The two approaches, with the platform agent and SteepRock Embedded Intelligence™ compared within approach 1
Approach 1: the vendor's AI over the vendor's data
Agentic AI tools sold with life sciences platforms go by many names: generative AI for medical affairs, AI insights for MSLs, AI analytics platforms for commercial teams. Most are built the same way. A large language model from one of a few commercial providers does the reasoning. The tool pulls records from the platform's data store and puts them in the prompt, so the model answers from that data rather than from memory. It can also call functions (search a database, read a document, write a record), and an orchestration layer decides which to call and in what order. A permission layer limits what retrieval can see. Around the model sit the governance controls: masking of sensitive fields, agreements that the provider won't keep or train on the prompts, and logs of what was retrieved and answered.
SteepRock Embedded Intelligence™ is built the same way. The differences are in what the agent can reach, the rules it follows and what it hands back.
The platform agent
A platform agent's strength is that it lives where your team already works. The account summary or pre-call briefing shows up in the application a rep or MSL opens every morning, and the agent can update a record right there. Its weakness is the other side of the same coin. It sees only what has been entered into the platform, or bought and loaded into it. The controls are the vendor's. And what comes back is a summary, a recommendation or a record update, which someone on your team still has to turn into a plan. A result is different from a plan. SteepRock Embedded Intelligence™ returns the result.
Approach 1, as we build it: SteepRock Embedded Intelligence™
Here is how SteepRock Embedded Intelligence™ is built, starting with how you reach it and ending with how its work is checked.
Delivery and access
SteepRock Embedded Intelligence™, SteepRock's life sciences AI platform, is delivered through our Opinion Leader Management System, OLMS™, and works across our Opinion Leader Analytics data layer, OLA®. OLMS™ handles identity, sign-in and role-based access. SteepRock Embedded Intelligence™ has no sign-in of its own: you reach it from an authenticated OLMS™ session. Each client gets its own single-tenant OLMS™ application and database, and our API is available to clients who contract for it.
The data it reaches
OLA® is the data layer we compile and maintain for AI-powered KOL management: KOL identification, KOL profiling and KOL mapping, and more, including healthcare organizations and payers. The data inputs behind OLA® are proprietary to SteepRock. Representative categories are shown below. Searches run in real time against evidence-based criteria. They surface key opinion leaders and emerging experts, and they reach every kind of provider, from physicians to nurse practitioners and pharmacists, down to the community level in the United States.
The two approaches, with the platform agent and SteepRock Embedded Intelligence™ compared within approach 1
Approach 1: the vendor's AI over the vendor's data
Agentic AI tools sold with life sciences platforms go by many names: generative AI for medical affairs, AI insights for MSLs, AI analytics platforms for commercial teams. Most are built the same way. A large language model from one of a few commercial providers does the reasoning. The tool pulls records from the platform's data store and puts them in the prompt, so the model answers from that data rather than from memory. It can also call functions (search a database, read a document, write a record), and an orchestration layer decides which to call and in what order. A permission layer limits what retrieval can see. Around the model sit the governance controls: masking of sensitive fields, agreements that the provider won't keep or train on the prompts, and logs of what was retrieved and answered.
SteepRock Embedded Intelligence™ is built the same way. The differences are in what the agent can reach, the rules it follows and what it hands back.
The platform agent
A platform agent's strength is that it lives where your team already works. The account summary or pre-call briefing shows up in the application a rep or MSL opens every morning, and the agent can update a record right there. Its weakness is the other side of the same coin. It sees only what has been entered into the platform, or bought and loaded into it. The controls are the vendor's. And what comes back is a summary, a recommendation or a record update, which someone on your team still has to turn into a plan. A result is different from a plan. SteepRock Embedded Intelligence™ returns the result.
Approach 1, as we build it: SteepRock Embedded Intelligence™
Here is how SteepRock Embedded Intelligence™ is built, starting with how you reach it and ending with how its work is checked.
Delivery and access
SteepRock Embedded Intelligence™, SteepRock's life sciences AI platform, is delivered through our Opinion Leader Management System, OLMS™, and works across our Opinion Leader Analytics data layer, OLA®. OLMS™ handles identity, sign-in and role-based access. SteepRock Embedded Intelligence™ has no sign-in of its own: you reach it from an authenticated OLMS™ session. Each client gets its own single-tenant OLMS™ application and database, and our API is available to clients who contract for it.
