# Cases Every engagement starts with a real problem in a real organisation. These are a few of the systems we've built alongside our partners, from first conversation to running production. ## Taxi & mobility A national taxi-fleet operator runs the operating layer behind a large share of the country's taxi fleet. That is three product lines spanning driver settlement, dispatch and B2B booking, and the data and processes that drive them had outgrown a one-team setup. Rather than fragmenting work across more developers, Foundry helped re-shape the organisation to be AI-leveraged: a Foundry-team Linear footprint, an agent-ready bridge into their existing stack, and Claude put into the hands of the people closest to operations. - **One-team ceiling** — A national taxi-fleet operator runs three product lines: driver settlement, dispatch, B2B booking. The data and processes behind them had outgrown a one-team setup. - **Headcount would have fragmented** — The obvious move was hiring more developers. That fragments the work without addressing why a single team had become the bottleneck. - **Reshaped as AI-leveraged org** — Foundry instead reshaped the organisation to be AI-leveraged: a Foundry-team Linear footprint, an agent-ready bridge into the existing stack, and Claude in the hands of the people closest to operations. - **Driver dashboard live, rolling out** — The first concrete output is a driver dashboard turning Databricks-resident ride and earnings data into a per-driver view of how the night went. The pattern is rolling out across the group. ## Commercial kitchen supply A commercial kitchen supplier's catalogue and pricing live in Navision, and quoting a full kitchen used to mean a sales rep digging through Navision by hand, copying numbers, and assembling a document. Foundry connected an AI layer on top of Navision: the rep describes the kitchen, the system searches the live catalogue, drafts the quote, and the rep approves. The pattern is in continuous operation under a drift agreement. - **Manual Navision quoting** — A commercial kitchen supplier's catalogue and pricing live in Navision, and quoting a full kitchen meant a rep digging through it by hand. - **Reps copied numbers by hand** — Every quote was a copy-paste exercise: pull numbers from Navision, assemble a document, hope nothing got out of step with the live catalogue. - **AI layer over Navision** — Foundry connected an AI layer on top of Navision: the rep describes the kitchen, the system searches the live catalogue, drafts the quote, the rep approves. Now in continuous drift. - **Productised as Foundry Quote** — The integration has been productised as Foundry Quote and is rolling to a second customer in a different industry. The next phase puts agents on top of the surrounding process. ## B2B sales (Navision) A B2B sales team has a five-step sales process: qualify, scope, configure, quote, close. It was being run out of spreadsheets and a Navision instance that wasn't designed for it. Foundry automated the whole flow end-to-end: a dashboard surfaces the live pipeline, a structured questionnaire captures customer context as deals move forward, and an AI layer produces the sales recommendation based on what's been entered and what Navision already knows. - **Five steps, no system** — A B2B sales team's five-step process: qualify, scope, configure, quote, close. It was being run out of spreadsheets and a Navision instance that wasn't designed for it. - **Reps assembled, didn't sell** — Reps spent their time pulling information together across three tools before they could make a single recommendation. The selling came last. - **Pipeline + AI recommendation** — Foundry automated the flow end-to-end: a live-pipeline dashboard, a structured questionnaire that captures context as deals move, and an AI layer that produces the recommendation from what's entered and what Navision knows. - **One screen, next action** — Reps stop assembling and start acting. The same Navision data that used to require three tools to interpret is now a single screen telling them what to do next. ## Skilled-trades staffing A ~45M DKK staffing organisation placing foreign craftspeople on Danish hospital builds was running the operation on a patchwork of spreadsheets and ad-hoc tools. Foundry delivered an administration system. It is the operational backbone the organisation now runs on, covering placement, time, billing and the document trail that holds it all together. - **Running on spreadsheets** — A ~45M DKK staffing organisation placing foreign craftspeople on Danish hospital builds was running the operation on a patchwork of spreadsheets and ad-hoc tools. - **Ops held together by hand** — Placement, time, billing and the document trail all lived in different places, held together by people remembering to keep them in sync. - **Administration system delivered** — Foundry delivered the administration system the organisation now runs on. It is the operational backbone covering placement, time, billing and the document trail. - **RedactIX layered on compliance** — On top of the same operation, RedactIX keeps compliance-sensitive information out of documents that leave the building. The combination is what made the engagement work. ## LexIX · Legal tech Lawyers spend disproportionate amounts of time reading contracts looking for the bits that aren't standard. LexIX flips it: the user supplies the firm's own contract standards, and the system reads incoming contracts, highlights the deviations, and explains why each one matters. Under the hood it's retrieval-augmented review over the user's own corpus. Their standards become the index, and the system does the matching. - **Reading for deviations by hand** — Lawyers spend disproportionate time reading contracts looking for the bits that aren't standard. It is the 5% that doesn't match what the firm normally signs. - **Hours spent on the 5%** — Most of a contract is template; the value of the review sits in the deviations. But finding them still costs the hours of a senior lawyer. - **RAG over the firm's own standards** — LexIX flips it. The firm supplies its own standards; the system reads incoming contracts, highlights deviations, explains why each one matters. Retrieval-augmented review over the firm's own corpus. - **Drift-aftale, local model next** — Delivered as a drift-aftale at 6.000 DKK/month, moving toward fully local model deployment so sensitive contract data never leaves the firm. ## RedactIX · Compliance & document redaction A staffing operation handles IDs, contracts and time sheets. Those documents include personal and commercially sensitive information that mustn't follow the document downstream. Foundry built RedactIX, an internal Foundry product, and deployed it inside the customer's operation. It reads incoming documents, matches them against a worklist of fields that need to be redacted, and produces a clean, share-able version with the sensitive content removed or replaced. - **Sensitive fields leave the building** — A staffing operation handles IDs, contracts and time sheets that carry personal and commercially sensitive information. It is content that mustn't follow the document downstream. - **Keyword redaction breaks on reality** — Find-and-replace approaches break on scanned PDFs, foreign-language documents and inconsistent layouts. The fields move around; the keyword approach can't find them. - **Field-level redaction engine** — Foundry built RedactIX and deployed it inside the customer's operation. It matches incoming documents against a worklist of fields to redact and produces a clean, share-able version. - **Scans, foreign-language PDFs handled** — Because it works at the field level rather than the keyword level, it copes with the messy reality that breaks simpler redactors: scanned PDFs, foreign-language documents, inconsistent layouts. ## Organisation-wide Claude rollout A mid-sized services organisation wanted to stop running AI as scattered, individual experiments and turn it into something the whole organisation works through. Foundry led the onboarding: Claude rolled out across teams, and the systems the organisation already lives in wired in as connectors. Those are email, calendar, the document store, the CRM and the data warehouse. The scheduling layer was switched on too, so recurring work can be handed to an agent instead of a person. The shape is deliberate: a curated set of approved connectors, a clear pattern for how new schedules and workflows get reviewed, and guardrails that let non-technical teams build automations without IT becoming the bottleneck. - **AI as scattered experiments** — A mid-sized services organisation was running AI as scattered, individual experiments. They were useful in pockets, but never something the whole organisation worked through. - **No organisation-wide lift** — Without a curated platform, every team was reinventing connectors and prompts, and IT was on the way to becoming the bottleneck for every new automation. - **Connectors + scheduling rolled out** — Foundry rolled out Claude across teams and wired in the systems the organisation already lives in: email, calendar, document store, CRM, data warehouse. Scheduling was then switched on so recurring work can be handed to an agent. - **Teams build their own automations** — Within weeks the first automated workflows run on their own: overnight reports, weekly data refreshes, document handoffs. The platform is the rail; what's built on top is increasingly driven by the teams.