Data & AI infrastructure
We help small businesses and startups build their data and AI foundations in the right order — right-sized, documented, and cheap to run. No enterprise licences you will never use. No platform you cannot maintain after we leave.
The order that works
Analytics & AI
Modelling & Governance
Foundations
Most projects fail because they start at 03. We start at 01 — and get you to 03 faster because of it.
Sound familiar
These four symptoms come up in almost every first conversation we have. They all trace back to the same root cause — the layers underneath were never built.
Finance, sales and ops each keep their own spreadsheet. Every meeting starts by reconciling them instead of deciding anything.
A person you pay well spends the first day of every week copying exports into a workbook. That is a salary being spent on clipboard work.
A warehouse, three SaaS subscriptions and a BI licence — none of them connected, all of them billing monthly.
You want an assistant over your own documents and records, but nothing underneath is clean or reachable enough to point it at.
The Axxa layer model
Our logo is a stack for a reason. Each layer only works if the one beneath it is solid — which is why we refuse to start with the dashboard, however much you want one.
The part everyone wants to start with
Dashboards, forecasts and AI assistants. Genuinely valuable — and the thinnest layer, which is exactly why it needs the two beneath it.
Where trust is actually manufactured
Raw data becomes agreed definitions. Tests catch breakage, access is deliberate, and every number traces back to something you can point at.
Get this wrong and everything above it lies
Sources identified, ingestion automated, one warehouse that is the agreed source of truth. Unglamorous, and the only reason the layers above ever work.
Cost control and governance are not a fourth layer — they wrap all three. Every decision gets made with the monthly bill and the Data Protection Act already on the table.
What we do
Take them individually or as one engagement. Most clients start with an audit and end up somewhere they did not expect — which is rather the point of auditing first.
The plumbing: getting data out of your operational systems and into one place you can trust, on a schedule, without someone exporting spreadsheets every Monday.
Dashboards people actually open. We start from the decisions you need to make, then work backwards to the metrics — not the other way round.
Lightweight governance that a ten-person company can actually maintain — enough structure to stay compliant and audit-ready, not a committee and a 90-page policy binder.
AI that works because the data underneath it does. We build the retrieval, evaluation and cost controls that separate a useful assistant from an expensive demo.
We would rather make you independent than indispensable to us. Every engagement ends with your team able to run, extend and debug what we built.
Cost-effective by design
Cheap infrastructure and good infrastructure are not opposites. Most of the cost in a small company's data stack is paying for scale it does not have yet.
We are not against big tools. We are against paying enterprise prices for enterprise problems you do not have yet. When you outgrow the simple version, the migration path is already documented — we design for it from day one.
Proof · a real rebuild
A medium-sized South African fintech was paying $3,163 a month for a warehouse and BI licences. We rebuilt the stack bottom-up in exactly the order above for roughly $70 — and they came out of it with more, not less: customer-facing and stakeholder analytics that integrate directly into the applications they already run.
Before
$3,163
per month — cloud warehouse plus per-seat BI licences
After
~$70
per month — usage-based warehouse plus one small VPS
Saved
~$37,000 / year
a 98% reduction, with more analytics delivered — not less
Production Postgres replicated into a warehouse every hour, incrementally, with per-table state so one bad table cannot wedge the other hundred and six.
Each table is its own orchestrated asset on an hourly partition, merging on a primary key with an updated_at cursor. Small reference tables full-replace because that is simpler and they are tiny. Hourly schedule, no manual steps — reporting is never more than sixty minutes behind the business.
Raw tables become conformed dimensions and facts, then department-owned marts. Column descriptions are persisted into the warehouse itself, so definitions live next to the data rather than in a wiki nobody opens.
Staging views feed conformed dimensions and incremental fact tables, which feed eight marts. Each mart is owned by the department that actually reads it, so nobody argues about whose number is right.
Thirty analytics-ready reports served over a documented API — customer-facing and stakeholder analytics that drop straight into the applications they already run, instead of yet another dashboard nobody opens.
