AI for subsidy advisors in 2027

Every AI demo for subsidy advisors shows the same thing. You type in a company profile, and within 3 seconds a list of 12 matching schemes appears. It looks impressive. It is also the smallest part of your work.

A match does not earn a single euro. The money is in the application that gets approved and in the final report that still holds up 2 years later. That is where your advisors' hours go, and that is where almost every AI tool leaves you on your own.

This article is for owners and managing directors of subsidy advisory firms with 10 to 200 people. You guide companies through WBSO, MIT, DEI+ and investment schemes, often for a fixed fee or a percentage of the grant. Your growth stalls on finding experienced advisors. It is not for you if you file 1 application a year for your own company, if you want a standalone writing tool, or if you are preparing a European tender.

In this article you will learn where AI for subsidy advisors saves real time, which of the 4 stages of a file to buy and which to build, why your previously approved applications are your most valuable asset, a worked example with 320 files a year, and what happens after launch.

‍

Key takeaways

• A subsidy file has 4 stages: finding, intake, writing and reporting. AI helps in all 4, but not in the same way.

• Finding is where every tool starts. In our worked example it is only 7% of the hours.

• The hours are in intake, writing and reporting. Together they are 93% of the work.

• Your approved applications are your real asset. Without that library, an AI writes tidy text that sounds like nobody in particular.

• An AI writes the first draft, fixed rules check the numbers, and an advisor signs off every application.

• In the example, the work drops from about 6,060 to 3,520 hours a year. That is room for around 230 extra files without a new advisor.

• If you work on a success fee, a better final report pays most, because a correction there costs you your own fee.

• Once your process is written down, you have something clients will buy on its own: an ongoing check on their subsidy options and their hours records.

‍

AI for subsidy advisors: where the hours really go

Ask the director of a subsidy firm where the time goes and you rarely hear "searching". You hear: the client's engineer who does not call back for 3 weeks. The project description that goes back and forth 4 times. The hours records that turn out not to add up in January.

Yet almost every AI product in this market starts with finding. From the vendor's side that makes sense. A match is easy to show, easy to sell and easy to measure. A better application is harder to demonstrate, because the result arrives months later.

The market is moving, though. According to preliminary figures from Statistics Netherlands, 27% of Dutch companies with 10 to 50 employees used AI in 2025. So your clients are experimenting too. Some of them now ask a chatbot to write their WBSO text, then call you to fix it.

To get clear on what to do with AI, it helps to cut a file into 4 stages. Each stage has a different answer.

‍

‍

Stage 1: finding, buy the database

Finding is the question of which schemes fit a client. A good advisor knows 80% of that by heart. The rest comes from a subsidy database, the RVO website or the province.

This is exactly the work AI does well. Putting a company profile next to a list of conditions is reading and comparing. We built a subsidy matching engine ourselves that cut hours of expert searching to seconds. That is exactly why we say it plainly: this is the easy part.

There are databases and tools that already do this, and they get better every year. They track schemes, deadlines and budget ceilings, which you do not want to maintain yourself. So buy the database. At most, build a thin filter on top with the rules only your firm uses. For example: no applications under 10,000 euro, or no schemes where you were rejected the last 3 times.

Where finding does pay: with your existing clients. A match that runs your whole client base against new schemes every quarter produces conversations with people who already trust you. That is not a product. That is a sales channel.

‍

Stage 2: intake, the conversation with the engineer

Intake is where a file stalls. You need the technical details: what is being developed, what is new, which technical problems the team expects. That knowledge sits with an engineer or product developer at the client, who has better things to do.

An advisor easily spends 4 hours per file here. An email with questions, a reminder, a call to book a call, a 1 hour conversation, writing it up, 3 more follow up questions. Most of that is waiting and chasing, not expertise.

