Atoms is an engineering team of twelve years. Four disciplines, one standard: the source nobody could parse, the process nobody would quote, the report that never reconciles, the platform a previous team could not finish. Twelve years on that class of problem, and nothing else.
Have a task or a problem you need solved? Send it over, whatever its size. A great many of our clients arrived with something small and stayed for years — the first job turned into the next one, and then into a system we have been running ever since. You get an answer within 24 hours.
systems built and shipped since 2014
records collected and delivered
operations executed a day
years working only on hard problems
Each one is a full practice with its own scale, its own failure modes and its own page. What they share is the part that decides whether a system still runs in year three: exceptions, state, proof, and somebody still operating it.
Extraction systems for the most heavily defended sources in the industry — Ticketmaster, Zillow, Amazon, Booking and their class — at millions of records a day, with coverage you can prove rather than hope for.
Hardest classBehavioural bot management at national scale, held at full coverage for years.
Whole departments replaced end to end — intake, judgement, execution and control — running millions of operations a day across systems that were never designed to talk to each other, including the ones with no API.
Hardest classCross-department processes over undocumented legacy systems.
A BI licence gives you a chart over one clean table. We build what sits underneath: every internal and third-party system joined into one governed model, any kind of report on top, drilling down to a single transaction.
Hardest classGroup-level consolidated reporting over a dozen incompatible systems.
We do not build brochure sites, landing pages or catalogues — agencies do that well and cheaply. We build internal operating platforms, client portals, marketplaces and real-time interfaces with real domain logic behind them.
Hardest classReplacing a core internal system without stopping the business.
Extraction feeds the automation, the automation feeds the reporting, and the platform is where people work. Most engagements use more than one.
The same people carry all four disciplines, so there is nobody to hand the blame to when the seam between two of them breaks.
Almost no engagement here began as a project of this size. What it usually looks like at the start is directly below.
Nobody arrives with a department to automate. They arrive with one thing that annoys them — a report that takes a morning, a list somebody retypes, a number nobody trusts — and they are not sure it is worth asking about. That is where most of the systems above came from.
One script that pulled tracking numbers out of a carrier's portal every morning, because a person was copying them by hand.
Seven years later
The document desk for the whole group: 240k documents a day, 81 roles released, running unattended.
One report the finance team could not get out of their ERP: margin by store, weekly, without three days of manual work.
Five years later
The group's reporting platform: 14 source systems, 400M rows a day, month-end close down by 73%.
A parser for one competitor's price list, to test a hypothesis nobody was willing to fund properly yet.
Six years later
The pricing data layer the product runs on: 9.4M records a day at 99.4% measured coverage.
Send the task or the problem, however small. There is no minimum and no qualifying call before anyone looks at it.
A two-line question is answered the way a department-scale brief is answered: in writing, within 24 hours.
When an off-the-shelf tool solves the problem, you get that answer instead of a quote. That reply has started more long relationships than it has cost us.
Almost anyone can show you a working prototype. The reason projects like these fail is always the same five things — and all five surface months after everyone has gone home.
On genuinely hard work, a quote written from a brief is a guess — the spread between the described version and the real one is routinely tenfold. So where the unknowns can move the number tenfold, the first step is a fixed-fee pilot on your systems and your data: one slice running end to end, with the measured numbers and the honest risks. By the end of it you know what the full build costs, what it will do and where the difficult parts sit — after a month, rather than after a year of a project that drifts.
We have not pivoted. Since 2014 the work has been sources with real defences, processes with real exceptions, reporting over real mess and platforms with real domain logic. That is why the estimate for a hostile integration is a recollection rather than a guess, and why our straight-through rates hold in month nine instead of month one.
At the scale we work at, the recurring line — compute, proxies, document processing, model calls, licences — decides whether the system is worth having at all. We design the execution model per case, use the expensive machinery only where it earns its place, and report cost per thousand operations from the first week.
Shipping a system is the start of it, not the end. Hosting, monitoring, on-call, upstream changes, tuning and new features are all part of one monthly arrangement, handled by the same engineers who designed it. That is exactly why we build with tests, runbooks, infrastructure as code and versioned policies: a system we will still be running in year eight has to be engineered to be operated, not merely delivered. Most of our clients are on their third or fourth system with us.
Every engagement runs against systems and data the client owns or has the right to use, under credentials they authorise, with every action logged, attributable and reversible. Where a source or a process carries a legal question, we put it on the table in the first conversation rather than months in — that has cost us work, and it is worth it.
No discovery call before there is something to discover, no deck, no phased proposal for a problem nobody has tested yet. Bounded work goes straight from brief to done. Where the unknowns are large enough to move the price tenfold, the pilot is the sales process — and it works the same whether the work is a parser, a department, a reporting platform or a product.
1Brief
The source, process, report or platform; the systems it touches; the volume it carries; what changes for the business if it works. Half the time the scope narrows and gets cheaper right here.