Every report maps onto a question the business actually asks — transaction volumes, failures by gateway, retention and churn, dormant high-value accounts, compliance review lists. Product teams integrate a number, not a query.
Why the bill collapsed
The old bill was mostly structural: a warehouse priced for scale they did not have, and per-seat licences for people who opened a dashboard twice a quarter. Neither was buying them better answers. Every component of the replacement is open source or billed on what it actually uses.
The $70 now covers a usage-based cloud warehouse and one small VPS running the orchestrator and the API. That is the entire bill.
What costs nothing
No per-seat licences. No per-row ingestion pricing. No Kubernetes. When they outgrow any single piece, the migration path is already documented — we design for it from the first week.
This is the reference build we bring to every engagement — not a diagram from a deck. A medium-sized South African fintech runs on it today.
How we work
Every stage produces something you own and could hand to somebody else. That is deliberate — it keeps us honest and keeps you free.
We map what you already have: systems, data, spend, people, and the decisions currently being made blind. You get a written findings document whether or not you hire us for anything else.
A right-sized architecture with named tools, real monthly cost figures, and a sequenced plan. You will know what it costs to run before anyone starts building.
We build in the order the layers demand — foundations first — shipping something usable every week rather than disappearing for two months.
Documentation, runbooks and training until your team can operate it without us. Stay on a monthly retainer if you want us close by; plenty of clients do not need to.
Pricing
One range, not a tier grid. Fixed scope and fixed price, agreed before anyone starts — and if the scope changes we tell you what it costs before we do it, not on the invoice afterwards.
Typical engagement
$500to$3,200
per engagement, priced on your company size and what you actually need
What moves the number
Early-stage startups get the audit side of things free: we look at what you have, tell you what we would do, and hand you the findings. No fee, and no obligation to hire us for the build. If the honest answer is that you do not need any of this yet, that is the answer you will get — and it will have cost you nothing.
Prices exclude VAT. Non-profits and educational institutions: ask about reduced rates.
Insights
Practical write-ups on the decisions small teams get asked to make — no email address required to read them.
What a genuinely useful warehouse, pipeline and BI setup costs a 15-person company — with the actual line items, not a vendor estimate.
A practical checklist of what the Act actually requires of a small business holding customer data — and the handful of things worth doing first.
Why retrieval assistants disappoint in production, and the four data problems that cause almost all of it.
Questions
Including the ones where the honest answer is “you probably don’t need us yet”.
Often, yes — and we will tell you so. If you have fewer than a handful of data sources and no recurring reporting pain, a well-structured spreadsheet and a clean database schema will serve you better than anything we would build. Ask anyway: for early-stage startups the audit is free, so finding out costs you nothing but an hour. The right time to hire us is when reporting has become somebody’s recurring job, or when you are about to make an expensive tooling decision you cannot easily reverse.
For most small businesses we work with, the monthly infrastructure bill lands somewhere in the tens of dollars — not thousands. The blueprint states the figure before any building starts, and we treat going over it as our problem to solve, not a change request to bill you for.
No, and the engagement is structured to make that credible. Everything runs on your cloud accounts under your credentials. The code is in your repository. Definitions are documented in plain language. If you decide to bring it in-house or move to another partner, you do not need our permission or our help.
Usually, and that is normally the cheaper answer. Most of what we find is sound infrastructure being used at 20% of its capability. The audit distinguishes between what is genuinely wrong, what is merely untidy, and what is fine and should be left alone.
You can, and we will build it — but the layer model is not a sales device, it is the reason AI projects fail. A retrieval assistant over inconsistent, unreachable data produces confident wrong answers, which is worse than no assistant. If your foundations are already sound, we go straight to the AI work. If they are not, we will show you exactly which gaps will sink it.
No. We know the East African context well — M-Pesa and payment rails, the Data Protection Act, the realities of local hiring — but our current reference build is for a South African fintech, and we work remotely with clients across East and Southern Africa without trouble.
Get started
No deck, no pitch. Tell us what is not working and we will tell you what we would do about it — including when the answer is that you do not need us yet.