AI agent for financial services the decision stays with your adviser
A financial adviser or broker lives off two things: a complete set of client documents and the time left before a decision. The automation takes over what eats the hours today: chasing what is missing, reading data out of documents, retyping it into the next application and collecting case status from lender portals. Assessing the client and choosing the product stay with the adviser.
- 5 portalsthat is how many places a broker working with several lenders opens every morning to see what moved. We reduce it to one list of cases
- 14 daysfrom the end of the workshop to the first automation running, tested on your own cases from a month that is already closed
- 0 decisionsthat is how many decisions about a client the agent makes. It does not assess creditworthiness and it does not recommend a product, that stays with the adviser
The problem
Why does an application go back to the client for the third time?
A case starts with a conversation and then spends two weeks as an exercise in collecting files. The client sends half of what you asked for, one document is six months old, another is missing because nobody told him this particular lender wants it. You write a second time, a third time, and the clock to the decision runs on your side, not on his.
The second half of the day goes on retyping. The same client details are entered into an application in one system, a form in another and a spreadsheet you keep yourself, because none of them shows the whole case. Then you still have to check what moved: log in to the first lender's portal, the second, the third. When the client rings to ask about status, the answer takes a few minutes when it should take a second.
The third thing is a line you have to draw before you switch anything on. Article 22 of the GDPR, regulation 2016/679, gives a person the right not to be subject to a decision based solely on automated processing where that decision produces legal effects concerning him or similarly significantly affects him. A refusal of finance is exactly that kind of decision. On top of that, financial services are a supervised market, so every step of the process has to be reconstructable after the fact. That is why the automation collects, checks and reminds, and a person decides.
- An application comes back because one document was missing and nobody checked before it went out
- The same client details typed in again, into the third form that day
- Case status checked by logging in to one more lender portal
- A client asks about terms that sit in your product table, and nobody remembers where
- A renewal date passes because the reminder lived in one person's head
Why now
What changed on the document and data side?
The first change is the reading. A model reads a scan of an earnings certificate, a bank statement or a contract well enough that the work stops being retyping and starts being review. The adviser sees the extracted fields next to the document and corrects whatever the model got wrong. The same extraction then feeds the next forms, so the same data is entered once.
The second change is in watching the case. The document requirements of every lender can be written into one table and compared with the state of a case automatically. Then the list of gaps goes out to the client on its own, after every change of status, rather than when somebody happens to remember. Not every lender exposes an interface, so we build the single status view out of what sits on your side: confirmation emails, export files, spreadsheets and entries made by the adviser. We do not log in to somebody else's portal with a script, and we say so at the start.
The third thing is an AI agent for the team. BOND, our company AI agent, answers advisers from your own procedures and product tables in 5 to 10 seconds, instead of sending them to a ring binder and to the colleague who happens to remember. The conversation with the client and the product recommendation stay human work, and the agent does not go near them.
What changes
What does an adviser get after the first month?
The list of gaps goes out on its own, not in the third round
After every change of case status the system compares the set of documents this lender requires with what the client has already sent, and mails him the missing items. The adviser steps in only when the client goes quiet.
One list of gaps instead of three rounds of chasing by email
Document data read, not retyped
Numbers, dates, amounts and company details come out of the file with no keyboard, including from a photo taken on a phone. The adviser reviews the extraction next to the scan and fixes whatever the model could not read.
One extraction feeds the next forms instead of the same details typed three times
Completeness checked before the application is sent
A rule compares the case with the lender's requirement list and holds it back when something is missing or the numbers do not agree. The application moves on only after the adviser gives it a green light.
Gaps surface before submission, not after a lender rejects the file
Every case status in one view
One table shows each case, its stage, the date of the last movement and what has been standing still the longest. Instead of a morning round of portals there is one list, the same one for the whole team.
1 view instead of 5 portals, refreshed every day
An AI agent answers advisers from your own material
Questions about product terms, required documents and what the procedure says get an answer in seconds, out of your tables and your documents. The agent issues no recommendation to a client.
