Microsoft detailed the deployment in a September 15 customer story, portraying Ask Jack as a source-grounded conversational layer around Jack Fertility’s mail-in sperm testing service. The agent is intended to explain concepts and help users interpret the vocabulary around their results, rather than issue diagnoses or recommend treatment. Jack Fertility’s own commercial site currently markets a £169 postal test and promotes unlimited questions to its “expert-trained AI.”
This is a useful Copilot Studio case study precisely because it is not a generic corporate FAQ bot. It puts a low-code agent in front of sensitive health questions, where the cost of an overconfident response is higher and where a superficial “responsible AI” checkbox would be inadequate. Microsoft’s account also reveals a significant operational change still ahead: the public version is paused during a site refresh, while Jack Fertility plans to move the agent into a secure post-test portal that could use a customer’s actual results.
That proposed portal integration is where the project stops being a relatively bounded education chatbot and becomes a system that demands far tighter data governance, testing, and clinical oversight.
Ask Jack’s reported speed came from changing the architecture
According to Microsoft’s customer story, Jack Fertility chose Copilot Studio after considering several approaches, including work through Google for Startups. Co-founder Lily Elsner said the startup avoided the need to build and maintain a traditional intent-and-dialogue-tree chatbot, claiming ongoing upkeep fell from two to three hours per week to one or two hours per month.
The basic technical idea is familiar: rather than mapping every expected wording of a question to a scripted answer, the agent retrieves information from a selected set of sources and generates a response. Microsoft says Ask Jack is grounded in curated Jack Fertility material, NHS information, World Health Organization material, and peer-reviewed research chosen with medical input. It says the agent supplies references with responses and directs users to a healthcare professional when approved sources do not support an answer or clinical judgment is needed.
For a small company, that is the genuine attraction of a managed agent platform. A team can spend less time on language parsing, bot hosting, orchestration plumbing, and manual response trees, and more time deciding the questions its agent must refuse, the evidence it can use, and when it must hand off. The speed claim is Jack Fertility’s estimate, not an independently audited cost comparison, but its description of the work is credible: configuration may be quick; safe deployment is still a testing project.
Microsoft’s own Copilot Studio documentation makes the same distinction indirectly. Knowledge sources can ground an agent in approved documents, websites, databases, and internal data rather than letting it rely only on general training. That does not validate the content automatically. Someone still has to select, review, update, and monitor the source material — and determine what the agent should do when retrieved text cannot safely answer the user’s question.
Medical boundaries are the product requirement
Microsoft says Jack Fertility ran user sessions that included scientific questions, slang, humour, sensitive wording, and different conversational styles. It also says the team specifically tuned the agent around reproductive-health terminology, which may be caught incorrectly by broad content-safety filters despite being necessary to answer a legitimate health question.
This is a substantive implementation problem, not a cosmetic one. Over-blocking can render a sexual-health agent evasive or useless; under-blocking can let it become too casual, too specific, or falsely reassuring. Jack Fertility says its design target is an approachable agent that explains evidence and prepares customers for conversations with clinicians, rather than a substitute clinician.
Microsoft explicitly notes in its own compliance documentation that Copilot Studio is not intended for use as a medical device, even though the service supports health-data scenarios and is covered under its HIPAA Business Associate Agreement in appropriate US deployments. In the United Kingdom, the Medicines and Healthcare products Regulatory Agency says that many health-related software and AI products can fall within medical-device or in-vitro diagnostic regulation when their intended purpose meets a clinical need.
The dividing line will depend on what Ask Jack actually does. An agent that explains the difference between sperm count and concentration from approved educational material occupies a different role from software that evaluates a particular person’s data and steers a care decision. Microsoft says Jack Fertility intends to preserve the agent’s educational scope and safety boundaries when it reaches the post-test portal. That intention is important, but it is not the same thing as demonstrating how the system will behave in production with individual results in context.
The portal plan changes the governance checklist
The public web agent was available as a pilot, Microsoft says, and Jack Fertility observed the heaviest use late at night. The company says the agent does not store chat history, a design choice that users reportedly welcomed. Yet the next proposed version could be connected to a secure portal and use an individual’s test results instead of hypothetical examples.
