Client portal for service firms: build the door, not the archive

Most client portals are a filing cabinet with a password. The client logs in 1 time, looks around, and goes back to mailing documents. Your team retypes them and chases the missing ones. The portal exists, and the work has not changed.

Ask almost any builder for a client portal for a service firm and you get the same list: login, documents, invoices, status. That is a window. The client may look inside. Your team's hours are somewhere else: at the front door, where a file comes in and is almost never complete the first time.

This article is for the owner, managing partner or operations lead of a Dutch B2B service company with 10 to 200 people. You process files by hand: subsidy applications, tax filings, certifications, inspections, policies or annual accounts. Each file needs documents from the client, and they arrive in pieces, by mail, under 4 kinds of file name.

It is not for you if you want a web shop with an order history or a helpdesk with tickets. It is also not for you if you handle 20 files a year. A shared folder and a good checklist will do, and we would rather say so.

In this article you will learn why clients do not log in, which 3 levels a client portal has, what a working front door does step by step, how the numbers come out at 1,500 files a year, and when to buy or build.

‍

Key takeaways

• A portal where the client can only look gets 1 visit. A portal where the client does something that saves them time gets used.

• There are 3 levels: the window (looking), the door (handing in a complete file) and the product (getting a result alone). Start with the door.

• Intake is usually the most expensive step nobody invoices. In our worked example, chasing and retyping costs 1,700 hours a year.

• Let AI check every document at the moment of upload. The client hears within a minute that the year is wrong, not 6 days later by mail.

• AI reads and compares, fixed code applies the rules, a person approves where a mistake is expensive.

• Do not count accounts. Measure the share of files that arrive complete without anyone calling.

• Buying fits when your intake is standard. Building pays when the rules for a complete file are your profession.

• The door is also the first half of a product. What the client hands in today, they can have checked on their own in year 2, for a fee.

‍

Why clients do not log in

A client has no need for a portal. They want their application out the door. Anything that does not visibly help with that is one more password.

A window portal asks something of the client and gives little back. They create an account to see a pdf you could have mailed. The status says "in progress". They knew that.

The pattern is wider than your office. Gartner asked 5,728 customers: 73% try self service at some point, but only 14% of issues are fully resolved there. The rest call or mail anyway. The most common reason: the customer found nothing that matched their question.

Apply that to a file. Your client's question is never "where is my invoice". It is "what do I still owe you, was what I sent correct, and when is it done". A portal that answers those 3 questions without anyone on your side typing gets used. The rest is decoration.

There is 1 more reason, and it sits with your own team. As long as an advisor accepts a mailed document, the client learns that mail works. The portal then loses to habit, every time.

‍

The window, the door and the product: 3 levels of a client portal

It helps to stop treating a client portal as 1 thing. There are 3 levels, and they differ in what the client does and in what you get back.

Level 1: the window

The client looks along. Documents, invoices, a status bar. This often already sits in the package you use, and switching it on is wiser than building it. It saves a few status calls. It changes nothing about how the work arrives.

Level 2: the door

The client hands in the file along a fixed route. The portal knows which documents belong to this type of file, checks each one on arrival and says at once what is missing. Your team receives a complete file in its own system, with no retyping.

This is where the hours are. It is also the level where AI adds something that was not possible 5 years ago: reading a document and understanding what it is before a person opens it.

Level 3: the product

The client gets a result alone. A first check on whether they qualify for a scheme. A calculation. A draft application. Your knowledge sits in the software, and the client pays per use or per month.

This is the level where a service firm becomes a software company. It is also the level that fails if you start there. Without the door you have no clean data, and without clean data the product checks rubbish.

‍

Start with intake, because that is where the time leaks

Ask an advisor where the week goes and they name the substantive work. Look in their mailbox and you see something else. "Could you still send the 2024 annual accounts?" "This scan is unreadable." "This is the quote, I needed the order confirmation."

Each of those mails takes 3 minutes to write. The real cost is the waiting. The file stalls, the advisor starts something else, and 6 days later they have to read back in. With 4 of those rounds per file you are 2 weeks further before the real work begins.