OutputWritten feasibility view in 3 days
2Pilot
Two to four weeks on your real systems and real data. One slice running end to end, with measured coverage or throughput, measured running cost, and the risks stated plainly.
OutputRunning pilot + risk list
3Design
Target operating model, exception handling, throughput, availability and running cost — written into the contract rather than into a slide. After the pilot, the price is fixed.
OutputFixed scope & fixed price
4Build
Every week ships something that actually runs, shadowing whatever does the job today before it takes over. Monitoring, runbooks and documentation are built alongside it, not bolted on at the end.
OutputA system running in production
5Operate
Hosting, on-call, upstream changes, tuning and new features as the business moves. Routine change is covered by the monthly fee; anything large is quoted before it starts.
OutputA system you stop thinking about
Two from each discipline. Every one of them arrived after another team stalled, quoted it as undoable, or delivered something that could not be maintained. Clients are under NDA — the engineering detail we walk through against your own case.
Full event, seat and price coverage from a source with behavioural bot management, refreshed continuously rather than nightly.
Listings, history and agent data unified across 40+ portals with contradictory identities, deduplicated to one entity per property.
Shipping, customs and supplier paperwork in 40+ formats, parsed, checked against the ERP and posted, for a European freight group.
Intake, credit checks, allocation, invoicing and dunning unified across an ERP, two CRMs, a WMS and a 20-year-old accounting core.
Consolidated commercial and financial reporting for a multi-country group whose subsidiaries had never agreed on a definition of margin.
Board KPIs, store operations and category management on one model, drilling from a headline number to the transaction behind it.
A 600-screen internal operating system migrated module by module, with both systems live and reconciled through the whole transition.
Real-time order, document and settlement visibility for thousands of business customers, sourced from a back office that had never been exposed.
Measured production numbers, agreed with each client before anything is published.
Clients, their systems and anything that would identify them stay out of public material.
We walk through the architecture, the exception profile and the failure modes in full, against your own case.
We will be running these systems for years, so we pick technology that stays boring under load and still has a deep talent pool a decade from now. Nothing exotic without a reason we can write down, and nothing that only works while somebody is watching it.
Per problem, not per fashion
Columnar where it earns it
Screens people work in daily
Our infrastructure or yours
Machine learning and language models are used where they measurably beat rules: extraction from messy documents, classification, entity resolution.
They are kept out of the path wherever a deterministic rule is cheaper, faster and explainable — which is most of it.
Every automated decision stays traceable to the policy version that made it, and to the input it saw.
The engineers who scope your system are the ones who build it and then keep it running. That continuity is the whole point: three years in, someone here still remembers why a particular decision was made, and that is worth more than any document. The team is senior by design — we grow only as fast as we can staff work we are willing to stand behind.
Who we are
We started in 2014 doing extraction nobody else would take, and grew into the three disciplines that kept turning up next to it — the automation the data fed, the reporting the business needed on top, and the systems people actually work in. Twelve years later it is the same team, the same standard and the same class of problem.
Distributed across Europe, working in English, overlapping with European and US East Coast hours. Long engagements are the norm: most clients are on their third or fourth system with us.
How we work
Work prices by how hostile the environment is, not by how many features are listed, and there is no floor it has to clear. The rule is the same at every size: if being wrong about the estimate is cheap, we quote the work directly; if it is not, the price comes after a paid pilot has measured the unknowns. Either way it is fixed before anything starts.
Model
You start on whichever rung fits, and most clients start on the first one. From the Operate rung onward it is a single monthly arrangement covering hosting, monitoring, on-call and continued development, so nobody on your side has to build an operations team around the system. Infrastructure runs on ours or inside your own perimeter — whichever your security and finance people prefer.
Not our work
Notice that none of these is about size. Small is not on the list and never has been. If your problem is real but not our kind of work, we say so in the first reply and point you at someone better suited — including a $20-a-month tool when that is genuinely the right answer. It costs us nothing, it saves you a quarter, and it is how a surprising number of our longest clients met us.
Legal & governance
Systems run on infrastructure we operate for you, or inside your own perimeter, against data and systems the client owns or is authorised to use, under credentials the client controls. Every action is logged, attributable and reversible, and every automated decision is explainable. Access, retention and deletion controls are designed into the system rather than bolted on afterwards.
GDPR alignment and internal-control requirements are implemented through technical and organisational controls. Confirming the lawful basis, the regulatory treatment and the approval authority for a specific source, process or jurisdiction remains with the client's legal and risk teams — we give them the documentation to do it, and we raise the questions at the start rather than months into the build.
If yours isn't here, put it in the form — we answer in writing, within 24 hours, and without a discovery call first. Most questions get a straight yes or no rather than a proposal.
Start here
Send the problem in whatever detail you already have — a paragraph is fine, a full brief is fine. You get a straight answer: whether it can be done, what makes it hard if anything does, and roughly what it costs. Whatever the size — no deck, and no discovery call before there is anything to discover.