The data it reaches
OLA® is the data layer we compile and maintain for AI-powered KOL management: KOL identification, KOL profiling and KOL mapping, and more, including healthcare organizations and payers. The data inputs behind OLA® are proprietary to SteepRock. Representative categories are shown below. Searches run in real time against evidence-based criteria. They surface key opinion leaders and emerging experts, and they reach every kind of provider, from physicians to nurse practitioners and pharmacists, down to the community level in the United States.
The two approaches, with the platform agent and SteepRock Embedded Intelligence™ compared within approach 1
Approach 1: the vendor's AI over the vendor's data
Agentic AI tools sold with life sciences platforms go by many names: generative AI for medical affairs, AI insights for MSLs, AI analytics platforms for commercial teams. Most are built the same way. A large language model from one of a few commercial providers does the reasoning. The tool pulls records from the platform's data store and puts them in the prompt, so the model answers from that data rather than from memory. It can also call functions (search a database, read a document, write a record), and an orchestration layer decides which to call and in what order. A permission layer limits what retrieval can see. Around the model sit the governance controls: masking of sensitive fields, agreements that the provider won't keep or train on the prompts, and logs of what was retrieved and answered.
SteepRock Embedded Intelligence™ is built the same way. The differences are in what the agent can reach, the rules it follows and what it hands back.
The platform agent
A platform agent's strength is that it lives where your team already works. The account summary or pre-call briefing shows up in the application a rep or MSL opens every morning, and the agent can update a record right there. Its weakness is the other side of the same coin. It sees only what has been entered into the platform, or bought and loaded into it. The controls are the vendor's. And what comes back is a summary, a recommendation or a record update, which someone on your team still has to turn into a plan. A result is different from a plan. SteepRock Embedded Intelligence™ returns the result.
Approach 1, as we build it: SteepRock Embedded Intelligence™
Here is how SteepRock Embedded Intelligence™ is built, starting with how you reach it and ending with how its work is checked.
Delivery and access
SteepRock Embedded Intelligence™, SteepRock's life sciences AI platform, is delivered through our Opinion Leader Management System, OLMS™, and works across our Opinion Leader Analytics data layer, OLA®. OLMS™ handles identity, sign-in and role-based access. SteepRock Embedded Intelligence™ has no sign-in of its own: you reach it from an authenticated OLMS™ session. Each client gets its own single-tenant OLMS™ application and database, and our API is available to clients who contract for it.
The data it reaches
OLA® is the data layer we compile and maintain for AI-powered KOL management: KOL identification, KOL profiling and KOL mapping, and more, including healthcare organizations and payers. The data inputs behind OLA® are proprietary to SteepRock. Representative categories are shown below. Searches run in real time against evidence-based criteria. They surface key opinion leaders and emerging experts, and they reach every kind of provider, from physicians to nurse practitioners and pharmacists, down to the community level in the United States.
Your own OLMS™ records sit beside OLA®: the interactions, MSL insights and medical information requests your team has logged; the HCP, KOL, digital opinion leader, organization and payer profiles you have access to; the engagement plans built on them; and the CRM call records your team imports into OLMS™. You ask one question in plain language, and SteepRock Embedded Intelligence™ answers across all of it.
Permissions
SteepRock Embedded Intelligence™ queries OLMS™ as you, inside your own session, so it inherits your role, territory and attribute permissions. There is no second permission model to maintain. Client administrators can't read other users' conversations.
Models and your own model account
The reasoning in SteepRock Embedded Intelligence™ is done by Anthropic's Claude models, under enterprise API terms. By default, they run under our provider account. No client data, system output or confidential information is used to train, validate or test any AI model. If you prefer, they can run under your own Anthropic, AWS Bedrock or Google Vertex account, so your own agreement with that provider governs the model processing. Either way, the source references, the rulebook and your users' permissions work the same. Running under your own account is still approach 1: our agent and our rulebook, with the same Claude models under your cloud account. It is separate from the MCP Gateway, which belongs only to approach 2.
Verification
AI hallucination is the first thing a life sciences compliance team asks about, and verifying AI outputs in pharma takes more than a disclaimer. When SteepRock Embedded Intelligence™ builds a deliverable from records in the system, it shows references to those records, and you can open them. A written rulebook governs every deliverable: don't invent values, label each claim as fact or inference, and say so when the data isn't there. In plain terms, it is instructed not to make up data, and the references let you check that it didn't. We screen user input for prompt-injection patterns. We don't filter the output after the fact; you review it before you use it. In a real-world test, the AI refused to give a number it couldn't support even when it would have been expedient to do so. We published the test, with the prompts and references, so you can check it yourself [1].