That transition introduces a different set of controls. A chatbot answering a public educational question can often be deployed without user authentication. A result-aware agent should be designed around authenticated identity, least-privilege access to the relevant data, strict boundaries between users, durable audit trails, and a tested process for removing or correcting information where applicable. The system also needs a clear answer to a deceptively simple question: is the user asking for general education, or is the agent using a protected result to personalize an explanation?
Copilot Studio has controls that can help with this work, including Microsoft Entra authentication, data-loss-prevention policies, restrictions on allowed knowledge sources and publishing channels, auditing through Microsoft Purview, and customer-managed encryption keys. But the platform’s capabilities do not configure themselves. Microsoft’s documentation notes that makers can choose unauthenticated access for agents shared by link, and that administrators must explicitly use data policies when they want to block that route.
For a health-data implementation, the practical minimum should include:
- The result-aware agent should require authenticated portal access and should not be exposed through an unauthenticated public chat channel.
- The agent should retrieve only the current user’s authorized data and should not be able to query broad result repositories through loosely governed connectors.
- The knowledge base should be versioned, clinically reviewed, and monitored for stale guidance rather than updated only through informal prompt edits.
- Test cases should include harmful reassurance, unsupported causal claims, prompt injection attempts, ambiguous symptoms, emergency language, and questions that seek treatment recommendations.
- Escalation language should be specific enough to send users to an appropriate clinician or urgent service when the conversation crosses an educational boundary.
The last point warrants emphasis. “Talk to a healthcare professional” is a necessary fallback, but it can become an empty one if an agent does not distinguish routine follow-up from potentially urgent symptoms, or if it has no tested way to recognize when it should stop explaining and redirect.
Jack’s test claims need clearer terminology
The company’s consumer material adds another reason to keep Ask Jack’s language tightly controlled: Jack Fertility’s own pages describe the service in inconsistent terms.
Its product page says the tests “are not clinical semen analyses,” do not provide a medical diagnosis or medical advice, and that some results are estimates. Its science page is more explicit, saying a fresh clinical semen analysis is performed within an hour of ejaculation and that a delayed postal sample is different; it describes the service as a “Sperm Test, not a semen analysis.” In particular, motility is estimated using measured motility, transit timing, validation data, and the company’s correction model.
Yet Jack Fertility’s terms of service call the standard product a “full semen analysis.” The same terms also say results may differ from past or future semen analyses, advise consumers that results are wellness information rather than diagnosis, and say the kits are registered general IVD medical devices with the MHRA.
The underlying limitations may be responsibly disclosed elsewhere, but the terminology should be reconciled before a conversational agent begins explaining results at scale. A user asking whether a result is “clinical” or whether a postal motility estimate is directly comparable to a fresh laboratory result deserves a precise answer. An agent should not smooth over an ambiguity in the source material simply because a more reassuring phrasing is easier to market.
Microsoft’s story says Ask Jack will say when a question cannot be supported by approved sources. In this deployment, that discipline must also apply when the approved sources conflict with each other. Retrieval grounding is only as reliable as the material being retrieved.
A customer story, not independent outcome evidence
Microsoft’s account reports that Ask Jack saved several hours of medical-adviser time each week and removed the need to hire a full-time person to answer inquiries. Those are company-reported operational benefits; Microsoft does not publish usage totals, response-accuracy results, rates of escalation, adverse-event data, or an independent assessment of the agent’s clinical safety.
The article does reveal one operational reality absent from the promotional framing: Ask Jack is not presently a continuously available public service. Microsoft says the public-site pilot has been paused while Jack Fertility refreshes its website, although the company plans to return it to the site as both a customer-acquisition tool and a preview of the portal experience. Jack Fertility’s current product page still advertises AI questions as part of the offering, so prospective customers should verify availability rather than assume the web agent is live.
For Copilot Studio administrators, Jack Fertility offers a compact lesson in what low-code agents are good at: they can remove a large amount of traditional chatbot engineering from a small team’s workload. In a health-adjacent service, though, that saved engineering effort has to be reinvested in source control, access controls, evaluation, clear product language, and escalation design.
The next release that matters is not the return of a polished chat box to Jack Fertility’s homepage. It is whether the company can demonstrate that its result-aware portal agent remains strictly educational, accurately distinguishes measured values from estimates, and protects the personal health information that gives the conversation its value.