Case Study

The answer at the point of need — building a virtual assistant for a pharmaceutical sales force

A knowledge-base chatbot built for 280 pharmaceutical sales representatives — giving them instant, approved answers to product and clinical questions right before or during a customer call.

Client
Leading Pharmaceutical Organisation
Sector
Pharmaceutical / Life Sciences
Timeline
Five months
At a glance
280
Medical sales representatives served by the virtual assistant
840+
Content assets created and ingested into the chatbot knowledge base
↓71%
Reduction in time reps spent searching for product information before calls
94%
Rep satisfaction with response accuracy after three months of live use

Medical sales representatives making critical customer calls with incomplete product knowledge.

The organisation's sales force covered six pharmaceutical products — including two recent launches — across 14 therapeutic areas. Reps were expected to answer highly technical questions from healthcare professionals, handle clinical data, counter objections, and stay within tight regulatory limits — often mid-conversation, with no time to look anything up.

The existing tool was a folder-based intranet — updated infrequently, hard to search on a phone, and structured around how the organisation filed documents rather than how a rep needed to find information. If they could not find the answer in 30 seconds, they guessed, escalated to a manager, or said nothing. None of those were good options in a regulated selling environment.

The brief: build a virtual assistant that reps could query mid-preparation, between appointments, or standing in a corridor — and get the right answer in under ten seconds, in plain language, with a source they could cite.

  • All content surfaced by the assistant had to be pre-approved by the medical affairs and regulatory teams — the assistant could not generate responses, only retrieve and surface approved content
  • The assistant had to work seamlessly on mobile devices — the majority of rep queries would come from phones between appointments, not desktops
  • Responses had to include source citations so reps could verify and share the underlying approved materials with HCP customers where appropriate
  • The solution had to integrate within the organisation's existing Microsoft 365 environment — a new standalone platform would not receive IT approval
  • Content across six products existed in inconsistent formats across multiple teams — no single content owner, no unified library

Before mapping questions or creating content, we did a full content landscape audit — because you cannot build a useful assistant on a disorganised knowledge base.

🗂️
Existing Asset Inventory
Catalogued 1,240 documents across six product teams — clinical summaries, detail aids, training decks, objection guides, regulatory references, internal FAQs. Assessed each for accuracy, currency, and whether it was actually suitable for chatbot ingestion.
Regulatory Clearance Review
Worked with medical affairs to classify every document: fully approved, approved with caveats, pending review, or retired. Only cleared content went in — 487 cleared in full, 203 cleared with excerption, the rest excluded.
🕳️
Gap Identification
Cross-referenced the cleared content against the question map built in parallel with the sales team. Found 214 question types with no approved content at all — meaning new content had to be created before the assistant could go live.
The audit finding that defined the project scope
Of the 1,240 documents in the existing library, fewer than 40% were both current and cleared for representative use. The audit revealed that the intranet was not just poorly organised — it was significantly out of date and in many cases contained superseded clinical data. Building the assistant on the existing content library without the audit would have created a tool that confidently gave representatives incorrect information. The audit added three weeks to the project. It saved the project.

Before building answers, we mapped every question a rep would ever need to ask.

The question mapping process was the most critical — and most time-consuming — phase of the project. A virtual assistant is only as useful as its ability to recognise what a user is actually asking. To achieve this, we had to understand the full landscape of questions a sales representative might pose, in the language they would actually use, not in the language of product documentation.

We ran structured question-elicitation sessions with 40 representatives across four regions — asking them to recall every question they had ever been asked by a healthcare professional, every piece of information they had ever struggled to find, and every situation in which they had wished they had a faster answer. Sessions were recorded, transcribed, and analysed for patterns.

Why natural language mattered

Reps do not ask "What is the mechanism of action of Product X?" They ask "How does it work in the body?" or "What happens when a patient takes the first dose?" The chatbot had to understand both — and map each to the same underlying answer. We catalogued over 340 natural language variants across the 580 distinct question types identified.

Product Mechanism & Clinical Profile 112 questions
"How does this work?", "What's the evidence base?", "How does it compare to the NICE guidance?"
Dosing, Administration & Titration 89 questions
"What's the starting dose?", "Can it be used in renal impairment?", "How quickly do patients see results?"
Safety, Side Effects & Contraindications 104 questions
"What are the common adverse effects?", "Are there any black box warnings?", "What about pregnancy?"
Competitive Landscape & Objection Handling 97 questions
"Why should I prescribe this instead of [competitor]?", "The data isn't as strong — how do you respond to that?"
Patient Access, Reimbursement & Formulary 78 questions
"Is it on formulary?", "What does the patient pay?", "Is there a patient support programme?"
Regulatory & Promotional Compliance 100 questions
"Can I share this study off-label?", "What's the approved indication exactly?", "What can I say about the Phase III data?"

