Governance-first AI, in the real world.

The challenge

Useful market intelligence is buried across dozens of fragmented sources, and most monitoring tools drown teams in hype rather than surfacing the few signals that actually matter for a decision.

What we did

We built an automated radar that ingests dozens of public sources, including news, hiring, company filings, events and regulatory feeds, then applies a weighted signal framework to score organisations on genuine intent rather than noise.

What we built

A two stage language model pipeline does the work. A low cost pass extracts and tags every item, then a higher quality pass writes concise, decision ready briefs. Entity resolution merges duplicates and public company data confirms size and fit. It runs weekly on public or permissioned data only, with scoring that stays explainable.

The challenge

Regional beverage teams need to track category shifts, competitor activity, launches, regulation and route-to-market movement across fragmented MEA sources. Global reporting often misses or flattens these signals, while manual scanning is inconsistent and time-consuming.

What we did

We configured the TrendRadar pattern for MEA beverage intelligence, combining regional news, beverage trade sources, marketing publications and targeted country, category and competitor queries. The system filters, deduplicates and classifies recent articles against MEA-specific signal categories before generating a sourced executive brief.

What we built

An n8n-based automation runs every two weeks, reading 49 regional, category and competitor feeds. A two-pass AI workflow tags and classifies articles, then generates a structured HTML report with an executive summary, trend deep-dives, competitor watch and country spotlight, delivered by email to the regional team.

The challenge

Global insights teams need a reliable read on category, competitor and market movement across fragmented trade sources. Manual scanning is time-consuming, inconsistent and difficult to scale, while generic reports often arrive too late or lack source-level evidence.

What we did

We configured the TrendRadar pattern for global beverage intelligence, combining industry publications and targeted news queries into a monthly monitoring workflow. The system filters, deduplicates and classifies recent articles against predefined beverage trend categories before generating a sourced executive brief.

What we built

An n8n-based automation runs monthly, reading 45 global beverage feeds and targeted news queries. A two-pass AI workflow tags and classifies articles, then generates a structured HTML report with an executive summary, radar overview, trend deep-dives, competitive moves and linked source evidence. Failure alerts notify AI Tappers if a run needs investigation.

The challenge

Corporate development and strategy teams need to spot acquisition-relevant brands before they become obvious. The signals are fragmented across funding news, distribution moves, facility changes, product launches, leadership changes and category press. Manual monitoring is inconsistent, and generic market reports rarely connect these signals into a target-level view.

What we did

We configured a dedicated radar for beverage M&A intelligence, combining European business sources, global beverage publications, M&A feeds and targeted news queries. Instead of summarising trends, the system scores individual brands against acquisition-relevant signals and filters out large corporations to focus attention on emerging or mid-size targets.

What we built

An n8n-based workflow reads 57 public feeds, deduplicates recent articles and uses a multi-stage AI pipeline to score brands against a 125-point acquisition signal framework. Brands above the alert threshold are passed through a secondary verification step before a structured brand alert report is generated and delivered by email.

The challenge

Corporate communications teams needed to produce frequent internal and external content in English and Dutch without losing brand consistency, factual accuracy or review discipline. Drafting, rewriting and translating repeat communications manually created delay, while generic AI tools risked unsupported claims, inconsistent tone and weak governance.

What we did

We configured CORA inside the client's Microsoft 365 environment, grounding it in an approved SharePoint knowledge base and defining clear communication tasks, tone guidance and review boundaries. CORA supports drafting, rewriting, translation and QA across common communications formats, while keeping final approval with the human owner.

What we built

A Microsoft 365 Copilot Agent Builder assistant that uses specified SharePoint sources before generating content. CORA supports English and Dutch communications, with pre-built prompts for press releases, LinkedIn posts, internal memos, complaint replies, sponsorship declines and brand-voice rewrites. The configuration is portable plain text, with no external APIs, custom code or third-party integrations.

The challenge

Marketing and innovation teams often develop ideas in rooms full of category knowledge, brand optimism and internal assumptions. The harder question is whether a time-poor, low-loyalty consumer would notice, understand or act on the idea before production or media spend is committed.

What we did

We configured FLOID around Swinkels' own sceptical consumer persona, grounding it in behavioural science, category thinking and brand context. The assistant reviews slogans, campaign ideas, product concepts and visual prompts, then returns structured feedback designed to explore the idea, challenge the assumption and push a practical refinement.