Nobody sees these hours either. They are not on the invoice and not in the planning. They hide inside "file in progress". If you first want to know where the hours really go, start by mapping the process, step by step, with a stopwatch.

Then there is GDPR. Payslips, passports and annual figures now travel back and forth by mail, sometimes to the wrong address. A door with a separate sign in per client is safer, and you can explain it when a client or a regulator asks.

‍

What a working front door does, step by step

A door that works has 6 steps. None of them is spectacular. Together they take the chasing out of the file.

1. A list per file type

The portal knows, per type of file, which documents are needed and under which conditions. A company with staff hands in different documents than a freelancer. The client sees only their own list, with 1 sentence of explanation and an example per document.

That list exists today, in the head of your best advisor. Writing it down is most of the work in the whole project. It is also work you have to do once anyway.

2. Getting in without fuss

The client gets a link by mail and is in. No password to forget. They can take a photo with a phone or drag 9 files at once. Every hurdle here costs you clients who fall back to mail.

3. A check at the moment of upload

This is the step that makes the difference. AI reads the document as soon as it arrives. Is this an annual account or a tax return? Which year? Which company? Is the scan readable? Is it signed?

The client gets the answer while still at the desk. "These are the 2023 annual accounts. We need 2024." They pick the right file and are done. The same mistake by mail costs 6 days.

The AI decides nothing about the file itself here. It reads and compares against the list. Whether a client qualifies for a scheme is decided by fixed code with your rules, and then by a person.

4. Reminders nobody has to remember

If something is still missing after 3 days, the client gets a message with exactly what. After 7 days, another. Only then does the file land on a list for your team. Your advisor calls 1 time, with a reason, instead of mailing 4 times.

5. Complete in your own system

A complete file lands in the system your team already works in. Data sits in the right fields, documents have a fixed name. The link with Exact, AFAS or your sector package is part of the build. Without it you have a nicer form and still someone retyping.

6. A status that says something

The client sees what has arrived, what is missing, who is working on it and what the next date is. Not "in progress". Where a mistake is expensive, a person approves before anything goes to the client or an authority, and the portal says so.

Want to write this out for your own file type before you talk to anyone? Use the free AI product requirements template. It makes you put the documents, the rules and the exceptions on paper. It also gets you quotes you can compare.

‍

Worked example: an advisory firm with 1,500 files a year

This is an example with assumptions, not a client. Take a subsidy advisory firm of 30 people that handles 1,500 files a year. Each file needs 9 documents from the client on average.

Today it goes like this. A file is complete after 4 rounds of chasing on average. Chasing, checking, renaming and retyping take 50 minutes per file together. Add 3 status questions of 6 minutes each. That is 68 minutes per file, and 1,700 hours a year. At an internal rate of 85 euro an hour, that is 144,500 euro of work nobody invoices.

Lead time is the second number. Between the first mail and a complete file sit 16 days on average. Nothing can be invoiced in that time.

With a working door we use these assumptions. 70% of files arrive complete without anyone calling. Those still take 10 minutes of checking. The other 30% need a person and take 35 minutes. Status questions drop to 6 minutes per file.

• Average per file: 23.5 minutes instead of 68.

• Per year: just over 580 hours instead of 1,700.

• Freed: about 1,110 hours, just over 94,000 euro of capacity.

• Lead time to complete: in our assumption from 16 days to 5.

Now the honest part. That 70% is an assumption, and it depends on your clients. At 50%, almost 1,000 hours still come free. And freed hours only become money when you fill them with files. If your team handles the same number of files afterwards, you mostly bought calm. That is worth something, but it is not in the bank.

What does it cost? An automation of this size sits between 18,000 and 45,000 euro with us, fixed per automation. Operations afterwards are 1,250 euro a month, so 15,000 euro a year. At 35,000 euro to build and 70%, it pays back in about half a year. At 45,000 euro and 50%, it takes about 8 months. Both only if the hours get filled.