Four types of content — each designed for a different kind of question, each structured for optimal chatbot retrieval.

🎬
Product Explainer Videos
Short animated videos (2–4 min) covering mechanism of action, clinical rationale, and patient profile for each product. Produced with professional voiceover and motion graphics. Linked from chatbot responses when a question requires visual explanation rather than text alone.
48 videos produced
📋
Structured FAQ Library
580 approved Q&A pairs — each written to answer the question in the rep's natural language, not in clinical document language. Every answer includes a plain-language response, a supporting data point, and a link to the full approved source document. Medical affairs sign-off on every entry.
580 Q&A pairs
📘
Product Quick Reference Guides
Single-page digital reference cards for each product — covering dosing, key clinical data, approved indications, and top-line safety information in a format reps can skim in under 60 seconds. Available as chatbot-linked PDFs and as in-assistant formatted responses for the most frequently accessed content.
24 product guides
🤝
Objection-Handling Frameworks
Structured objection-handling guides for the 40 most common HCP objections — written as conversation frameworks, not scripts. Each includes the objection, the approved response approach, the supporting evidence, and the regulatory guardrails on what can and cannot be said. Accessible via natural language query or direct topic search.
40 objection frameworks

Microsoft Copilot Studio was selected for its native integration with the organisation's Microsoft 365 environment, its retrieval-augmented generation architecture, and its ability to restrict responses strictly to approved content — a non-negotiable requirement in a regulated pharmaceutical context.

Phase 1 — Foundation
Knowledge Base Configuration
All approved content — Q&A pairs, product guides, objection frameworks, and video links — was structured and ingested into Copilot Studio's knowledge base using a consistent tagging taxonomy: product, topic category, question type, and regulatory clearance level. The taxonomy governed how the assistant retrieved and surfaced content.
Content was ingested as structured SharePoint documents, with metadata tags driving retrieval precision. Untagged or ambiguously tagged content was held back until correctly classified — preventing the assistant from surfacing content it could not reliably categorise.
Phase 2 — Dialogue Design
Conversation Flows & Intent Mapping
The 580 Q&A pairs were mapped to 340+ natural language intents — covering the full range of phrasings a rep might use for the same underlying question. Copilot Studio's natural language understanding engine was trained on product-specific terminology and pharmaceutical sales language, reducing misinterpretation on technical queries.
Fallback flows were designed for queries outside the knowledge base — rather than generating a response, the assistant acknowledged the gap, routed the rep to the relevant product manager, and logged the unanswered query for knowledge base expansion review.
Phase 3 — Compliance Layer
Regulatory Guardrails & Source Citation
Every response was configured to include a source reference — the specific approved document the response was drawn from, with a direct link. The assistant was programmed to surface compliance warnings on sensitive topic categories (off-label queries, comparative claims, safety questions) — reminding reps of what they could and could not say to HCP customers.
Medical affairs retained editorial control via a review dashboard — flagging high-volume queries, reviewing the accuracy of responses, and pushing content updates through a structured approval workflow that bypassed the full IT release cycle for minor content corrections.
Phase 4 — Deployment
Phased Rollout & Feedback Loop
The assistant was deployed first to a pilot cohort of 40 representatives across two regions for a four-week period. Usage data, unanswered query logs, and rep feedback were reviewed weekly. Forty-two new Q&A pairs were added and 18 existing responses were refined before full deployment to all 280 representatives.
The pilot cohort included representatives from both high-performing and average-performing segments — ensuring the assistant was tested against the full range of query types, not just the questions that top performers already knew how to find answers to.
🔒
Why Microsoft Copilot Studio — and not a general-purpose chatbot
The critical requirement in a regulated pharmaceutical environment is retrieval, not generation. General-purpose AI chatbots can generate plausible-sounding answers that are not grounded in approved content — an unacceptable risk when a medical sales representative might share a response with a healthcare professional. Copilot Studio's architecture retrieves from the knowledge base and generates responses grounded strictly in that content. Combined with source citation and fallback routing for out-of-scope queries, this made it the only viable solution for a pharmaceutical deployment without a full custom build.

What made this project genuinely hard — and how each obstacle was navigated.

