Incumbents make you rent a generic firehose per seat, forever — or build from components you can't engineer. This is the third option.
Per-seat platforms charge $10,000–$20,000 per analyst per year — $50,000–$100,000+ for a small team — and what arrives is a horizontal product your analysts must still triage by hand. Encoding your taxonomy breaks their unit economics, so they never will.
You pay per instance, not per seat. Every analyst in your organisation gets access for roughly what one or two incumbent seats cost — and the database, the corrections, and the accumulated evaluation set are yours to keep.
Everything domain-specific is your data. The system underneath never changes.
A new deployment is a re-configuration and a source-feasibility audit — not a rebuild.
No terminal, no GitHub account, no engineer on your payroll.
The verification loop is the moat.
Every credible buyer of intelligence tooling has the same fear: unverified machine output published under their name.
Silent acceptance is structurally impossible
Every category, country, and entity the model returns is re-checked server-side against your controlled vocabulary. Anything low-confidence or unresolvable routes to a review queue where your analysts confirm, correct, or reject it. Corrections log as auditable events, fix the live record immediately, and become verified ground truth.
Your experts correct the AI — they are not replaced by it
The system does the reading; your experts do the judging. That is the safety story, the accuracy story, and the reason analysts who fear automation will actually adopt it.
Every marketing claim is re-runnable
Accuracy is measured against a client-labelled evaluation set by an automated harness — not asserted in a testimonial. Incumbents publish quotes; we publish measurements you can reproduce on your own data.
Registered sources, never open-web crawling
We sell derived structured records and analysis — never redistributed article content. Per-source terms review is part of the feasibility audit, priced in before a line of ingestion code exists.
Four categories, four structural failures.
Pricing excludes the boutique buyer; horizontality forbids depth
Per-seat economics punish exactly the small expert teams that produce specialist intelligence, and volume is their feature while triage stays your problem.
They sell components to buyers who need systems
Everything ends at structured text. The review workflow, accuracy measurement, brief generation, and analyst dashboard — the last mile — is left to engineering staff you don't have.
Mention-counting, not record-building
They answer "who talked about us." An intelligence team needs "what happened, classified how, verified by whom" — with a controlled vocabulary and measured accuracy.
Rigour at annual latency, with no client deliverables
Hand-compiled academic datasets prove the demand for expert-verified structured data — and prove manual compilation can't deliver it at operational cadence.
Any domain with a defined source universe and a controlled framework.
Plus analysts who need structured records and briefs — not another alert stream.
Bilateral-corridor trackers
China–LatAm, Gulf–Africa, EU–Africa, India–Africa. The reference-data swap is near-mechanical.
Regulatory & policy monitoring
Pharma, energy, fintech, trade policy. The taxonomy is the regulatory framework; sources are official gazettes and trade press.
Sector & competitor policy intelligence
The taxonomy is your own strategic framework — which is precisely why a horizontal platform can never encode it.
Development-finance tracking
DFI commitments, sovereign lending, project pipelines — the datasets that prove the demand, at daily cadence.
Diplomatic-activity tracking
Conferences, delegations, bilateral visits — the entries model already accommodates them.
The scoping audit answers exactly that — with a go/no-go verdict and live probes of every candidate source, before anyone commits to a build.
The method is part of what you're buying.
Decision-complete planning
Planning documents you sign off before any code is written. No ambiguity carried into the build.
Cheap probes first
Live technical probes disqualify unworkable sources before they can burn a build cycle. This already saved the reference engagement weeks.
Fixed scope, fixed fee
The quote from scoping is the price. Scope changes are decisions you make, not invoices that arrive.
Measured, then handed over
Accuracy measured against your ground truth, and the instance handed over with its database and documentation.
Three stages, one sequence. Start at the first.
Only the scoping fee is charged today — and it is credited in full toward your build if you proceed.
Scoping engagement
A go/no-go verdict and a fixed build quote. Standalone value if you walk.
Scoping + full build
The finished pipeline, branded and measured, handed over as an asset you own.
Charges the $4,500 scoping fee and holds your build slot.
Build + retainer
Keeping the pipeline accurate and the intelligence flowing — not "hosting and bug fixes".
Get the AI Project Scoping Kit.
The first step we take with every client is scoping the problem before touching the build. We've packaged that step as the AI Project Scoping Kit — a working system you can run yourself, useful even if you never hire us.
Get the Scoping Kit — $49A scoping conversation takes thirty minutes: your sources, your taxonomy, your deliverables.
Book thirty minutes