Customers/Agios Pharmaceuticals
×

Agios ran the biggest launch in company history on the team it already had.

A launch is a multiplication problem for a lean analytics team: more stakeholders, more questions, and a same-day clock replacing a comfortable one. Agios saw it coming a year out. Rather than add headcount, the commercial analytics team put a governed virtual analyst on the warehouse it already trusted — live in weeks, and now running pre-call planning, market mapping, and root-cause analysis for the whole commercial organization.

2/3 Less
Dashboard Build & Maintenance

Static Excel and PowerPoint reporting gave way to live apps and dashboards that can be changed in natural language.

Months → Days
Work That Used To Queue

And days to hours for ad hoc questions — answered live in the session instead of over rounds of vendor back-and-forth.

+0
Headcount Added

The same handful of analysts covered market access, field, and leadership through the launch and past it.

The Story

The queue was survivable pre-launch. Then the math changed.

Before
  • Answers came back on a vendor's clock. Ad hoc questions routed through offshore vendors — overnight to weeks per answer, plus rounds of back-and-forth before anything reached the person who asked.
  • Dashboards moved through a multi-vendor chain. Mockups in one place, the build in another, data modeling in between. New dashboards took weeks; enhancements landed every few months.
  • Every request ran the same manual pipeline. Gather the request, run it manually, slice and dice, send it back. A lean team serving marketing, market access, sales, and leadership — and a launch about to multiply the volume against fixed headcount.
  • Horizontal AI couldn't carry commercial numbers. ChatGPT, Claude, and Copilot earn their keep on email and slides, where some variance is fine. When someone acts on the number, the same question has to return the same right answer every time.
With Tellius
  • A governed virtual analyst on the existing warehouse. The warehouse stayed the single source of truth. Tellius sits on top as the semantic, reasoning, and agentic worker layer, so every answer runs on definitions the business already trusts.
  • Chosen on consistency, not demo polish. Agios scored vertical AI platforms on semantic-layer strength, reasoning depth, speed, and whether answers held up after five follow-ups. Tellius ranked highest.
  • Validated metric by metric before anyone logged in. Business rules and KPI definitions were tested against the reports the business already trusted, then against the questions stakeholders actually ask — until the answers came back right every time.
  • Live in weeks on a foundation that was already ready. Because the warehouse was analytics-ready, setup took weeks rather than quarters. Existing dashboards ran in parallel first, then a pilot group, then everyone.

The wave they saw coming

Before Launch · Turnaround
Field
"Who are the top targets in my territory?"
about a week
Home office
"Can you refresh the launch readiness deck?"
a few weeks
Leadership
"How does the forecast look by scenario?"
next review
Watch · PMSA Webinar Highlights

Deepanshu Arora on building it before the pressure hit

Highlights from the PMSA session on how an AI-native commercial team prepared for launch scale — what they evaluated, what they built, and what they'd tell a team starting now.

In Action

One question before a call.
Then an app for the whole field team.

A live example of how a rep gets a physician 360 ahead of a conversation — and how the same governed answer becomes something the rest of the team can use without filing a ticket.

Prep me for tomorrow's call with Dr. Rivera — patient mix, diagnosis history, recent activity.
DA
Physician 360 · CRM activity joined to claims and profilingROW-LEVEL ACCESS APPLIED · TERRITORY ONLY
Diagnosed Patients in Claims Footprint
TreatedDiagnosed, untreated
1260Dx < 6 mo6–12 mo> 12 moLapsedTime since first diagnosis
The newly diagnosed cohort is the opening. Dr. Rivera's largest untreated group sits within six months of first diagnosis — the window where a conversation changes the path. Source: patient claims · ICD-10 history
Call cadence dropped last quarter. Interaction frequency is below territory average despite a growing diagnosed panel. Source: CRM call activity
Comorbidity profile flagged. Relevant treatment history and comorbidities surfaced for the conversation, with the underlying SQL exposed on request. Source: HCP profiling · Tellius reasoning layer
Make this an app the whole field team can use.
DA
KAIYA APP
Pre-Call PlannerPublished to field · row-level access · editable in natural language
Illustrative example — the physician, metrics, and scenario shown are hypothetical.

Do not assume the AI will automatically know your organization. You have to train it like a new employee. Do that groundwork diligently, because it is going to serve you for a long time to come.

Deepanshu Arora
Associate Director, Commercial Advanced Analytics · Agios Pharmaceuticals
Deep Dive · The Architecture

A governed virtual analyst,
built on the warehouse they already had

Nothing was ripped out. The existing warehouse stayed the single source of truth, and Tellius went on top as the semantic, reasoning, and agentic worker layer. That is what makes the speed safe: every answer runs on the same definitions the business already trusts, with the SQL exposed underneath.

