Job Description

Hi 👋 We’re Legl.

Legl is building the operating system for modern legal services.

We help law firms and regulated businesses replace manual, fragmented workflows with intelligent software from client onboarding and compliance to payments, risk, and reporting.

Legal work is high-stakes. It’s regulated, complex, and deeply human. The software supporting it has historically been slow, manual and brittle. We believe it doesn’t have to be that way.

We’re backed by leading European VCs (Series B), scaling quickly, partnered with over 550 law firms including 40 of the UK’s top 200, launched in the UK and Australia - and entering our next phase of growth.

AI-Native by Default

AI-centricity is expected and part of how we work day to day. We encourage and expect everyone to treat it as a core tool rather than a novelty. Across the company people use AI to do their jobs better, faster and at higher quality, and we invest in the tooling to make that real. We also expect people to have the judgement to know when AI's output is wrong - speed without correctness is worse than useless. If you see AI as a threat to your craft rather than a multiplier of it, this won't feel like home.

What You’ll Do

As a Data Business Analyst at Legl, you'll:

• Own business problems, not just data requests. We obsess with the business problem / goal behind any number, dashboard or report request. You'll reframe asks, challenge needs, and define real problems worth solving before defaulting to build.

• Investigate, prioritise and sequence. Take ambiguous, high-stakes questions, run discovery with stakeholders across the business, and make the call on what's worth building and in what order — comfortable pushing back, scoping down, or saying "not yet."

• Build the technical foundations others can't. This is the spine of the role. Model the data properly — including the genuinely hard problems that are currently blocked or hand-waved: joining product/Django data to HubSpot reliably at the right grain, modelling slowly changing dimensions (e.g. historic price and SKU lists), and building defensible COGS models. You'll bring engineering discipline (version control, testing, review loops, documentation) in service of reliable, reusable outputs — and you'll have a clear view on the implementation trade-offs (the quick 90% answer today vs. the fully modelled version with a maintenance process behind it) and which one the business actually needs.

• Own correctness and validation. AI makes it cheap to produce work that looks finished but isn't. Part of why this role exists is to make sure that doesn't happen: you own testing and validating data models, the knowledge layer, and Ada's outputs so that wrong answers get caught before anyone trusts them — not by chance, but by design.

• Own the semantic / knowledge layer as a single source of truth. Define each metric once so the same number is computed the same way everywhere, and keep the knowledge layer that AI tools rely on accurate and current — treating that as an ongoing workstream, not a one-off.

• Deliver data-led solutions end to end. Turn a problem into something that ships and changes a decision — a metric, a model, a dashboard, an automation — and own it through to adoption and impact, not just delivery.

• Translate between technical and commercial. Operate as the conduit between technical & commercial stakeholders; quantifying builds & impact in a way which resonates with the audience.

• Use AI as leverage, and know exactly where it breaks. Lean on LLMs for speed — but the differentiator here isn't offloading SQL, it's the technical judgement to verify what AI produces, spot where it's confidently wrong, and understand the access, permissioning and governance that agent-style patterns require. AI lowers the floor on plausible-looking output, which raises the premium on someone who can tell the difference.

This Role Is a Great Fit If…

• You think in problems and decisions, not tickets — you instinctively ask "what's this actually for?" before building anything, and you measure yourself on impact, not output.

• You're comfortable owning ambiguity: you can take a vague, high-stakes question, run discovery, prioritise, and commit to a direction without waiting for perfect inputs.

• You have real analytics-engineering craft, and you've done the hard parts. Strong data modelling and hands-on experience building tested, version-controlled transformations on a cloud data warehouse (e.g. Snowflake, BigQuery, Redshift) — and you've personally untangled messy technical problems: getting systems to join at the right grain, modelling slowly changing dimensions, building cost models. You treat data as a product, not a pile of queries.

• You own metric definitions as a single source of truth, and you've untangled the kind of mess where the same metric is calculated differently in two places.

• You can review and validate what AI produces. You're fluent enough in SQL and data modelling to know when a plausible-looking answer is actually wrong, you already use LLMs day-to-day, you build validation into your work as a matter of course, and you have a clear view on where AI should and shouldn't be trusted.

• You're a genuine translator — equally credible with engineers and commercial leaders — and you bring a view on access, permissioning and governance rather than treating it as an afterthought.

This Role Is Not a Great Fit If…

• You wait to be handed fully specified requirements, and see the job as fulfilling requests rather than solving problems.

• You optimise for shipping output over changing decisions — counting dashboards rather than measuring impact.

• Your strength is one-off queries and dashboards, and you've never built or owned a modelled, tested transformation layer — or never had to solve the harder modelling problems (grain, joins across systems, slowly changing dimensions, cost models).

• You're comfortable shipping AI-generated work that looks right but you can't verify the technical detail underneath — you reach for the quick plausible answer rather than understanding the implementation trade-off.

• You treat AI as either magic or a threat, and haven't actually shipped LLM-enabled work.