The Self-Funding Headcount

What ten weeks of Embedded Intelligence did for one commercial engagement team

How a nine-person team at a global medical technology company recovered more than a full employee's worth of capacity.

By Adria Stapleton, J.D. | Vice President, Research, SteepRock Inc.

August 2026

ROI claims in life sciences technology tend to arrive as round numbers or enticing percentages with no visible math or traceability. This article takes the opposite approach. It walks through ten weeks of measured usage data from a live deployment of SteepRock's Embedded Intelligence (EI) layer at a global medical technology company, separates what was measured from what was assumed, and shows the calculation in full. The client cannot be named here, but the underlying data — usage logs, deliverable counts, completion rates — is theirs, drawn from production deployment between late April and early July 2026.

What Embedded Intelligence Is

Embedded Intelligence is SteepRock's unified intelligence layer: an AI capability that works natively across a client's own data, applicable proprietary and licensed sources, and SteepRock's Opinion Leader Management System (OLMS) and Opinion Leader Analytics (OLA) search engine. Rather than answering questions in isolation, EI performs cross-domain inquiry, analysis, and insight generation — and produces finished work products: customer profiles, congress strategies, advisory-board plans, executive presentations, and intelligence digests, assembled directly from the data already in production.

The value of EI is not measured in answers delivered but in deliverables completed — the research-and-synthesis work that otherwise consumes hours of skilled staff time per item.

Ten Weeks - Measured

Between April 29, 2026 and July 7, 2026, a nine-person commercial team at a top 10 global med tech organization put EI into daily use. The headline facts below are measured from usage logs, not estimated:

Adoption was full and sustained. All nine licensed users were active in every one of the ten weeks. This was not a pilot with a burst of curiosity followed by decay — usage held week over week, embedded in the daily workflow.

The work was substantial. The team completed 143 deliverables in the period, concentrated in the highest-value engagement work: KOL and customer profiling — the single most time-consuming category of manual research — accounted for nearly half of all engagements, followed by congress strategy, executive slides and briefs, and advisory-board planning. Annualized, the run-rate is approximately 756 deliverables per year, or about 2,270 recovered staff hours.

The output was dependable. Roughly 90% of tasks completed successfully (292 successful deliverable runs out of 323 total turns), which means the time saved was realized rather than lost to rework.

Figure 1. 143 deliverables in 10 weeks, by type — KOL/customer profiling dominates. Source: EI conversation records (n = 143).–

The Math, With Assumptions Labeled

The model is simple: measured work volume × time saved per deliverable × fully loaded cost per hour.

Two of those three inputs are measured. Deliverable volume, active users, and completion rate come straight from the usage logs. The other two are modeled — stated management assumptions: hours saved per deliverable (a base case of 3 hours, which is conservative for a full customer profile or a congress plan) and a fully loaded employee cost of $250,000–$300,000 per year, or $120–$144 per hour on a 2,080-hour basis.

Run the model across three scenarios and the reading is straightforward: even the base case clears the fully loaded cost of one employee. A nine-person team, ten weeks in, is already generating approximately one full-time employee of additional capacity. EI is, in effect, a self-funding headcount.

Figure 2. Annualized value under three time-saved scenarios. The shaded band marks the fully loaded cost of one employee ($250–300K); the base case clears it.

How sensitive is the result to those assumptions? The table below shows the annualized value at three hours-saved levels and both ends of the loaded-cost range, so readers can locate the break-even point and the upside for themselves. Against a fully loaded employee cost of $250,000–$300,000, the model breaks even at roughly 2.3–3.3 hours saved per deliverable — below the base case, and well below what a full customer profile or congress plan plausibly saves.

Value @ $120/hr

Hours saved per deliverable

Value @ $144/hr

FTE equivalent

1.5 hrs (below break-even)

Publications, congresses, guidelines, advisory boards

Social media, podcasts, video, online communities

~0.5

3 hrs (base case)

Deep, institutional, slow-building

Fast, broad, conversational

~1.1

6 hrs (upside)

Peer specialists, institutions

Clinicians, trainees, patients, public

~2.2

Sensitivity of annualized value: 756 deliverables/yr × hours saved × loaded hourly cost. FTE equivalent on a 2,080-hour basis. Even the 1.5-hour case — half the base assumption — returns more than half the cost of an employee.