An answer in 5 to 10 seconds, from your own base, not from the model's general knowledge
Renewals and instalment dates watched without a calendar in someone's head
An approaching renewal, an instalment and the end of a fixed rate period land on the adviser's list with the lead time you set yourselves. The client gets the reminder by whichever route you chose.
Lead time set separately for each type of product
Proof
Do you work this way yourselves?
Yes, and to be straight about it: our measurement does not come from a financial advisory firm, it comes from our founders' agency. That is a services business running dozens of client cases in parallel, where repetitive delivery work ate hours the same way it does in your office. The AI agent has been running in production there since March 2026. Below is the number, together with what it does 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
And in your firm
What if it does not work for us?
Four sentences we hear most often from advisory firms and brokers, and what we say back to them. The rules we follow with personal data are written out in the guide AI and GDPR.
We are not letting a model assess our client.
Neither are we, and that is how we build it. The automation collects documents, reads the data out of them, checks that the case is complete and watches the dates. Creditworthiness, product choice and the decision on the application stay with the adviser and with the lender. Article 22 of the GDPR speaks plainly about a decision based solely on automated processing that produces legal effects, and that is the line we do not cross.
Our data is income, liabilities and a client's history.
We agree the scope in the workshop and write it down before any rollout: the model sees specific sources and specific fields needed for the task, not the whole case database. Processing runs at a provider you hold a data processing agreement with, in the region you name. Client consents and how long you keep documents after a case closes go into the workshop too, because without that the automation only repeats today's mess faster.
The lenders we work with have no interface at all.
Some of them do not, and we are not pretending there is a way around it. We do not log in to somebody else's portal with a script. We build the single view out of what sits on your side: confirmation emails, export files, the spreadsheets you keep anyway and adviser entries made in one click. Even a view like that pulls to the surface the case that has been standing for two weeks, which today nobody sees.
Clients come to us for a conversation, not for an automation.
Good, because that is not what we take away. The automation lifts off collecting files, retyping data and watching dates, which is work the client does not pay for. The conversation, the product choice and explaining the terms stay where they were, you simply have more hours in the week for them.
- CRM
- HubSpot, Pipedrive, Salesforce, Microsoft Dynamics
- Client documents
- Google Workspace, Microsoft 365, a shared inbox, drives, a photo from a phone
- Lenders and providers
- confirmation emails, export files from portals, status spreadsheets, adviser entries
- Commissions and invoicing
- Xero, QuickBooks, Fakturownia, Comarch Optima through a connector inside your network
- Calendar and notifications
- Google Calendar, Outlook, Slack, Microsoft Teams, email
- Data and the automation layer
- Google Sheets, BigQuery, SQL databases, n8n, Make, custom API integrations
What the automation takes over in a case, and what stays with a person
The split we open every conversation in this industry with. The right hand column matters more than the left, because it decides where responsibility sits when a case goes wrong.
| Stage of the case | What the automation takes over | What stays with a person |
|---|---|---|
| Collecting documents from the client | the list of gaps after every change of status, and reminders on a fixed rhythm | deciding what this particular case and this particular lender require |
| Reading data out of documents | numbers, dates, amounts and company details pulled from the file with nothing retyped | approving the extraction and correcting it where the document is unreadable |
| Completeness check before submission | comparing the case with the requirement list and flagging the gaps | the decision to send the application |
| Filling in the application | carrying the same data into the next forms | signatures, declarations and contact with the lender's case handler |
| Watching the statuses | one view of cases with the stage, the date of the last movement and what has stood longest | the call to the lender when a case is stuck |
| Assessing the client and choosing the product | nothing | the adviser, every time |
| Renewals, instalments and dates | a reminder to the adviser and to the client ahead of the date | what is said in the call and the client's decision |
The rows are a starting point and we sharpen them in the workshop, case by case. The row about assessing a client is not up for negotiation and stays empty on the automation side.