Tellius — AI worker layer
workflows · apps · answers
Semantic + reasoning
definitions · rules · guardrails
Your data
warehouse · data lake
Semantic + reasoning layer

Metric definitions and business rules, validated metric by metric against trusted reports. The same answer for every user, with the SQL exposed.

Agentic workflows on a schedule

The most-asked questions run on a cadence, delivering scorecards and briefs to email, Slack, or Teams.

Agentic apps for everyone else

Dashboard-like apps built and changed in natural language, with deeper exploration sitting underneath the default answer.

Guardrails from day one

Row-level access so a rep sees only their own territory, defined workflows, and a validation agent checking output against source data.

Implementation

Weeks, not quarters

The objection Deepanshu hears most is that this has to be a multi-year project. It wasn't — and the reason is unglamorous. Agios had already done the data work. "AI has skills. What it lacks is context, and that's where you come in."

01

AI-ready data

The warehouse was already analytics-ready and structured as the single source of truth. Tellius connected to the data lake directly. His rule of thumb: if a person can manage the data, AI will be able to manage it too.

02

Context + rules

Business context, KPI definitions, and guardrails set up in a few weeks — including row-level access, so a rep sees only their own territory from the first login.

03

Validate + pilot

The step most teams skip. The team compiled the questions market access, field, and home office actually ask, then tested against that list until the answers came back right every time.

04

Roll out wide

Existing dashboards ran in parallel first so the business could check the new answers against the old ones. Then a pilot group. Then everyone.

What It Grew Into

The ask was a virtual analyst. The use cases outgrew the name.

1
Pre-Call Planning: Physician 360A full view of a physician before the conversation — from patient mix down to diagnosis history, joined across CRM activity, claims, and profiling data.
2
Market Mapping + PrevalenceSizing prevalence in new markets to inform where the commercial effort goes next.
3
Customer + Patient SegmentationSegmentation work that runs alongside market mapping rather than queuing behind it as a separate project.
4
Root-Cause "Why" AnalysisWhy a trend broke, why a geography is underperforming — with the full diagnostic picture. Multi-week projects now run in minutes, while the team is still exploring.
5
Agentic Apps + DashboardsLive apps that replaced static Excel and PowerPoint — which matters most post-launch, when a static report is stale before it circulates.
6
Live Course-Correction"If we need to go back and forth, we do it live. We course-correct immediately instead of waiting those couple of days."
The Operating Model · One Foundation, Different Front Doors
GOVERNED · ROW-LEVEL ACCESS
1
POWER USERS
Ask anything

People who want to interrogate the data directly — including an RVP of Sales who asks his own questions and builds his own custom reports.

Trained on the prompts that work — and, deliberately, on what the platform can't do
2
AGENTIC WORKFLOWS
Encoded once, run the same way every time

For the questions many people ask, the team encodes the steps to follow and the structure of the output, so a familiar question is handled identically each run.

Scorecards and briefs delivered to email, Slack, or Teams on a schedule
3
EVERYONE ELSE
Answers served, not searched

Agentic apps that behave like dashboards on the surface — default answers up front, deeper exploration underneath. Users don't care what's beneath. The analytics team cares a great deal.

Built in a fraction of the time and changed in natural language
Advice

What Deepanshu would tell a team starting now

01

Start before the pressure builds

Agios started about a year ahead of its launch. "If you start when the pressure hits, panic kicks in and you are not diligent enough." The diligence that makes this work is a luxury of teams that aren't already underwater.

02

Separate governed from general

Commercial numbers need the same right answer every run. General-purpose AI covers the rest — email, slides, general analysis. Don't cross the streams.

03

Don't skip the data groundwork

Analytics-ready data and the semantic layer were most of the effort, and they're what make the speed safe. Define your success metric before the pilot starts — for Agios it was turnaround time on the most frequently asked questions.

04

Stay close to the technology

Building dashboards from a prompt wasn't on the radar at kickoff. The idea arrived months into production and landed on a foundation that was ready for it. "AI is evolving every day whether we like it or not."

What Is Tellius

Decision AI for the enterprise — the intelligence layer connecting your data to your decisions.

Tellius gives commercial teams an AI worker that shows up with the work done. It connects structured and unstructured data, applies your business rules through a governed semantic and reasoning layer, and delivers finished work where your team already works.

Live in weeks on the stack you already own.

Used by 8 of the top 10 pharmaceutical companies
5× Gartner Magic Quadrant Visionary, 2022–2026
Production-ready in weeks alongside existing BI tools