AI helps here in 2 ways. First, it makes the questions specific. Instead of a standard questionnaire, the system asks questions based on what is already known: last year's application, the website, the project plan the client already shared. An engineer answers 6 targeted questions faster than 30 general ones.

Second, it summarises the answers in the shape your advisor needs. A recording of the intake call becomes a structured project outline, with what is still missing at the top. The advisor reviews that outline in 10 minutes instead of spending an hour writing up notes.

In the worked example below, intake drops from 4 to 2.5 hours per file this way. Not to zero. The conversation with the client stays, because that is where you hear what is not in the documents.

‍

Stage 3: writing, your approved applications are the asset

Writing is the largest item. A WBSO application easily takes an experienced advisor 6 hours. An MIT or DEI+ application with a full project plan and budget quickly runs to 24 hours.

This is where most AI tools go wrong. A general language model writes fluent text about "new technology that will change the market" that reads like every other application. A reviewer who reads many applications sees that at once. What is missing is exactly what makes your firm good: how your advisors argue technical novelty, which phrasing worked before and which led to questions.

That lives in your archive. Every approved application from the last 5 years is an example of what works. Every rejection or follow up question from the RVO is an example of what does not. Together they form a standard that exists nowhere else.

At one subsidy advisory we saw 5 advisors working 5 ways in 5 tools. Everyone did good work, but nobody could take over anyone else's file. The first step there was not AI. The first step was writing down 1 way of working, with the best applications as examples. Only after that could an AI write drafts that sounded like the firm.

It then works like this. The structured project outline from intake goes to the model together with 3 to 5 similar approved applications. The model writes a first draft in the structure the scheme requires. Fixed code then checks what is not language work: do the hours match the budget, does the project fall inside the period, are all required sections there. The advisor sharpens the draft and signs off.

In the worked example, a WBSO application drops from 6 to 3 hours this way, and a large application from 24 to 15 hours. The thinking stays with the advisor. The typing does not.

Want to put this process on paper before you have anything built? Our free AI product requirements template helps you write out the stages, the rules and the cost of a mistake in a form any developer understands.

‍

Stage 4: reporting, where success fees hurt

The stage people notice least costs the most over time. For WBSO, your client has to record per person, per project and per day how many hours went into research and development. After the year ends, the realised hours are reported. That is how the RVO's WBSO page describes it, and it also notes that about 20,000 companies use the scheme each year.

The client keeps those hours records, but the problem comes back to you. Hours that do not match the project description, projects that turned into something else in practice, an engineer booking time on 3 projects at once. At an audit or final assessment, it all has to add up.

If you work on a success fee, this is directly your money. A correction lowers the grant, and with it your fee. And a client who has to pay money back does not forget who wrote the application.

This is where AI is least visible and most useful. A system that puts the client's hours records next to the project description every quarter spots deviations early. Hours on a project that was not in the application. An employee booking more research hours than they work. A project period that has already ended. The AI reads and flags, fixed rules do the arithmetic, and the advisor calls the client before it becomes a problem.

In the worked example, reporting drops from 3 to 2 hours per WBSO file and from 10 to 7 hours per large application. The real gain is not even in those hours. It is in the corrections you prevent.

‍

Worked example: 22 people, 320 files

Take an example firm with 22 people, 14 of them advisors. This is an example with assumptions, not a client. Each year the firm handles 260 WBSO files, 60 larger applications such as MIT, DEI+ and investment deductions, and 200 subsidy scans for new prospects.

This is what the hours look like today:

• Finding: 200 scans of 2 hours is 400 hours.

• Intake: 320 files of 4 hours is 1,280 hours.

• Writing: 260 times 6 hours plus 60 times 24 hours is 3,000 hours.

• Reporting: 260 times 3 hours plus 60 times 10 hours is 1,380 hours.

Together that is about 6,060 hours a year. Finding is 7% of it. That is the part most tools advertise.

With AI in all 4 stages, and an advisor signing off every step:

• Finding: 200 scans of half an hour is 100 hours.