What you get back
SteepRock Embedded Intelligence™ brings AI analytics for medical affairs and commercial teams into one workspace: plain-language questions, closed-loop analysis, automated insights, summaries, tables and charts. From the conversation you turn results into the deliverables your team uses: KOL profiles, territory models, congress plans, HCP engagement plans, referral pathway analyses and executive presentations, exported as PowerPoint, Excel or PDF. Every one of these is a result, and together they are what we mean by Results as a Service.
For example, an MSL might ask: "Which cardiologists in my territory chaired a session at ACC this year and are running an open heart failure trial?" What comes back is a ranked list with the session and the trial cited for each name, each linked to its source record and ready to export to Excel.
What you get back
SteepRock Embedded Intelligence™ brings AI analytics for medical affairs and commercial teams into one workspace: plain-language questions, closed-loop analysis, automated insights, summaries, tables and charts. From the conversation you turn results into the deliverables your team uses: KOL profiles, territory models, congress plans, HCP engagement plans, referral pathway analyses and executive presentations, exported as PowerPoint, Excel or PDF. Every one of these is a result, and together they are what we mean by Results as a Service.
For example, an MSL might ask: "Which cardiologists in my territory chaired a session at ACC this year and are running an open heart failure trial?" What comes back is a ranked list with the session and the trial cited for each name, each linked to its source record and ready to export to Excel.
What this means to you
This is the agent for a team that wants the work done. You ask once, in plain language, and get back the profile, territory model or congress plan itself, built from the research, trial, grant, regulatory, claims and congress sources pharma teams work from, plus your own records, with references to every one, and built under the rulebook. Because your own interaction records are part of the reasoning, the deliverable reflects what your team already knows as well as what the field shows. Teams under contract have used it for launch brand plans and territory models, congress strategies, hospital referral pathway analyses, influence maps, KOL profiles, monthly monitoring reports and dashboards.
Approach 2: your AI over the vendor's data
Here you keep the AI assistant your employees already use and give it the vendor's data. Some platforms now offer a version of this, exposing their records or agent actions to your assistant on their terms. We offer it for our data through the MCP Gateway.
Your assistant connects to the gateway as a governed source, and the gateway connects to OLMS™ and OLA® as the user who is asking. Every request is read-only and runs under that user's OLMS™ permissions, so the assistant gets only the records that person is allowed to see, each with a link back to the source record in your application. Nothing is exported in bulk; each request returns only what it asked for. Your assistant does its own reasoning on those records. Our rulebook and guardrails stay inside SteepRock Embedded Intelligence™ and don't travel with the data, so the rules that govern the answer are yours.
What this means to you
This is for a company that has standardized on one AI and wants its life sciences data inside it. You can run it alongside SteepRock Embedded Intelligence™: our agent builds the finished work, and your own assistant draws on the same data in the same conversation as everything else it knows. We govern access: every request runs as the user who asked, under that user's OLMS™ permissions, and is logged and revocable. Once the records reach your assistant, your own AI policies and your agreement with your AI provider govern them. What you give up is our rulebook. The references come with the records, but your assistant's answers are only as disciplined as the rules you give it.
Side by side
The table sets the three side by side.
Within approach 1, the two agents differ in what they reach, the rules they follow and what they return. A platform agent helps with the work inside its application; SteepRock Embedded Intelligence™ hands the work back finished. Across the two approaches, the choice is yours: our agent producing finished work, your own AI on our data under your rules, or both at once. For our agent, you also choose whether the models run under our account or yours.
Results as a Service
We've helped life sciences manufacturers identify, profile and engage KOLs, DOLs, HCPs, HCOs and payers for 25 years, and since 2025 we've offered Results as a Service. In practice that means what SteepRock Embedded Intelligence™ returns: the finished deliverable, assembled from OLA® and your OLMS™ records, referenced to its sources, built under the rulebook and exported as PowerPoint, Excel or PDF. We can also produce defined deliverables for a team, generated in OLA®, reviewed with the team and tagged in its OLMS™.
A result, in our sense, is a finished and referenced deliverable that your team checks before using. It is not a promise of a business outcome. And because the MCP Gateway returns records to your own AI, which writes its own answer, Results as a Service applies to SteepRock Embedded Intelligence™ rather than the gateway. Both are available now.
Results as a Service
The finished, referenced deliverable, checked by your team before you use it.
Jason Cater is Senior Vice President, Technology, at SteepRock, Inc. SteepRock Embedded Intelligence™ is a trademark of SteepRock, Inc. OLA® is a registered trademark of SteepRock, Inc.
References
1. SteepRock, Inc., “When the Number Is Wrong, the AI Should Say So,” 2 September 2026, steeprockinc.com/when-the-number-is-wrong-the-ai-should-say-so.