A note on the sample: we acknowledge that a 10-week deployment with nine desk users is a limited window for long-term generalization — seasonality, workload shifts, and broader user heterogeneity may change run-rates over time. That said, the extrapolation is intentionally conservative: we annualized observed deliverable run-rates without inflating the measured hours-saved per deliverable. Presenting measured results alongside clearly labeled modeled assumptions preserves transparency and ensures the extrapolation errs on the side of caution.

Speak with the client

The company featured in this case study has not been named at their request. Readers who would like to hear about the experience firsthand may request a reference: contact SteepRock, and we will ask the client whether they are willing to discuss their deployment of Embedded Intelligence directly.

The overlap matters more than the distinction

Here is the part strategy teams most often get wrong: KOL and DOL are not mutually exclusive categories. They are different modes of influence, and the map of who holds each is constantly shifting. Some of the most influential voices in a therapeutic area are hybrids — guideline authors with large digital followings. Others are pure DOLs with modest publication records but enormous reach among community physicians. Some deeply influential KOLs have no digital footprint at all

Treating these as one list produces blind spots in both directions: engage only traditional KOLs and you miss day-of-data sentiment; chase follower counts and you miss the experts who will actually write the guidelines.

Medical affairs leaders increasingly describe the answer as an integrated influence map — one view that combines digital influence signals with traditional KOL identification, so field teams can prioritize experts who carry weight in both worlds [6, 7].

How life sciences teams identify DOLs

Follower counts are the least informative signal available, not least because congress-period conversation attracts noise: one analysis of ESMO congress activity found heterogeneous sources competing for attention in the same hashtags, including substantial commercial and low-quality content [2]. Rigorous DOL identification weighs signals like these instead:

Relevance. Does this voice consistently discuss the disease area, mechanism, or treatment landscape in question — or is healthcare a minor part of a general feed?
Engagement quality. Are clinicians and researchers responding, sharing, and debating this person's content, or is engagement primarily from the public?
Network position. Is this voice connected to, cited by, or amplified by other recognized experts?
Velocity and timing. Is this person shaping the conversation at key moments — data readouts, congress sessions, label changes — or commenting after consensus forms?
Trajectory. Rising voices matter as much as established ones. DOL influence changes far faster than KOL standing, and today's active fellow with a sharp tweetorial habit may be tomorrow's hybrid expert.

Because these signals live across platforms and change daily, DOL identification at scale is an analytics problem, not a manual research task.

Why the distinction changes engagement strategy

KOLs and DOLs warrant different engagement models. Traditional KOL engagement runs through advisory boards, investigator relationships, congress meetings, and publication planning. DOL engagement centers on scientific exchange and education — making sure accurate, balanced information reaches the digital conversations where sentiment is actually forming — and it carries compliance considerations specific to public digital channels, a topic medical affairs bodies have flagged as a priority as DOL collaboration matures [6].

The practical implication for medical affairs teams is a single stakeholder view that tracks both modes of influence: who publishes, who presents, who posts, who gets amplified — and how each expert's beliefs about the science are evolving across all of those channels at once.

How SteepRock helps

SteepRock approaches this as one problem, not two.

OLA™ (Opinion Leader Analytics) is a proprietary advanced search engine within SteepRock's Embedded Intelligence service, spanning scientific, commercial, regulatory, and digital data domains across life sciences. OLA identifies and profiles both KOLs and DOLs in any disease area, analyzing publications, clinical trials, congress activity, claims and referral data, and social and digital media in real time — surfacing the digital voices traditional KOL lists miss alongside the institutional experts they overlap with.

SteepRock Embedded Intelligence is the unified, agentic intelligence layer that operates natively across those data domains as a single environment. It brings generative AI directly into the workflow — summarizing expert profiles, synthesizing engagement history, drafting insights, and answering natural-language questions across your stakeholder data. For KOL/DOL strategy, that means a team can ask questions no single data source can answer — “Who are the rising DOLs in our therapeutic area, and how has expert sentiment shifted since the congress readout?” — and get a synthesized, evidence-linked answer drawn from publications, congress activity, and digital conversation at once.

