a statement from the team.
- After the workshop you hold a list of questions, systems and permissions that stands up without us
- The pilot shows real usage in the query log before the full rollout is decided
- Permissions per integration and a log for audit, which is the answer for your security team
- Company data is not used to train models, and the instance is dedicated
What it costs
The workshop has a published price, the rollout is quoted after it
The process workshop costs EUR 3,500 net per week and runs from 1 to 7 weeks. For BOND one or two weeks are usually enough, because the work consists of writing down the questions and pointing at the systems that hold the answers.
We quote the BOND rollout after the workshop, once it is clear how many systems are involved and what state they are in. Model usage you pay for directly to the provider, with no margin from us: EUR 0.005 to 0.07 per query at provider list prices, which for typical company use is pennies per question.
Process workshop
EUR 3,500 net
per week, usually one or two for BOND
A self-contained piece of work. It ends with a list of questions, systems and permissions plus a rollout plan you can execute without us.
- The list of questions asked in the company every day, ranked by the time they cost
- A map of the systems and data that hold the answers
- Permission scope per integration, read or read and write
- Rollout order and the metrics to measure at 30 days
- Most popular
Workshop and BOND rollout
Quoted after the workshop
scope follows from the number of systems and their state
The option most companies choose. A working agent 4 to 6 weeks after the workshop, tested in a pilot on your own data.
- Everything in the workshop
- Integrations with the CRM, invoicing, spreadsheets and the data warehouse
- Company context: vocabulary, metric definitions, department names
- Pilot with a group of 5 to 10 people, then production with monitoring
- Team training and a query log for audit
BOND and personal agents
Quoted after the workshop
for companies where the agent works for named people
Everything in the BOND rollout plus the Personal Agent, an AI agent configured for one person and their scope of work.
- Everything in the BOND rollout
- A Personal Agent for selected people, wired to company data through BOND
- Extension to further systems after the 30 and 90 day measurement
- Reacting to changes in the source systems
The workshop ends with a document that stands up at board level without us: the list of questions, the systems, the permission scope and the rollout order. If it turns out your data sits in one system and its own report is enough, we say so, and the matter ends at the workshop rather than at a rollout you do not need.
Frequently asked questions
What is the BOND business AI agent?
It is an AI agent built by JustAutomate and wired into your company systems: CRM, invoicing, spreadsheets, the data warehouse. It answers business questions asked in a plain sentence, in 5 to 10 seconds, using current data it has access to. The whole team asks about data instead of logging into several systems and searching by hand.
How much does a BOND rollout cost?
The published price is the process workshop: EUR 3,500 net per week, usually one or two weeks for BOND. We quote the rollout after the workshop, because the cost depends on the number of systems, their state and the permission scope. Model usage you pay directly to the provider, with no margin from us.
How long does a BOND rollout take?
From 4 to 6 weeks from the workshop to an agent in production, roughly two months from the first call. That covers connecting the integrations, the company context and a pilot with a group of 5 to 10 people. The first answers on your own real data appear during the pilot.
Which systems does BOND integrate with?
CRM (HubSpot, Pipedrive, Salesforce, LiveSpace), accounting (Fakturownia, Xero, sevDesk, Comarch Optima), commerce (Shopify, WooCommerce, PrestaShop, BaseLinker, Allegro), team tools (Google Workspace, Microsoft 365, Slack, Notion, Google Drive) and data (BigQuery, Google Sheets, GA4, PostgreSQL). If a system has an API, it can be connected as a separate module.
Does company data go to external servers?
Company data does not reach any external training service. BOND runs on a dedicated instance in zero data retention mode, and every integration has limited permissions agreed before the rollout. All queries are logged and available for audit, so you can check what was asked and what the answer was built from.
How is BOND different from a chatbot on a website?
A chatbot answers from content uploaded to it beforehand, and usually talks to your customers. BOND answers from your company systems at the moment of the question, and talks to your team. On top of that it performs operations: adding a note, moving a deal, posting a summary to a team channel, within the permissions you grant it.
Can BOND change data in our systems?
Only where you allow it. Integrations go in on read by default, and write is granted one item at a time: a specific object in the CRM, a specific Slack channel. We agree the scope in the workshop and it can be changed after the rollout without rebuilding the agent.
What happens when BOND does not know the answer?
It says the data is missing or that it has no access to it, and names the system that is absent. We set that behaviour on purpose, because an answer you cannot check against a source is useless in a company. Unanswered questions are collected during the pilot and they define the scope of the next integrations.
Let us check which questions BOND would answer in your company
The free consultation takes 30 minutes. You write down the three questions asked most often in your company, and we tell you which systems can answer them and what has to be tidied up along the way. We show a working agent on sample data, with no access to your own systems. You write directly to Pawel, no reception and no tickets.
We run at most two rollouts at a time, so that someone is actually sitting on each of them.
Book a free consultationPS. If answering a question about numbers in your company means logging into three systems and waiting until the evening, that is exactly the problem. BOND answers from the same data in 5 to 10 seconds and names the source, and it starts with a workshop, not with connecting systems.