AI in brand licensing: what 60+ tests on our demo org taught us

You can ask a chatbot almost anything. Ask it which approved products never sold, or which licensee is behind on statements, and it has nothing to go on. Those answers live in your agreements, approvals and royalty reports.

That gap is where AI in brand licensing gets interesting. We have now run more than sixty AI tests on our Flowhaven demo org. Flowhaven is built on Salesforce, so AI tools such as Claude can connect to it through an integration.

Everything below was tested on demo data with fictional brands, or is still on our test list. None of it is a shipped feature inside Flowhaven.

The four shifts of AI in brand licensing

After all those tests I keep landing on the same picture. AI in licensing is not one change. It’s four, and most of us are still on the first one.

1. From finding to asking. Getting an answer used to mean navigating to a record or building a dashboard. Connected to the system of record, it’s a sentence.
2. From asking to being told. The system knows what needs whose attention and brings it to them in Slack, email or by voice.
3. From doing to approving. The work gets redesigned step by step. AI proposes, a person confirms.
4. From running to growing. When the routine runs itself, your team’s time moves to market research, new categories and new partners.

Most AI talk in our industry is still about shift one. Impressive the first time, and close to solved. The real work is in shifts two, three and four.

Finding and asking: questions nobody could answer before

One of my favourite questions came from our Customer Success team, not from me. “Which products were approved in the last twelve months but never appeared on a royalty report?”

Almost impossible to answer in most licensing setups, because approvals and royalties live in different systems owned by different people. With both in one record, Claude answered it on our fictional demo data.

Then it volunteered a contrast nobody asked for. Some partners turned approvals into sales at a healthy rate. One had heavy design activity and almost nothing sold, often an early signal a year before the renewal talk.

I also asked the demo org for our top 5 licensees in Finnish. The org is in English. Claude answered in Finnish without being told to, and got the vocabulary right: minimitakuu for minimum guarantee, rojaltit, nettomyynti.

Your licensee in Madrid, your agent in Tokyo and your finance lead in Helsinki don’t have to speak the system’s language.

None of this works without the record. Anyone can buy the model. Very few have the record.

Because the answers come from your agreements, approvals and royalties, you can check every one of them.

Approvals: treat the queue as a diagnosis

We asked Claude to read every product approval comment between a licensor and its licensees, and score the tone on both sides. The licensee side was impatient. Fair enough, they had deadlines.

The licensor side scored lower. Plenty of “please change the base colour” and a few “no good”. No context, no reason, no next step.

Every comment like that costs a revision round, and a little bit of the relationship.

Next we asked Claude to analyse why products were rejected. Without being asked, it proposed fixes: brief licensees better, explain expectations up front, remind internal teams of the rules that trip people up.

If most of your rejections come from the same logo rule, you don’t have an approvals problem. You have a style guide problem.

We’re also testing what faster approvals are worth in money. Pulled forward sales simply arrive earlier, which helps cash flow. Truly added sales come from selling days you’d otherwise lose, when a season window or sell off period closes.

Only the second is new money. Separate the two and you can rank partners and categories by where approval effort actually pays. More on the process in our guide to the licensed product approval process and on product approvals in Flowhaven.

Royalties: knowing what normal looks like

Ask a royalty manager about the hardest part of their month and you’ll rarely hear “the maths”. You’ll hear that every licensee reports in their own format.

AI is unusually good at translating that: mapping the file to your royalty structure, matching SKUs and territories, calculating against the agreement’s rates. The key is surfacing the lines that won’t reconcile instead of quietly swallowing them.

We also asked Claude what stands out in a royalty report. It found the expected things: royalties submitted but not invoiced, reports loaded months late, validation checks never run.

Then it said something a spreadsheet never would: this data is too perfect to trust.

Round totals. Unit prices that never move. One transaction date per quarter.

Its advice was to check against the licensee’s own records before you invoice. For where money usually slips through, read where licensing royalties leak. Then see how royalty management in Flowhaven checks every line against the agreement.

Brand assurance: designed, and next on our test list

Some of the most promising use cases are still on our test list. I’ll be clear about which is which.

The rebrand sweep is next on our test list. After a logo refresh, brand assurance checks every open submission by eye for the old logo. The idea is to let AI flag every item still carrying the retired logo, with a confidence note.

After a person confirms, it adds the same clear comment to each one. The reviewer’s eye is saved for the real judgement calls.

The customs check is designed, and we’re setting up to test it from a phone. A customs notice lands: shipment held, photos attached. Is it ours?

AI matches the held goods against approved products and their images, then checks the licensee’s rights for that territory. It returns a draft authorisation letter, or a “not ours” finding with the evidence.

Legal reviews every letter before it goes anywhere. This is evidence assembly, not legal advice, and counsel gets the full picture in minutes instead of days.

The style guide question is one we’re exploring. It’s 11pm in Japan, and a licensee wonders if they can use the secondary logo on a dark background. Today that’s an email, a wait, and often a rejection later.

What if they could ask the brand’s style guide and get an answer based on the guidelines and their granted rights? Including “please check with the brand team” when that’s the honest answer.

Planning: how many people does a programme need?

We gave Claude a demo org with a few dozen active agreements, around twenty licensees and about a hundred product submissions a year. It landed on roughly three full time people across licensing, brand assurance and royalties.

Then it built the profit view: people, software, expos, travel, creative. It added KPIs nobody asked for, like guarantee coverage and how much revenue depends on a single agreement. Licensing deserves a P&L, not just a spreadsheet of royalties.

What we learned: draw the lines first

The biggest lesson wasn’t about prompts. It was about where to draw lines.

Internal updates can run on a schedule. Anything addressed to a licensee is drafted by AI and sent by a person.

A daily programme recap posting itself to the team’s Slack channel is great. A revision note to a partner gets drafted, reviewed, then sent by a person.

External communication is not the place to discover an error.

• AI proposes, a person confirms before anything changes on a record.
• Legal reviews anything with legal weight, from authorisation letters to trademark evidence packs.
• Answers come from your own records, so every one can be checked.
• We test on demo data with fictional brands first.

What comes next

Being told, approving instead of doing, growing instead of running. That’s where our roadmap at Flowhaven is heading. On the test list right now: German and Japanese after Finnish, the customs check and the rebrand sweep.

What to do next

Start with the record. AI tools such as Claude connect to Flowhaven through an integration, and the answers are only as good as the data behind them. See how Flowhaven fits your stack on our integrations page.

And which shift is your team on today?