OLMS™ (Opinion Leader Management System) then gives medical affairs and commercial teams one compliant system of record for engaging both KOLs and DOLs — connecting external influence signals with internal engagement history, so the influence map and the engagement plan live in the same place.

See how OLA and Embedded Intelligence map KOL and DOL influence in your disease area:

The overlap matters more than the distinction

Here is the part strategy teams most often get wrong: KOL and DOL are not mutually exclusive categories. They are different modes of influence, and the map of who holds each is constantly shifting. Some of the most influential voices in a therapeutic area are hybrids: guideline authors with large digital followings. Others are pure DOLs with modest publication records but enormous reach among community physicians. And some deeply influential KOLs have no digital footprint at all.

Treating these as one list produces blind spots in both directions: engage only traditional KOLs and you miss day-of-data sentiment; chase follower counts and you miss the experts who will actually write the guidelines.

Medical affairs leaders increasingly describe the answer as an integrated influence map, a single view that combines digital influence signals with traditional KOL identification so that field teams can prioritize experts who carry weight in both worlds [6, 7].

How life sciences teams identify DOLs

Follower counts and raw volume are among the least informative signals available. A network analysis of the ESMO 2018 congress conversation illustrates why: commercial accounts made up under 10% of tweeters yet produced roughly 16% of tweets and drew nearly a quarter of all retweets. Even so, they were far less likely than non-commercial voices to be cited or mentioned by their peers [2]. Volume and even retweets can be inflated by sources that carry little genuine influence among clinicians. Rigorous DOL identification weighs signals like these instead:

Relevance. Does this voice consistently discuss the disease area, mechanism, or treatment landscape in question, or is healthcare a minor part of a general feed?
Engagement quality. Are clinicians and researchers responding, sharing, and debating this person's content, or is engagement primarily from the public?
Network position. Is this voice connected to, cited by, or amplified by other recognized experts?
Velocity and timing. Is this person shaping the conversation at key moments such as data readouts, congress sessions, and label changes, or commenting only after consensus forms?
Trajectory. Rising voices matter as much as established ones. DOL influence changes far faster than KOL standing, and today's active fellow with a sharp tweetorial habit may be tomorrow's hybrid expert.

Because these signals live across platforms and change daily, DOL identification at scale is an analytics problem, not a manual research task.

Why the distinction changes engagement strategy

KOLs and DOLs warrant different engagement models. Traditional KOL engagement runs through advisory boards, investigator relationships, congress meetings, and publication planning. DOL engagement centers on scientific exchange and education, ensuring that accurate, balanced information reaches the digital conversations where sentiment is actually forming. It also carries compliance considerations specific to public digital channels, including privacy and data usage, fair market value, pharmacovigilance obligations when listening to public conversation, and the distinction between passive listening and active engagement. Medical affairs bodies have treated this as a priority as DOL collaboration has matured [5, 6].

The practical implication for medical affairs teams is a single stakeholder view that tracks both modes of influence: who publishes, who presents, who posts, who gets amplified, and how each expert's beliefs about science are evolving across all of those channels at once.

How SteepRock helps

SteepRock approaches this as one problem, not two.

OLA®(Opinion Leader Analytics) is a proprietary advanced search engine within SteepRock's Embedded Intelligence service, spanning scientific, commercial, regulatory, and digital data domains across life sciences. OLA identifies and profiles both KOLs and DOLs in any disease area, analyzing publications, clinical trials, congress activity, claims and referral data, and social and digital media in real time. It surfaces the digital voices traditional KOL lists miss alongside the institutional experts they overlap with.

SteepRock Embedded Intelligence is the unified, agentic intelligence layer that operates natively across those data domains as a single environment. It brings generative AI directly into the workflow, summarizing expert profiles, synthesizing engagement history, drafting insights, and answering natural-language questions across your stakeholder data. For KOL and DOL strategy, that means a team can pose a question no single data source can answer, such as which rising DOLs matter in a therapeutic area and how expert sentiment has shifted since a congress readout, and receive a synthesized, evidence-linked answer drawn from publications, congress activity, and digital conversation at once.

OLMS™ (Opinion Leader Management System) then gives medical affairs and commercial teams one compliant system of record for engaging both KOLs and DOLs, connecting external influence signals with internal engagement history so that the influence map and the engagement plan live in the same place.