← ALL INSIGHTS
CASE STUDY — BotSupply · HEALTHTECH / AI · CONVERSATIONAL AI PLATFORM

BotSupply: Governed AI Assistant for Clinical Workflows.

BotSupply wanted to prove a feature: point their platform at a website's FAQ page and get a working chatbot back. Proving it meant building an automation platform on Microsoft's FAQ bot and solving the data extraction problem underneath — work their own team did not have the domain expertise for, and that agencies would not commit to at proof-of-concept scale.

BY Henry, Engineering Editor | August 5, 2026 · AUGUST 5, 2026
case-study-botsupply-clinical-ai
70%
FASTER CHATBOT BUILD & DEPLOY FOR ORACLE'S CUSTOMERS
FEATURE SHOWCASED · $120K RAISED · DANISH AI IBM AWARD 2017
99% UPTIME AGAINST SLA
AI MODELS TRAINED AND RELEASED WITH ZERO DOWNTIME
PARTNERSHIP RAN OVER 4 YEARS · ACQUIRED BY XTENDOPS USA
01 — THE PROBLEM

A feature that had to be proven, not described

The founders wanted to showcase a capability rather than a roadmap: a customer adds their FAQ page link, and the platform returns a working chatbot. That meant building an automation platform on Microsoft's FAQ bot and solving data extraction from arbitrary FAQ pages. Their options were poor. A single hire could build it slowly. Agencies would not commit to research work and short proofs of concept. The internal team lacked expertise in the data extraction domain. Freelancers could not become the long-term team they wanted alongside them.
02 — RESEARCH FIRST

POCs built to be shown, not filed

Research-led, not assumption-led. We delivered several proofs of concept, built to a standard where they could be presented at tech events across the Nordic region rather than demonstrated internally and shelved. That is a different engineering bar: the extraction has to hold up against FAQ pages nobody curated in advance, and the failure modes have to be understood well enough to stand in front of an audience. The feature was showcased, BotSupply raised $120k, and the work won the Danish AI IBM Award 2017.
03 — THE PLATFORM

A no-code builder on a dockerised backend

The engagement widened from a proof of concept to the complete backend engineering of the platform. A no-code UI let customers build chatbot logic, train it instantly and deploy in minutes without downtime. Pre-built libraries of entities and intents removed the cold start. A generic integration layer carried conversations onto Facebook, Slack, Teams and web without per-channel rework, and the platform supported Oracle ODA, DialogFlow and Rasa as interchangeable NLP engines. It ran dockerised, monitored, and scaled to what each customer needed.
04 — RUNNING IT

Uptime as a contract, not an aspiration

Notification and monitoring components were built to hold 99% uptime against the SLA, and state-of-the-art AI models were trained and released with zero downtime — a harder requirement than it sounds when the model is the product. The partnership ran for over four years across infrastructure and AI, with 24x7 support. It carried the company through funding difficulty and ultimately through the acquisition by XtendOps USA. Oracle's customers used the result to build and deploy chatbots 70% faster.

Need a capability proven
before you can sell it?

Talk to the teamAgentic AI service →
FAQ

Questions about this engagement

Why did agencies and freelancers not fit? The work began as research and short proofs of concept, which agencies would not commit to. The internal team lacked expertise in the data extraction domain the feature depended on. Freelancers could deliver a piece but not become the long-term team working alongside them, which is what the founders actually wanted. The requirement was a dedicated team that consults as well as delivers and behaves like an in-house partner.
What made the proofs of concept different? They were built to be presented at tech events in the Nordic region, not demonstrated internally. Extraction had to work against FAQ pages nobody had curated in advance, and the failure modes had to be understood well enough to stand behind in public. That standard is what turned a feature demonstration into $120k raised and the Danish AI IBM Award 2017.
What does the platform actually do? It lets a customer build chatbot logic through a no-code UI, train it instantly and deploy in minutes without downtime. Pre-built libraries of entities and intents remove the cold start. A generic integration layer carries conversations to Facebook, Slack, Teams and web without per-channel work, and Oracle ODA, DialogFlow and Rasa are supported as interchangeable NLP engines behind it. The platform is dockerised, monitored and scaled per customer.
How do you release AI models with zero downtime? By separating model release from platform availability and building the notification and monitoring components that make the SLA measurable rather than assumed. The platform held 99% uptime against its SLA while state-of-the-art models were trained and released behind it. When the model is the product, a release window that takes the service down is a commercial problem, not just an operational one.
How long did the engagement run? Over four years, spanning infrastructure through AI innovation, with 24x7 support. It began as a proof of concept, widened into complete backend engineering of the platform, carried the company through a period of funding difficulty, and ended with the acquisition by XtendOps USA.