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 of it, multi-level and drilling down to a single transaction — at any scale and any complexity.
Twelve years, 300+ analytical systems, 400M+ rows refreshed a day across 90+ kinds of source system. 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. You get an answer within 24 hours.
dashboards and reporting systems shipped
kinds of system connected as a source
rows refreshed a day on one platform
years building analytics on messy data
From a wall-mounted operations screen that updates every thirty seconds to a consolidated group P&L that has to reconcile to the cent across eleven legal entities. Same platform underneath, different surface on top.
One page the whole board reads the same way: plan versus actual, entity and segment splits, and a defensible answer when someone asks where a number came from.
Consolidated · governed
The screen a duty manager watches: orders, queues, SLA clocks, capacity and anomalies, refreshed in seconds rather than overnight.
Seconds · alerting
Multi-entity consolidation, intercompany elimination, multi-currency, cost allocation and a month-end close that stops being a two-week manual exercise.
Reconciled · auditable
Retention, LTV, payback and margin per cohort, channel and product — computed on event-level history rather than assembled by hand each quarter.
Event level · historical
Stock, movement, lead times, service level and shortfall forecasting across warehouses, suppliers and channels that each count stock differently.
Cross-system · forecast
Spend from every ad platform joined to revenue in the CRM and the billing system, on one attribution model everyone has agreed to.
Joined spend · revenue
Fixed-format regulatory returns and internal control reports, generated on schedule with lineage from every figure back to its source record.
Lineage · fixed format
The one that has never reconciled, or that three teams have quoted and abandoned. Two weeks, fixed fee, one slice running on your real data —.
Bring it to us
Every report we ship carries its definitions with it: what the metric means, which source records it came from, when it was last refreshed and what changed since the last version. A number without that provenance is a slide, not a report — and it is the reason most dashboards quietly stop being trusted in month four.
Third-party CRMs, internal systems, the warehouse, the ad accounts, the billing engine, the 2004 database nobody wants to touch and the spreadsheet that four people still edit by hand. If it holds a number the business reports on, it becomes a source.
Where a documented API exists we use it. Where it does not, we read the database directly under agreed constraints, take change-data-capture off the replica, pull file and protocol exchanges, or build the adapter ourselves — the same integration engineering our automation practice does every day. A system without a marketplace connector is a normal source, not an exclusion.
Two systems rarely agree on what a customer is, when revenue is recognised, or which currency rate applied on the day. We build the entity resolution, the conformed dimensions and the history — including slowly changing attributes — so a figure computed across four systems means one thing and keeps meaning it after a reorganisation.
Metrics live in a governed semantic layer, versioned like code, with a written definition, an owner and a version history. Every dashboard, export and API call reads from the same place, so two departments cannot arrive at a meeting with two different revenue numbers and no way to settle which is right.
Sources break, backfill, rename fields and restate history. Loads are incremental and idempotent, every run is tested against control totals and freshness rules, and a bad load is quarantined and replayed rather than published — because a dashboard that is quietly wrong is worse than one that is visibly late.
The board number and the invoice behind it have to be the same truth. We build reports that drill from a single group-level KPI through entity, segment, product and cohort to the individual transaction — every level computed from the same model, so the totals always tie.
Complexity is not only depth. Comparison against plan, budget versions and last year; rolling and period-to-date windows; multi-currency at the correct historical rate; allocations and eliminations; restated history after a reorganisation; row-level access so a country manager sees their country and the CFO sees everything — all of it belongs in the model, not in a spreadsheet somebody maintains beside the dashboard.
Power BI, Tableau, Looker and Metabase are good tools, and we build on them when they fit — the failure is almost never the chart. It is the six things underneath, and every serious reporting project meets all six. Here they are, and what we do about each.
The marketplace covers the popular SaaS tools and stops exactly at your internal systems, your legacy core and the supplier feed that arrives as mail attachments — which is where the interesting numbers live.