What we built

A Microsoft 365 Copilot Agent Builder assistant inside the client's tenant, using Swinkels' persona material, behavioural science documents and decision-tree logic as its knowledge base. FLOID gives marketing teams a fast first challenge on slogans, campaigns and product ideas, with responses structured around explore, challenge and refine prompts.

The challenge

Royal Swinkels had real momentum with AI across the business, but adoption was beginning to fragment across teams, tools and licensing models. Leadership needed a governed way to scale what was working, protect data and ownership, and avoid turning early enthusiasm into long-term vendor dependency.

What we did

Discovery mapped where AI could create value across functions and where governance, ownership and data boundaries mattered most. We then shaped a deployment model that separated contained MVPs from future industrialisation, allowing teams to prove value quickly while preserving portability and client ownership.

What it produced

The work produced a governed portfolio of AI systems: CORA for corporate communications, FLOID for marketing challenge and refinement, TrendRadar for global and regional market intelligence, and M&A Radar for acquisition signal monitoring. Microsoft-native assistants operate inside the Swinkels tenant, while radar systems use public-source automation with portable workflows and no hidden data persistence.

The challenge

A global dry bulk operator had complex, distributed operations and growing interest in AI, but no shared evidence base for where to start. Leadership needed to identify where AI could reduce manual information work without disrupting safety-critical workflows, weakening accountability or introducing new operational risk.

What we did

We ran a structured Discovery across operations, commercial and technical stakeholders, combining interviews with role-tailored questionnaires. We mapped current workflows, surfaced pain points, and scored candidate use cases against value, feasibility, operational risk and governance readiness.

What it produced

Discovery produced a prioritised AI roadmap and technical report leadership could act on. The recommended first build was a governed Knowledge Assistant over the company's own operational knowledge: designed for Azure, bounded by human oversight, and structured around explicit ownership before wider deployment.

The challenge

A long-established, family-run manufacturer had growing AI activity across the business, but little coordination. Teams were trialling assistants site by site, sensitive documents were entering consumer AI tools, and leadership lacked a clear view of value, risk or return. With four largely autonomous sites, uneven processes and operational data spread across ageing systems, the business needed an evidence-led view of where AI could realistically help before committing further investment.

What we did

We ran a four-week Discovery across sales, factory general managers and IT, combining one-to-one interviews, role-tailored questionnaires and a baseline team survey. Rather than asking teams to invent AI use cases, we mapped real workflows and sub-processes, then assessed them through a governance lens covering shadow AI exposure, data risk, GDPR and EU AI Act alignment.

What it produced

Discovery produced an AI opportunities playbook: a prioritised view of where AI could genuinely move the needle, two to three quick-win pilots, and a phased roadmap that starts with one site before wider rollout. The recommendation was to build internal capability, governance and adoption discipline first, rather than buy more tools.

The challenge

A facilities and property management group was running complex cost allocation, maintenance, claims, statutory compliance, tenant reporting and field operations through spreadsheets and disconnected tools. Leadership wanted to understand where AI could reduce manual effort without disrupting the transparent open-book model their tenant relationships depend on. Part of the team worked outside English, so Discovery had to reflect the language of the operation.

What we did

We ran a compressed Discovery using role-tailored questionnaires and interviews in the client's own language. We mapped the operational workflows behind cost allocation, supplier invoices, claims, inspections, annual reconciliation, tenant reporting and field activity, then assessed where AI could reduce effort without weakening transparency, auditability or control.

What it produced

Discovery produced an AI Capability Catalogue: one Foundation layer and nine independently buildable modules, each separately priced, dependency-mapped and sequenced. The catalogue also gave leadership recommended packages and monthly run-rate options, so they could choose what to build, when to build it and how much to commit at each stage.

The challenge

RoadMender was scaling quickly, but operational knowledge lived inside people, spreadsheets and disconnected systems. Leadership wanted to grow machine deployment and council reach without adding proportional operational load, while improving visibility across commercial activity, factory workflows, field operations and sales.

What we did

We ran Discovery interviews with leadership and operational stakeholders, mapping where manual reporting, individual knowledge dependency and disconnected workflows were creating scale risk. We then prioritised AI and automation opportunities across commercial operations, customer intelligence, machine monitoring, field support and leadership reporting.

What it produced

Discovery produced a 90-day operational acceleration roadmap, sequencing quick wins, core systems and longer-term intelligence capabilities. The roadmap covered council intelligence, customer reorder alerts, CRM and dashboard foundations, delivery note and invoice automation, machine activity monitoring, fitter support and tender intelligence.