01
Regulatory sign-off at speed
Medical affairs approval is thorough but slow. With 580 Q&A pairs requiring review, standard sequential approval would have extended the project by months and created a bottleneck that threatened the go-live timeline.
How we resolved itWe introduced a tiered review system — low-risk factual content (dosing, approved indication) through an expedited 48-hour review track; complex or sensitive content (safety, comparative claims) through the full medical affairs review cycle. This reduced average approval time by 60% without compromising rigour on high-risk content.
02
Content fragmentation across product teams
Six product teams had developed content independently over several years — in different formats, different naming conventions, and with different levels of medical affairs oversight. There was no single content owner, no master library, and no version control system.
How we resolved itWe appointed a single content custodian from each product team as the point of contact and version authority. All content was migrated into a unified SharePoint library with standardised naming, metadata, and review date fields. The chatbot content structure effectively became the organisation's first properly governed product knowledge library.
03
Reps asking questions the chatbot could not answer
Even after the question mapping exercise, real-world use during the pilot revealed a long tail of queries the assistant could not match — highly specific clinical edge cases, regional formulary questions, or queries that combined multiple topics in a single sentence.
How we resolved itThe fallback flow was designed to capture, not just decline. Every unanswered query was logged with the exact phrasing used, routed to the relevant product manager for response within 24 hours, and added to the knowledge base if the answer was approvable. Within eight weeks of go-live, the unanswered query rate dropped from 14% to 4%.
04
Rep adoption: convincing the field to trust it
Experienced sales representatives who had spent years building their own product knowledge systems — personal notes, saved emails, informal networks — were initially sceptical. "Why would I trust a chatbot over my own experience?" was a common pilot feedback theme.
How we resolved itWe involved high-credibility representatives in the content validation process — asking them to review Q&A pairs in their therapeutic area for accuracy and practical usefulness. Their names appeared in the assistant's onboarding as "validated by field experts." Peer credibility drove adoption far more effectively than a management mandate would have.
05
Keeping content current after go-live
A pharmaceutical product portfolio does not stand still. New clinical data, updated labels, revised formulary positions, and evolving competitive landscape meant the knowledge base required continuous maintenance — something the initial project budget had not fully accounted for.
How we resolved itWe designed a lightweight quarterly review process with each product team — a structured two-hour session to identify any content requiring update, retirement, or addition. The Copilot Studio management interface allowed approved content updates to be published without IT involvement, reducing the maintenance overhead to a manageable ongoing commitment.
06
Balancing helpfulness with regulatory safety
There was a persistent tension between making the assistant maximally helpful — giving reps rich, detailed answers — and keeping responses within the boundaries of what had been approved for promotional use. Too restrictive and reps would abandon it; too liberal and it created compliance risk.
How we resolved itThe solution was a response architecture that separated the plain-language answer (always concise and safe) from supporting detail (available on request, with a compliance flag where relevant). Reps who needed more depth could request it; the system surfaced the depth with an explicit regulatory context label so reps always understood what they could and could not share with an HCP.
What made this work

The content came first — the technology came second

The most common failure mode for chatbot projects is leading with the technology and retrofitting the content. The assistant's intelligence is entirely a function of the quality, structure, and completeness of its knowledge base. We spent 60% of the project on content — auditing, creating, tagging, and approving — before the Copilot Studio build began in earnest. The technology build took six weeks. The content foundation took fourteen.

This sequencing felt uncomfortable for stakeholders who wanted to see the chatbot working quickly. The pilot results — 94% response accuracy and a 71% reduction in information search time — vindicated the approach comprehensively.

The unexpected outcome

The assistant became the organisation's first single source of truth

A side effect of building the knowledge base was that it forced the organisation to resolve content conflicts that had existed for years unnoticed — cases where two product teams had different approved data points for the same clinical question, or where the intranet contained both a current and a superseded version of a product guide.

The chatbot could not have two answers to the same question. This forced decisions. And those decisions — made by medical affairs and product teams together — produced a governed, versioned, single-source content library that the organisation had needed for years but had never had a compelling reason to build.

The right answer, in ten seconds, wherever in the world a rep needed it.

Within three months of full deployment, the virtual assistant was handling an average of 1,400 queries per week from 280 representatives. The 94% accuracy rating — assessed through a monthly sample review conducted by the medical affairs team — held steady across all six product areas and all query categories.

The operational impact extended beyond individual rep performance. The regional sales managers reported that call preparation time dropped significantly, post-call query escalations fell by over two-thirds, and new representatives reached full product knowledge competency two weeks faster than the prior cohort — because they had a tool that could answer their questions immediately rather than waiting for their manager's availability.

"I used to spend 20 minutes before every important call digging through the intranet trying to find a clinical reference I half-remembered. Now I ask the assistant and I have it in 10 seconds. I have genuinely changed how I prepare for calls because of this tool."

— Senior Medical Sales Representative

94%
Response accuracyMaintained across all six product areas and all query categories after three months of live use
↓71%
Information search timeReduction in time reps spent locating product information before and between calls
↓68%
Post-call escalationsReduction in queries escalated to product managers or medical affairs after customer calls
1,400+
Weekly queries handledAverage weekly query volume across all 280 representatives within 90 days of full deployment

Have a sales force that needs the right answer at the moment it matters most?

Learning matters. Let's build something that travels with your people — and answers when it counts.