How we work
Five steps from three rounds of chasing to one list
Process first, tools second. Every step ends with something you can judge before you decide on the next one. The first three do not send a single message to a real client.
Process workshop
One week. We take two or three cases with different profiles and walk them from the client's first question to the lender's decision. We note how many times you go back for a document, how many times the same details get typed again and where the case most often stands still. The output is a document, not code.
One place for documents and one requirement list per lender
We collect the document requirements of every lender into one table and build one intake for client files, whether they arrived by email or as a photo from a phone. We run it on cases that are already closed, so nothing leaves the office.
The list of gaps and the completeness check
We add the comparison of a case with the requirement list, plus the message to the client listing what is missing. The adviser judges its wording and its timing on closed cases, before anything goes to a live client.
Document extraction and one status view
We switch on field extraction from documents and the case table with the stage and the date of the last movement. The adviser's corrections give us the threshold: above it an extraction moves on with one click, below it goes to manual handling.
Production and measurement
We switch over with error monitoring and a queue where a held case sits with its reason, ready to retry. At 30 and at 90 days we look at the indicators agreed before the start: the share of applications sent with nothing missing, the number of chasing rounds and the time from the first enquiry to a complete set of documents.
Before and after
How the week changes for an adviser
Before
- A morning round of lender portals to see what moved
- A third email to the client about the same missing document
- The same client details typed into one more form
- The client reminds you about the coming renewal, not the other way round
After
- One morning list of cases with the stage and the date of the last movement
- The list of gaps went out on its own, the adviser steps in when the client goes quiet
- The extraction from the document feeds the next forms, the data is entered once
- A renewal date is visible with the lead time you set yourselves
The first effects show 14 days after the workshop, on one case path and one lender. The share of applications sent with nothing missing and the time from enquiry to a complete set of documents are judged at 30 and at 90 days, because only by then has a full cycle gone through the system.
For the person signing this off
What the owner of an advisory firm gets out of it
The first thing is capacity. Today the number of cases you can run in parallel follows from how many hours the team can survive collecting documents and retyping data. Once the automation takes that part over, a conversation about another lender partner or another adviser stops opening with the question of who is physically going to handle it.
The second thing is about risk and about people. Knowledge of what each lender requires and where each case stands stops sitting in one person's head, so a holiday or a resignation no longer stops a client. The trail of what was sent and when stays in the system, and responsibility for assessing a client stays where it was, with the adviser.
- Collecting documents stops setting the limit on how many cases you can run
- Creditworthiness and the product recommendation stay with a person, every time
- The history of a case can be reconstructed later, without digging through four inboxes
- The first three steps of the rollout send nothing to a live client
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. In advisory and brokerage one or two weeks is usually enough, because you have to walk two or three different cases rather than one. We quote the rollout after the workshop, once it is clear how many lenders you work with, how many of them can be read automatically and what state your CRM is in.
CRM licences, the running cost of scenarios in n8n or Make and the cost of model usage go straight to those providers at their own list prices, and a single query costs the provider roughly EUR 0.005 to 0.07. We add no margin and we do not handle those payments. There is more about the workshop itself on the process workshop page.
Process workshop
EUR 3,500 net
per week, and in advisory work usually one to two weeks
A self-contained piece of work. It ends with a map of how a case travels and a 30 day plan you could carry out without us.
- Walking two or three cases with different profiles, from enquiry to decision
- The list of required documents per lender, written into one table
- A map of the points where a case stalls and goes back to the client
- The line between the automation and a person, written down before anything is switched on
- Indicators to measure at 30 and at 90 days
- Most popular
Workshop and case handling rollout
Quoted after the workshop
scope follows from the number of lenders and from the state of your CRM
The option most firms choose. One place for documents, the list of gaps and the completeness check, first run 14 days after the workshop.
- Everything in the workshop
- One intake for client documents, whatever route they arrived by
- The list of gaps sent after every change of case status
- Field extraction from documents and a completeness check before submission
- One case status view with the date of the last movement
- Training for the person who runs this on your side
Rollout and an AI agent for advisers
Quoted after the workshop
for teams where the same questions about terms come back every day
Everything in the case handling rollout plus BOND, the company AI agent that answers from your procedures and your product tables. Agent rollout takes around 2 months.