• Intake: 320 files of 2.5 hours is 800 hours.

• Writing: 260 times 3 hours plus 60 times 15 hours is 1,680 hours.

• Reporting: 260 times 2 hours plus 60 times 7 hours is 940 hours.

That is 3,520 hours, a drop of about 2,540 hours or 42%. Look at where those hours come from. Finding saves 300 hours. Writing saves 1,320 hours. The stage that is in every demo is 12% of the gain.

What is that worth? At 1,500 billable hours per advisor, 2,540 hours is about 1.7 advisors. At 80,000 euro fully loaded per advisor, that is around 135,000 euro a year in capacity. At an average of 11 hours per file after the change, it is room for about 230 extra files with the same team, if you can sell them.

There are costs on the other side. Say the Blueprint, the knowledge base of your approved applications and 2 automations (intake with writing, and the hours check) cost 75,000 euro together. Add operations at 1,250 euro a month per automation, so 30,000 euro a year. That leaves about 105,000 euro a year, and the build pays back in about 9 months.

Be strict with yourself. If you only get half the savings, that is about 68,000 euro in capacity minus 30,000 euro operations. Payback then becomes about 2 years. Still worth it, but no miracle. Anyone who promises you a miracle has not seen your archive yet.

‍

What not to automate

Not everything in a subsidy file belongs to an AI. A few limits we hold to:

• Judging whether a project is truly technically new. An AI can gather arguments, the advisor decides.

• Submitting. An application never goes to the RVO or a province without a human signature.

• Bad news to the client. If a project does not qualify, an advisor makes the call.

• Small firms with little volume. Under 100 files a year, a good checklist and a shared folder of example applications pay back faster than software.

That last one matters. Software solves a volume problem. If you do not have a volume problem, you have a knowledge problem, and you solve that on paper first.

‍

Who works on it and what happens after launch

3 questions come up in every conversation, so we answer them here.

Who works on it? With us, 2 engineers, by name, from the first analysis through year 2. They learn your schemes, your archive and your advisors, and they are also the people who get called when something breaks.

What happens after launch? Schemes change every year. Rates, tier limits, forms and conditions shift. In operations, a named engineer watches the alerts, rules and models are updated when a scheme changes, and the AI cost per file stays under a cap. You get a 1 page report every month. The response time is set in the Blueprint.

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

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.

For the order and the prices: the Blueprint costs 2,500 to 10,000 euro and takes 10 to 20 working days. You get your process mapped, the bottlenecks priced in hours and euro, the AI opportunities ranked, a roadmap with fixed prices per milestone and a first rough prototype in your own repository. Automating a major process then costs 18,000 to 45,000 euro, fixed per automation. You can also use the Blueprint to build it yourself.

‍

From service to product

There is one more reason to write your process down. Once the system runs your client base against new schemes every quarter and checks the hours records, you have something clients will buy on its own. An ongoing subsidy check, with an advisor behind it. That is no longer a one off fee but a subscription.

That is a next step, not the first one. We wrote earlier about how to take it in productising your consulting service with AI, with a subsidy firm as the example. If you want to go further, to a full platform where clients log in themselves, the building blocks are in our article on building an AI subsidy platform.

The advice stays the same. Start with the stage that costs you the most hours, not the stage that makes the best demo.

‍

How to start this week

You do not need software to start. You need an afternoon and a spreadsheet.

• Count last year's hours per stage: finding, intake, writing, reporting. Estimates per advisor are good enough.

• Collect your 20 best approved applications in 1 folder, per scheme.

• Write down which questions the RVO asked back most often, and why.

• Pick the stage with the most hours. That is your first automation, not finding.

• Decide who in your firm guards the rules. Without an owner, every knowledge base goes stale within a year.

Want to shape it into something you can discuss with a developer? Download the free AI product requirements template and fill it in for 1 scheme.

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. Book your conversation here.