AI9OS vs. Hiring a Research Team: The Real Cost Comparison

2026-08-11 · Philip Choo · AI9OS

A platform and a research team do not compete for the same budget line, and treating them as substitutes is how practices end up paying twice. A platform buys capacity — how much open-source collection one desk can put through in a day, and how well that collection survives being questioned later. Headcount buys judgment — the decision about what a finding means, which no amount of parallel collection produces. This post is the comparison a licensed agency actually has to make, including the cases where the answer is "hire someone."

Why we won't quote you a cost-per-analyst figure

The usual version of this comparison runs the arithmetic: an analyst's fully loaded annual cost against a platform subscription, and the platform wins by a margin that looks decisive on a slide.

We are not going to run that arithmetic, because it is a true lie — arithmetic that is correct and still misleads. It works by quietly comparing a platform's licence cost against a person's total cost, while leaving out the platform's real costs: the time to learn it, the review burden it creates, and the analyst you still need in order to use its output responsibly. A figure assembled that way is not a comparison. It is a conclusion with numbers attached.

What follows compares the two structurally instead. If a number matters to your decision, build it from your own caseload — and put the omitted costs back in.

What headcount actually buys

An additional investigator buys four things a platform cannot supply:

1. Judgment about meaning. A platform can tell you two records agree. Only a person can tell you whether agreement is corroboration or the same source counted twice.

2. Accountability with a name on it. Regulators, courts and clients hold a licensed human responsible. Software is not a signatory.

3. Elastic scope. A person can be redirected mid-matter to something nobody anticipated. Software does what it was built to do.

4. Client and counsel handling. The parts of the work that are conversation, not collection.

A practice with no capable investigator does not have a tooling problem. It has a hiring problem, and buying a platform first will produce output nobody is qualified to sign.

What a platform actually buys

A platform earns its place on two axes, and it is worth being precise about both because they are often conflated.

Throughput. Manual open-source work is sequential — one investigator, one browser, one tab. A platform runs the same category of checks concurrently across every entity in a matter. The techniques are not new; the concurrency is. We set out where that line falls in AI9OS vs. Manual OSINT.

Evidentiary durability. This is the axis buyers underweight, and it is the one that decides matters. A folder of hand-named screenshots is an assertion; a tamper-evident, hash-chained record is a demonstrable history. Headcount does not fix this — adding a second investigator to a practice that stores evidence loosely produces twice as much evidence stored loosely. See chain of custody in the deepfake era for why that gap has widened.

There is a third, less obvious return: a platform enforces decisions that otherwise depend on memory. A no-contact list held in someone's head is a promise; a no-contact list checked before every collection task can run is a control. That distinction survives staff turnover. A person's discipline leaves when they do.

Where the platform is the wrong answer

Three cases, stated plainly, because a comparison that never concludes against the vendor is marketing:

• Your bottleneck is intake, not collection. If matters are arriving unscreened, more collection capacity multiplies the wrong work. Fix the case-acceptance gate first; it costs nothing and prevents more loss than any tool.
• Your volume genuinely is low. A practice running a handful of light matters a month will not recover the learning curve. Manual work is slower per matter and that may simply be acceptable.
• The work is not open-source. Interviews, physical surveillance, and anything requiring a subject's cooperation are outside what any OSINT platform does. A tool that appears to reach those is doing something you should refuse.

The comparison that actually decides it

Ask what your practice is short of. If you are turning down matters because nobody can get through the collection in time, that is a capacity constraint, and capacity is the thing a platform is genuinely good at. If you are accepting matters you should decline, or you cannot defend how an exhibit was handled when opposing counsel asks, no amount of headcount fixes it — those are process constraints, and a platform that enforces process structurally is the cheaper fix.

And if what you are short of is someone who can look at a contradictory set of findings and say what it means, hire that person. That capability has never been the thing automation was going to replace, and any vendor implying otherwise is selling you the wrong half of the product.

General information for practitioners, not legal advice. AI9OS is an open-source-intelligence technology platform; investigation services are conducted solely by licensed agencies under Singapore's Private Security Industry Act.