The AI cost itself is small. Reading a document costs cents. We put a cap on that cost per file, so a client who uploads 300 photos does not eat your margin.

‍

Buy or build

Buy when your intake is standard. Accountants, administration offices and brokers have packages with a client environment built in. If your list of documents looks like that of every other firm in your sector, switch that environment on. Building would waste the money.

Build when the rules for a complete file are your profession. If which document is needed depends on the scheme, the legal form, the year and 3 exceptions only your advisors know, no package will fit. That is also exactly the knowledge your clients pay for.

There is a middle road where many firms start: a form with a rented flow behind it. That works for 1 file type and 50 files. After that you hit the limits we describe in automating workflows with AI: nobody owns it, and you see a mistake only when the client calls.

On market prices: quotes for a portal vary widely, because "portal" can mean anything from a login page to a full platform. So ask for a price per level and per integration. How such a price is built up is in what custom software costs in 2026.

‍

What it asks of your own office

Software is the smaller half here. The rest is behaviour, and it starts with your own people.

• 1 owner. Someone in your firm is responsible for the document list and the rules. When a scheme changes, that person changes the list.

• 1 file type first. Pick the type with the most volume. Not the hardest, not all 6 at once.

• Mail closes. After a transition month, nobody accepts documents by mail. Whoever mails gets the link back, politely. This is the rule the project succeeds or fails on.

• A queue for exceptions. What the AI is not sure about goes to a person, with the reason. That queue has an owner and is emptied every day.

• 1 number. The share of files that arrive complete without chasing. Measure it every month. The number of accounts says nothing.

Expect resistance in month 1. A few long standing clients will call to say it used to be easier. They usually mean: for me. After 2 files they notice they no longer get 4 mails, and it passes.

‍

From front door to product

A door that has run for a year gives you something you did not have before: the same data per file, in the same shape, checked on arrival. That is the material level 3 runs on.

Take the subsidy firm from the example. The client uploads documents for 1 application. With the same documents, software can also check which other schemes they qualify for. We built a subsidy matching engine before that took that search from hours to seconds. Put something like it behind the door, and the client has a reason to come back without you calling.

That is a product. The client pays per check or per month, and your advisors do the part that needs judgment. We call it the Platform, built in stages, from 25,000 euro, usually in year 2. Not earlier. First the door has to show that clients want to walk through it.

‍

Who works on it, what happens after launch, and what if month 1 disappoints

These are the 3 questions to ask any builder. These are our answers.

Who works on it: 2 engineers, by name, from day 1 into operations. You know who writes your code and who gets the alert when something stalls.

What happens after launch: monitoring, a named engineer on the alert, a cap on the AI cost per file, a 1 page report every month and updates when a model or a rule changes. The response time is set in the Blueprint.

What if month 1 disappoints: Exit after month 1: pay for what is delivered, keep everything.

It starts with the Blueprint: 2,500 to 10,000 euro depending on the number of processes, in 10 to 20 working days. You get the process as it runs today, the bottlenecks in hours and euro, a roadmap with fixed prices and a first rough prototype in your own repository. It is yours, also if you build it in house afterwards.

Codelevate turns Dutch B2B service companies into software companies: we map the expert process you run by hand, automate it so the same team handles the volume, and build the platform your customers pay for, with 2 engineers by name from the blueprint to year 2.

‍

How to start this week

You do not need a builder for the first 4 steps. You need an afternoon and 30 finished files.

• Pick 1 file type with the most volume.

• In 30 finished files, count how often someone chased and how many days it took to complete.

• Write down the document list with its conditions. Let your best advisor shoot holes in it.

• Note the 5 most common mistakes: wrong year, wrong document, unreadable, not signed, wrong entity.

• Only then decide: switch it on in your package, or have it built.

For steps 3 and 4 you can use the free AI product requirements template. With a filled in version you know within an hour whether a package is enough.

Send us what you have. 1 of the 2 engineers reads it with you in 20 minutes, and you get 1 of 3 answers in writing the same day: do it in house, hire, or the Blueprint.