Verification-
AI does the first 80% of a verification workflow. Expert humans close the last 20%. A daily learning loop keeps moving the line. What you buy is verified outcomes with an audit trail, not hours.
Who it’s for
Teams running high-stakes manual workflows across media and production: rights and royalties, production finance and tax credits, deliverables and content operations, in film and TV, music, and publishing. This is the work someone signs their name to before it ships or bills, where a demo that’s right 80% of the time is worthless, because the errors are where the money is: the unmatched royalty, the disallowed claim, the rejected delivery. Every item that stays unmatched is money that doesn’t reach the person it belongs to, and a question someone eventually has to answer.
The moment this bites hardest is when a catalog changes hands. After an acquisition, a library purchase, or a distribution migration, two systems’ versions of the truth have to be reconciled before anyone can bill against them.
We’re a Toronto operator with a Canada focus, whether the team doing the work is two people or twenty; for rights and royalty data, Canadian data residency is part of the offer. The method comes out of 25 years building ML systems for media companies, including Pandora and SiriusXM, and seven issued US patents in personalization.
- Unmatched royalties and metadata matching: recordings, compositions, editions, and the statements behind them
- Catalog and rights reconciliation: titles, terms, territories, and participations, checked against the contracts
- License and billing reconciliation: accounts, tariffs, and invoices that have to match reality
- Tax-credit claim verification: eligibility, labour breakdowns, and the supporting documents behind every line
- Document verification against eligibility rules: residency, entitlement, supporting paperwork
- Deliverables QC: packages checked against the distributor's spec before the distributor checks them
- Review queues and classification held to a professional standard
Two shapes, one loop underneath
Managed verification
Give us the backlog; we return verified results with an audit trail. Our reviewers work inside our loop, and your experts only adjudicate escalations and spot-check. For teams that have the problem but not the staff, or a backlog that outgrew the headcount. It's the lowest-adoption-risk way to start.
Judgment-loop pilot
We embed with your own experts; the app updates daily from their corrections. Your team keeps doing the job inside a working app that learns from every verdict. For teams that have the people and want the loop.
Both run on the same machinery, the human-guided learning loop: proposals from the system, one-click verdicts from experts, and an overnight improvement process gated by accuracy against the record.
Already have matching automation? We don’t replace it. The loop runs as a verification layer on top: your engine’s output becomes the proposals, and your experts’ verdicts give it the measured accuracy record it never had.
The engagement
- 01
Sit in
We join the people doing the manual workflow and learn the job as they actually do it.
- 02
Measure the baseline
The workflow as it stands: volume, time, error patterns. Every later claim is relative to this number.
- 03
First improvement, in days
A working app your experts judge with one click: agree, fix, or reject.
- 04
Daily updates
Overnight, verdicts become the next version. Each one is measured against the record before it ships.
- 05
The plateau
Improvement flattens. That's the finding, not the failure.
- 06
The frontier report
A measured map: what's machine-ready, what runs assisted, what stays human territory.
What you provide: access to the people doing the work, sample items, and a daily verdict on what shipped. That’s it. No requirements documents, no workshops, and no integration project; the loop runs alongside the tools you already use.
Because everything is measured against the day-one baseline, the engagement produces its own before-and-after: what the workflow cost, and what it costs now.
What you walk away with
A working improvement
To the workflow you already run, updated daily, with a plain-language changelog naming which of your corrections drove each change.
The judgment dataset
Your experts' own ground truth, captured as structured data. It's your asset, portable and yours to keep, and it travels with the catalog through migrations, reorganizations, and changes of ownership.
The frontier report
The evidence-backed answer to what machines can take at current capability and what still needs your people. It's the number to build a staffing plan around, and the artifact you take to whoever approves the next step.
The pilot
Fixed scope, fixed price: typically C$20–50K over 8–10 weeks. Scope tracks the size of the workflow, so a smaller catalog sits at the smaller end, and the window fits inside an off-season, which means the loop is trained before the next peak rather than during it. Your data and your workflow specifics remain yours; the measurement tooling remains Signal Foundry’s.
Our incentives run the opposite way from an hours-billing vendor: the product gets cheaper for you as it works. We’re measured by how few humans the workflow eventually needs, and by how much the remaining ones matter.
Fair questions
Is this going to replace my team?
No, and we'll show you exactly why not, with your own data. The engagement's final deliverable maps which parts of the workflow genuinely require human judgment. In every workflow we've measured, that number is not zero.
How fast do we see something?
A working improvement your team can judge in the first days of the engagement. The measurement starts on day one.
What do we actually own at the end?
The improved workflow, and the judgment dataset your experts created. That ground truth is portable and yours to keep.
Why daily updates?
Because feedback decays. A correction acted on overnight teaches the system while the context is fresh, and your experts see their judgment show up in the product tomorrow. That's what keeps them engaged.
What's in the audit trail?
Every verified item carries the trail of how it was decided: what the system proposed and how confident it was, whether it cleared the accuracy gate or went to a person, who judged it and what they decided, and which version of the system did the work. All of it is measured against your experts' own record.
Who sees our data, and where does the work happen?
The work runs from Toronto, with Canadian data residency for rights and royalty data. Your data and workflow specifics remain yours throughout, and the judgment dataset your experts create leaves with you. Security questions get answered by a person: hello@signalfoundry.ai.
What happens when the engagement ends?
The improved workflow and the judgment dataset leave with you; the dataset is useful with or without us. The pilot is self-contained. Fixed scope, and nothing you keep is licensed on an ongoing basis. The measurement tooling is ours, but nothing you walk away with depends on it.
Book a discovery conversation
Tell us about the workflow. If it’s a fit, the first thing we do is come sit with the people who run it.
hello@signalfoundry.ai