SparsemaxLabs

Prototyping · data processing · technical evaluation — St Helens, England

Test the idea in daylight before you build it in the dark.

Sparsemax Labs is a small software practice for organisations facing a technical question they cannot answer by reading brochures. We build working prototypes, process messy data into something you can reason with, and evaluate tools and approaches against your actual conditions — then tell you plainly what we found.

Bring us a question Try the evaluation matrix

Illustrative image: azurite mineral samples in small trays on a pale green bench

Fig. 1 Specimens before analysis. The interesting part is always what you did not expect to find.

Three questions we hear most

Q. Will this actually work with our data?

Often the honest answer is “partly”. We take a representative sample under an agreed data-handling arrangement, build the smallest thing that exercises the risky part, and measure. You get the result, the conditions it holds under, and where it breaks.

Q. Should we build this ourselves or buy a product?

We compare two or three realistic options against criteria you weight — fit, cost of ownership, lock-in, skills needed, time to useful — using hands-on trials rather than vendor demonstrations. We do not take referral fees from anyone.

Q. Our data is a mess. Where do we even start?

With an inventory and a single useful output. We trace where the data comes from, what each field means in practice, and build a repeatable processing pipeline for one output that matters. Then extend.

Three kinds of investigation

How we judge whether a prototype worked

A prototype succeeds when it answers its question, not when it looks finished. Before writing code we agree, in writing:

  1. The question. One sentence. “Can we match supplier invoices to orders automatically for at least most lines?” is good; “explore AI for finance” is not.
  2. The evidence. What measurement, observed with which data, would convince a sceptic in your organisation.
  3. The boundary. What we deliberately leave out — security hardening, scale, edge cases — so nobody mistakes the prototype for a product.
  4. The time box. A fixed number of days. If we have not answered the question by then, that itself is a finding.
  5. The next decision. What you will do with a yes, a no, or a “yes, if”.

Data processing, without the fog

Before

Spreadsheets exported from three systems, columns renamed by hand each month, duplicate customers, dates in four formats, and one person who knows which tab to trust. Reports take two days and nobody is fully sure of them.

After

A documented pipeline that ingests the same exports, applies named, tested rules, flags records it cannot resolve rather than guessing, and produces the output in minutes. The person who knew the tabs now reviews exceptions instead of rebuilding them.

We work in tools your team can maintain — commonly Python and SQL, sometimes your existing data platform — and hand over code, tests and a plain-English description of each rule. We process personal data only under a written agreement and only the minimum needed.

Working in daylight means

Questions that fit our way of working

The common thread is a decision waiting on evidence. These are the shapes of question we are best suited to — not past projects, but the kinds of thing worth bringing us.

feasibilityCan our scanned delivery notes be read reliably enough to remove manual keying, and what share would still need a person?

selectionWhich of three scheduling platforms copes with our shift patterns without customisation we would regret?

processingCan monthly reporting from four systems run in minutes instead of days, with every number traceable?

performanceWill a search approach still respond quickly when the catalogue is ten times larger than today?

usabilityWill field staff actually use a simple mobile form, or will they keep writing on paper?

integrationDoes the supplier's API expose what we need, or will we be exporting spreadsheets forever?

What we ask of your time

Investigations go faster with less of your time than you might expect, but not with none. Typically we need one short framing conversation, an owner who answers questions within a day or two, access to representative data or systems arranged through your own processes, and an hour with a domain expert each week. At the end, we ask for an hour to walk through findings with the people who will decide.

In return you get a fixed time box, a fixed fee for that box, a weekly note on progress and surprises, and a final findings document that someone who was not involved can read and understand. If the question changes part-way — which happens, and is often the most useful outcome — we stop, agree the new question in writing, and only then carry on.

Interactive

Weigh build, buy or wait

Move six sliders to reflect what matters to you — speed, control, cost, skills, data sensitivity, certainty of need — and the matrix shows which route your priorities lean towards, and why.

Open the matrix

Checklist

Is your data ready?

Twelve questions about ownership, access, definitions and quality. Find out what to settle before a data project starts, so the first weeks are spent on analysis, not archaeology.

Open the checklist

From the notebook

What we will not tell you

We do not claim research breakthroughs or novel algorithms. Our work applies established techniques carefully to your situation. A prototype is evidence, not a guarantee that a production system will behave the same way at scale. We do not sell software licences, take commissions from vendors, or give legal, financial or regulatory advice; where those questions arise, we will point them out so you can take them to the right adviser.

We also decline work where the answer has already been decided and the investigation is meant to decorate it. If a stakeholder needs a particular result, we would rather say so at the framing stage than spend your budget producing a report nobody can rely on. And where a question is too large for a time box, we will suggest splitting it into smaller ones, each with its own decision at the end, so that you can stop as soon as you know enough.

Write down the question you keep circling.

Send it to us as it is, however rough. Within two working days we will reply with clarifying questions, a suggested time box, and our honest view of whether a prototype, a data pipeline or an evaluation would answer it best — or whether you need none of them.

Open an enquiry