- Everything in the case handling rollout
- An AI agent answering the team in 5 to 10 seconds, from your own material
- Reminders about renewals, instalments and contract dates
- A case overview for the owner, refreshed every day
- Access for the team, with the data scope agreed and written down before the start
The workshop ends with a document that stands up without us: a map of how a case travels, the requirement lists per lender and a 30 day plan. If it turns out that one place for documents and a decent template for the gaps email is enough, we say so plainly and show you how to set it up. The matter ends at the workshop then, rather than at a rollout you do not need.
Frequently asked questions
Will the AI agent assess a client's creditworthiness?
No, and that is not how we build it. The agent collects documents, reads the data out of them, checks that the case is complete and watches the dates, which is the work done on a keyboard today. Assessing the client and the decision on the application stay with the adviser and with the lender. Article 22 of the GDPR gives a person the right not to be subject to a decision based solely on automated processing where it produces legal effects.
Will the agent tell a client which product to take?
No. The agent answers your team, not your client, and it works from your procedures and your product tables so that an adviser has the terms to hand in 5 to 10 seconds. The recommendation to a client is the adviser's act and it stays there, together with the responsibility for it and the paperwork it leaves behind.
What happens to client data, given it is income and liabilities?
We agree the scope in the workshop and write it down before any rollout: the model gets specific sources and specific fields needed for the task, not access to the whole case database. Processing runs at a provider you hold a data processing agreement with, in the region you name. Part of the flow can stay on your own infrastructure if you prefer it that way.
How will you collect status from lender portals that have no interface?
We do not walk around portals and we do not log in to them for you with a script. We build the single view out of what can be read reliably: confirmation emails, export files, the spreadsheets you keep anyway and adviser entries made in one click. A case with no such source still shows in the view, with the date of the last movement and a flag that it has been waiting the longest.
How long does a rollout in a financial advisory firm take?
One to two weeks of workshop, then 2 to 4 weeks of rollout, with the first working automation usually standing 14 days after the workshop. We start on cases that are already closed, so nothing goes out to a client or to a lender during the early steps. The AI agent for the team comes later and its rollout takes around 2 months.
Can this be wired into our invoicing and commission software?
Yes, but the route differs by kind of software. Cloud tools such as Xero, QuickBooks or Fakturownia expose an API over the internet, so a connector in your network is not needed. Programs installed locally, Comarch Optima among them, keep their data in a database on your own server with no public address, so a connector running inside your network does the work and calls out for tasks itself. The first pass runs on a copy of the data.
Do we have to change our CRM?
No. We leave HubSpot, Pipedrive or Salesforce where they are and add a layer that watches documents, completeness and statuses. Migrating a CRM is a separate project measured in quarters, and the first automation is meant to exist 14 days after the workshop, so we keep the two apart.
We are a team of three, is it too early for us?
It may well be too early, and we will say so on the consultation rather than sell you a rollout. With a dozen or so cases a month the biggest value usually comes from the list of gaps alone and from the reminders about dates, which is a narrow scope for narrow money. Full document extraction and an AI agent pay back once collecting paperwork starts setting the limit on how many cases you can run.
Let us find where the time goes in your cases
The free consultation takes 30 minutes. You tell us how many cases you run in parallel, how many lenders you work with and in how many places you check status today. We tell you what can be taken over straight away, what needs one place for documents first, and what we do not touch at all because it stays with the adviser. You do not have to give us access to any system at this stage.
We run at most two rollouts at a time, so that someone is genuinely sitting on each of them.
Book a free consultationPS. If somebody in your office wrote to a client for the third time this week about the same document, that is exactly this problem. The list of gaps can go out on its own after every change of status, and the adviser can step in at the conversation. Assessing the client stays with him anyway. You write directly to Pawel, the founder, not to a shared inbox.