About SystemMatch
SystemMatch takes a site brief written the way a customer actually writes one — "three floors, four offices each, lift lobby, perimeter cameras" — and returns a zone-by-zone design, a multi-vendor comparison, a compatibility check and a draft RFQ. The compatibility check is the part that matters, and it is the part that isn't AI.
Where this came from
SystemMatch exists because of a specific order, on a specific job, that went out wrong. It was not an exotic mistake. It is one of the most common errors in access control, and it is invisible on paper right up until an engineer is standing at a door with a reader that will not enrol.
A Gallagher Controller 7000, a Gallagher 8H HBUS module, and third-party HID readers. Everything on the order was a real, current, in-stock part.
Gallagher's HBUS bus talks to Gallagher's own T-Series readers. It does not talk to an HID reader. The 8H module supplied eight HBUS ports — the wrong module family entirely, not a missing accessory. Result: wrong goods on site, a returns process, a re-order, an engineer visit written off, and a delay the customer saw. Nobody had been careless. The information needed to catch it was spread across three manufacturer documents that nobody reads at quotation stage.
A price list would not have caught this. A CRM would not have caught this. A general-purpose AI would have produced a confident, tidy, wrong answer — because the wrongness is not in the language, it is in a hardware fact you either know or you don't. That gap is the whole product.
The job as it is done now
An email, a site-visit note, a page of a tender. Rarely a schedule, almost never a drawing you can take off from.
Usually the platform the estimator has installed most recently, or whichever manufacturer's rep called last. Failure point one: the design reflects familiarity rather than fit.
Copied from the last similar job. Failure point two: the interface module, the door contact, the request-to-exit device and the PSU headroom are the lines most often carried over incorrectly or dropped.
Then everyone waits. Failure point three: the RFQ doesn't ask the distributor to confirm the one part nobody is certain about, so nobody confirms it.
Cutting the four hours is worth something. Catching step three is worth an order of magnitude more, because a wrong module doesn't cost you the module — it costs the return, the re-order, the second engineer visit, the programme delay, and a customer who now checks your next submission twice.
What SystemMatch produces
Every card below also says what that output deliberately does not do. A pre-sales tool that overstates its scope is worse than no tool, because it invites you to stop checking.
Each area gets the full access-control chain — reader, the exact interface module, the controller it lands on, and why — plus camera type and count. Structured so an installer could order against it.
Does not: produce cable schedules, containment, or connector-level BOM. That is a design office's job, not a pre-sales tool's.
The same requirement expressed in more than one manufacturer's kit, side by side, with specification values taken from the sourced catalogue rather than from the model's prose. The report shows each entry's verification status.
Does not: carry prices. Pricing is a live commercial negotiation with your distributor, and a stale number in a proposal is a liability.
The brief and the generated design are both tested against a maintained rules list. Anything that will not work, or that is missing a required part, is flagged with severity and the fix.
Does not: guess. If no rule covers your combination, it says the combination is not covered — it does not invent reassurance.
A structured enquiry to your distributor: scope, schedule of equipment, commercial requirements, and a points requiring confirmation section that names the exact part numbers nobody is certain about.
Does not: send itself. You read it, you edit it, you put your name on it.
A branded PDF proposal you can put in front of a customer, and one Excel working file with the schedule, the comparison and the open points on separate sheets — one workbook, not a folder of them.
The part that isn't AI
This is the most important paragraph on this page, so it is worth being blunt about the architecture.
A language model drafts the design. It is good at reading a messy brief and turning it into zones, counts and sensible equipment types. That is a language problem and it solves it well.
A separate rules engine runs the compatibility check. It is ordinary deterministic code matching against a hand-written list. The model cannot add to that list, cannot edit it, and cannot talk its way past it. Every rule was written from a manufacturer's own documentation, carries a link to that source, and is switched off until a person with the trade knowledge to judge it has signed it off by name.
The practical consequence: a warning you see is a warning somebody stood behind. Here is a real one, exactly as it is held in the rules file.
Gallagher HBUS bus will not talk to a third-party reader
Gallagher's native HBUS reader bus only communicates with Gallagher's own readers. A third-party reader (for example HID) cannot connect over HBUS.
Requirement: an OSDP or Wiegand interface module must be ordered alongside the controller to bridge the third-party reader.
id gallagher-hbus-thirdparty-reader
source Gallagher HBUS Modules product documentation · confidence high · signed off by a named reviewer before going live
Because the failure mode of a confident wrong answer in this trade is expensive and slow to surface. A tool that is silent where it doesn't know is more useful than one that is fluent everywhere. The rules list grows slowly and deliberately, one signed-off entry at a time. That is a feature.
Where the product data comes from
The catalogue behind the comparison holds 80 products across 17 manufacturers. Every single record carries the URL it was taken from, the date it was checked, and a confidence marker. Where a specification has not been confirmed against a primary source, the report labels it Unconfirmed rather than filling the gap with something plausible.
SystemMatch does not bulk-crawl manufacturer or distributor websites. That is a deliberate decision, taken for two reasons. The first is that mass scraping generally breaches those sites' terms and, in the UK, runs into database right — and an integrator's procurement review will ask about it. The second is more practical: a scraped catalogue is not an asset. Distributors publish the same data free. Anyone can copy it in a week.
What cannot be copied in a week is a list of the mistakes that actually get made, each one verified by somebody who has made or seen it.
Source URL and a checked-on date on all 80 products, so a specification in your proposal can be traced back to the manufacturer's own page.
An unverified value is shown as unverified. The comparison table marks it, and the PDF carries the same marking through to the customer copy.
Where a product is in the catalogue, its sourced values replace whatever the model wrote. The catalogue record is the authority for those fields, and its human-verification status remains visible.
Being straight about it
If you are evaluating this against an in-house sales tool, these are the boundaries you would find in week two. Better you have them in minute two.
Who built it
SystemMatch is a UK product, built by someone working in building security systems — CCTV, access control and intercom — rather than by a software company forming a view of the industry from the outside.
That shows up in small decisions more than large ones. The design output structures the full reader-to-controller chain because that is the chain that goes wrong. Prices are excluded because a stale price in a customer document is worse than no price. The RFQ has a "points requiring confirmation" section because the questions you fail to ask the distributor are the ones that come back as variations. Compliance items are worded as questions because that is how a responsible estimator raises them.
It is currently in private testing with a small group of named colleagues, each on their own access code, so that the rules list is stress-tested against real briefs before it is opened up more widely.
The most useful thing you can do is give it a brief you have already quoted, and check what it flags against what actually happened on that job. That is the only test that means anything, and it is the one this product is built to survive.
You can run one real brief with the self-service free trial. Private testers also receive an individual access code and report allowance while SystemMatch is in testing.
Open the app and enter it. Bring a brief you have already quoted — that is the test worth running.
Open the appOpen the app and choose Start a free trial for one full report and one watermarked download. Ongoing beta access is still issued individually.
Start a free trial