What we doWe build the adapter for anything: undocumented APIs, direct database and CDC, protocol and file exchange, internal systems written in-house. Any source, no exclusions.
Each team builds its own logic in its own workbook. Six months later no two reports agree, nobody can say which is right, and the meeting is spent arguing about the data instead of the business.
What we doA governed semantic layer: one versioned definition per metric, one owner, and every surface reading from it.
The dashboard answers what happened but not why, so the real analysis happens in a spreadsheet built by hand each month — and that spreadsheet becomes the report the business actually runs on.
What we doDrill-through from group KPI to source document, every level computed from the same model, with totals that reconcile at each step.
An extract that worked at ten million rows takes forty minutes at four hundred million, refreshes overnight if it finishes at all, and quietly gets sampled until the numbers stop being true.
What we doAn engineered warehouse: incremental loads, partitioning, pre-aggregation and query design so response stays interactive on billions of rows.
A nightly batch is fine for a monthly review and useless for a duty manager, a trading desk or a marketplace — and one broken load means the screen shows a confident number that is two days old.
What we doFreshness is a per-metric requirement: streaming and CDC where seconds matter, batch where they don't, and a visible last-updated stamp with alerting when it slips.
Logic buried in a proprietary workbook nobody can review, a per-seat licence that makes “show it to the whole warehouse team” a budget conversation, and no way to move without rebuilding.
What we doModels, transformations and definitions as code, in open storage formats, on infrastructure we run for you. Unlimited internal viewers — no per-seat meter on who is allowed to look at a number.
We build at any scale — a single operational screen for a department, or a reporting platform that serves thousands of people across a group. The architecture is chosen for the size of the problem, not sold as one fixed stack.
Operating envelope
Volume, freshness, depth and concurrency are fixed on day one and written into the SLA — not discovered in month four when the morning refresh stops finishing before the morning. Small engagements start small on purpose: the same model, sized to what you actually have.
Where the volume goes
Illustrative profile of one production platform. Your envelope is fixed in the design phase and written into the SLA.
Every analytical system we ship runs on the same contour. Blocks change per client; the property that does not change is that a figure on a screen can be followed back to the source record that produced it, and a bad load is caught before anyone reads it.
The difference between a BI contractor and an analytics engineering team shows up half a year after launch — in whether anyone still opens the dashboard, and whether they believe it when they do.
Before anything is built we establish who reads the report, what decision it changes, and what number would make them act differently. Half of the dashboards we are asked for shrink to a third of their size at this step — and the ones that survive get used, because they answer a question somebody actually has on a Monday morning.
A new report is not finished when it renders. It is finished when its totals have been tied to the source systems and to whatever the business used before — usually a spreadsheet — and every difference has been explained rather than averaged away. That exercise is where the real data problems surface, and it is the reason our reports survive their first audit.
Eliminations, allocations, historical currency rates, plan versions, restated hierarchies and row-level access are all modelled once, centrally, and tested. They do not live in a filter someone remembers to set, or in a workbook that only one analyst can maintain — which is how organisations end up with a reporting function that cannot take a holiday.
Transformations, metric definitions, tests, infrastructure and runbooks, all as code. Shipping is the start of it, not the end: hosting, monitoring, on-call, upstream changes, tuning and new features all sit inside one monthly arrangement, handled by the same engineers who designed the thing. A system we will still be running in year eight has to be engineered to be operated, not merely delivered — and most of our clients are on their third or fourth system with us.
