How Aperture is built, sold and staffed — including the agent desk.

Aperture sells indication intelligence to corporate development, search & evaluation, and M&A teams in biotech and big pharma. The product is a living briefing — epidemiology you can model, a treatment algorithm with share and residual unmet need, a classified pipeline, price, KOLs, centres, and an asset deep-dive with PoS, time-to-market and peak sales.

We do not sell a 12-week transformation. We sell the object those projects always collapsed into: the six pages the IC actually used.

• The same OA has been sold at €100–150k for a decade. Buyers are tired.
• Labels, CT.gov, FDA letters, registries and open social are machine-readable enough for a first pass.
• Gene therapy safety events (Elevidys 2025) and modality races (ATTR, SMA) punish anyone working from a six-month-old deck.
• Corp Dev headcount did not grow. The number of assets they are asked to look at did.

Primary ICP: VP / Director Corp Dev, Search & Eval, CBO office, in EU and US specialty / rare / cardio-metabolic franchises.

Buyer job: “I have ten days and a partner who wants a view on this indication before we take the meeting.”

Not the buyer: full commercial DD on a live auction (we will lose to the bulge-bracket and should). Medical affairs wanting a congress booth. Banks wanting a CIM appendix.

Snapshot €12k / 5 days. Brief €28k / 10 days. Mandate €48k / 15 days. Desk retainer €9.5k / month.

Contribution: a Brief should take ~4–6 partner-hours of review plus agent compute, against €28k. That is the whole point. If a Brief takes two consultant-weeks we have rebuilt the firm we left.

Year-1 target (directional): 18 Briefs + 8 Snapshots + 3 Mandates + 2 Desk seats ≈ €750–900k revenue, two partners, no associates. Profit is the point of year one, not headcount.

1. Ship the Duchenne brief as the credential. SMA and ATTR as teasers.
2. Twenty named accounts from the old book — Corp Dev, S&E, CBO. Personal note, link to the brief, one ask: a 30-minute scope.
3. Three warm intros into EU big-pharma rare and US mid-cap BD.
4. One webinar / breakfast: “What Elevidys’s boxed warning does to a DMD model.”
5. By day 30: first paid Snapshot, even at a friends-and-family number, so the motion is real.
6. By day 90: three paid Briefs, one Desk conversation, a second full indication (our pick: SMA or a mutation-cut oncology).

Humans do not scrape CT.gov at midnight. Agents do. Humans decide what the IC is allowed to believe. That split is the operating system.

Workflow, in order:

1. Scope agent. Indication, geography, decision type (in-license, out-license, competitive, IC). Outputs a question tree and a data-need list.
2. Epi agent. Registries, labels, HTA, papers. Builds the funnel (prevalent → diagnosed → staged → treated → addressable). Flags every number that is an estimate.
3. SOC agent. Guidelines, labels, share proxies (claims if we have them, otherwise KOL + launch + analogue). Efficacy, safety, residual unmet need per product.
4. Pipeline agent. CT.gov / EU CTR, company decks, congress abstracts. Classifies phase, modality, pathway, route. Late-stage gets endpoints and last readout.
5. Access / price agent. WAC, known net ranges, HTA opinions (HAS, NICE, G-BA).
6. KOL + CoE agent. Trial PIs, last three congress chairs, ERN / MDA / NAC-style networks, volume centres.
7. Voice agent. Advocacy sites, X, open caregiver forums, KOL social. Themes, not quotes we cannot defend. No scraping of closed Facebook groups.
8. Valuation agent. PoS priors by phase × TA × modality, haircut for class safety, TTM from catalysts, peak build from the epi model.
9. Editor agent. Writes the snapshot and the so-whats. Forced to take a view.
10. Partner review. We cut, we argue, we sign. Nothing ships without a name on it.

Grok (and Grok Automations) is the default runtime because we already live in it and because the same model can search, write and be scheduled.

• In-product. The “Ask the brief” desk on a signed-in briefing — constrained to that indication’s evidence pack. That is already wired in this workspace.
• Build-time agents. For each new indication we run a Grok project with the ten roles above as named chats / custom instructions, plus a shared evidence folder (PDF labels, CT.gov exports, a sources.json). The partner is the last node.
• Grok Automations (scheduled). Weekly: “Diff CT.gov + FDA + EMA + company IR for assets on the watchlist. Write a 12-line delta. Only page a partner if phase, label or safety moved.” Monthly: epi and price refresh.
• What we will not do. Unattended publishing. Scraping closed patient groups. Inventing a claims database we do not have. Calling a number “fact” when it is a model.

Other stacks (n8n + a model, a custom orchestrator) are fine later. They are not the week-one problem. Week one is: one indication, one signed brief, twenty conversations.

• Every figure is labelled: label / registry / Aperture estimate.
• Teaser case studies hold back price and peak on purpose — that is the Brief.
• Patient voice is thematic and directional until we have a consented panel (Mama Health-class). We say so.
• We do not pretend to own claims or EHR. If a client has them, we plug them into the same funnel.

A self-serve indication generator for strangers. A marketplace. An oncology module with 400 tumour types. A mobile app. Those are how this becomes a software company before it is a desk. First we sell Briefs.

• Named company, offer, and landing that can be sent tomorrow.
• Open Duchenne brief with interactive epidemiology, treatment, pipeline, landscape, two asset dives.
• SMA and ATTR teasers so the portfolio is visible.
• Contact capture. Auth for export, saved scenarios, ask-the-brief.
• This playbook, so we do not re-litigate the model in Slack.