A reporting platform that a board reads is a production system with an uptime, not a project that was delivered once. These are the commitments in the support agreement; tiers and windows are set per project against how critical the reporting is.
| Commitment | Target | What it means in practice |
|---|---|---|
Freshness |
Per metric | Every metric carries an agreed freshness window — thirty seconds, hourly or daily — shown on the report itself and alerted on when it slips. Nobody has to guess how old a number is. |
Correctness |
Every load | Tests, control totals and reconciliation against the source systems run on every load. A load that fails them is quarantined and replayed rather than published to the dashboard. |
Query latency |
< 2 s p95 | Interactive response on the agreed model size and concurrency, including drill-through. Slow queries are treated as defects, not as the nature of big data. |
Monitoring |
24 / 7 | Failed and late loads, schema drift at a source, volume anomalies and metric jumps are detected by us and raised to you — not noticed by a director during a board meeting. |
Engineering response |
From 4 h | Incident tiers with named windows. A source system upgrade, a changed field, a new entity or a redefined metric is routine work covered by the retainer, not a change request. |
Availability |
99.9 % | Measured on completed refreshes against schedule and on dashboard availability, reported monthly. You get the same monitoring we watch. |
Each of these came to us after a BI rollout stalled, an analyst left with the logic in their head, or a consultancy delivered a dashboard the business stopped trusting. Clients are under NDA — the engineering detail we walk through against your own case.
Consolidated management reporting for a multi-country retail group, from board KPI down to the individual receipt, closing daily instead of monthly.
What was hardFour ERPs with different charts of accounts, intercompany eliminations, three currencies at historical rates, and a close assembled by hand in 40+ spreadsheets.
Live order flow, courier capacity, SLA breaches and anomaly alerting on one screen for operations, plus the same numbers as a daily executive view.
What was hardEvent streams and transactional databases that disagreed by construction, a thirty-second freshness requirement, and thousands of concurrent viewers at peak.
Portfolio, vintage and risk reporting for a consumer lender, with fixed-format regulatory returns generated on schedule and traceable to source records.
What was hardRestated history after a portfolio migration, six levels of drill-down required by the regulator, and a rule that every figure must be reproducible as of any past date.
On real reporting, a quote written from a brief is a guess — the gap between the data as described and the data as it is runs to a factor of ten. So we build one slice on your actual systems first, for a fixed fee.
1Brief
Who reads it, what they decide, which metrics matter, where the data lives and how fresh it has to be. Often the scope narrows and gets cheaper right here.
OutputMetric list & source map
2Pilot
Two to four weeks on your real sources: one report built through the full stack, its totals reconciled against your current numbers, plus the honest list of data problems we found.
OutputWorking report + data risk list
3Design
Data model, metric definitions, refresh windows, access rules, volume envelope and running cost — written into the contract, not into a slide.
OutputFixed scope & fixed price
4Build
Every week ships a report or a source that runs in your environment and gets used. Monitoring, runbooks and documentation are built alongside it, not bolted on at the end.
OutputA system running in production
5Operate
On-call for loads and freshness, handling source-system changes, new metrics and new reports as the business asks for them. Routine work sits in the retainer.
OutputNumbers nobody argues about
Analytics prices by the state of the data and the number of systems that have to agree, not by the number of charts. So we price in two steps, and we say no early when the work is not ours.
Model
You start on whichever rung fits, and most clients start with one small first task. 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 it. Infrastructure runs on ours or inside your own perimeter — whichever your security and finance people prefer.
Not our work
If your reporting is real but not our kind of work, we will say so in the first reply and point you at someone better suited — including an off-the-shelf tool when that is genuinely the right answer. That costs us nothing and saves you a quarter.
Legal & governance
Platforms run inside the client's own perimeter, read only the sources the client authorises, and enforce row- and column-level access so each viewer sees exactly what their role allows. Personal data is minimised at ingest, and who saw which report is itself logged.
GDPR alignment and internal-control requirements are implemented through technical and organisational controls. Confirming the lawful basis, the retention rules and the reporting obligations for a specific dataset and jurisdiction remains with the client’s legal and risk teams — we give them the lineage and documentation to do it.
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.
Bring us the hard report
Send the report you need, the systems the data sits in and how fresh it has to be. You get a straight answer: whether it can be built, what specifically makes it hard in your data, and roughly what it costs to build and to run. No deck, no discovery call before there is anything to discover.