The Synthesis Company

What do they actually do

The Synthesis Company builds software that turns very large sets of academic papers into usable, checked summaries and evidence tables so teams can run defensible literature reviews much faster. The system ingests large corpora, automates screening for relevance, extracts study details and results, and links every claim to a traceable citation an expert can verify (homepage). They report running systematic reviews over tens of thousands of papers in weeks and emphasize high citation accuracy on internal tests (YC profile).

Today they deliver reviewed systematic‑review artifacts that research teams and businesses can use directly. Early work and collaborations include academic groups and institutions such as Stanford, Harvard, and Johns Hopkins (YC profile). Public updates describe a verification approach that combines document search, bibliometric scoring, and guided reasoning traces to reduce citation errors (LinkedIn). The team’s open‑source tooling and DSPy contributions underpin their production prompt and extraction infrastructure (GitHub, YC profile).

Who are their target customer(s)

• Academic researchers and systematic‑review teams: They must screen, read, and extract data from thousands of papers by hand, which takes months and risks missing relevant studies or introducing errors. They need reproducible evidence tables and traceable citations for peer review and publication.
• Pharmaceutical R&D and regulatory affairs teams: They need audit‑ready summaries and exact citations to support filings or label decisions, but manual reviews are slow, costly, and hard to defend under inspection. Missing or mis‑extracted studies can delay approvals or create compliance risk.
• Clinical guideline authors and medical‑society panels: Committees must produce defensible, up‑to‑date recommendations from a rapidly growing literature with limited reviewer bandwidth. They need synthesized evidence that is transparent enough for committee debate and public scrutiny.
• Health‑economics and market‑access teams: They require precise effect sizes, population definitions, and real‑world evidence from diverse sources to build models and payer dossiers. Inconsistent extraction or missed studies undermines pricing, reimbursement negotiations, and launch timelines.
• AI/ML research teams and evaluation groups: They want large, expert‑validated datasets and reasoning traces for training or benchmarking scientific reasoning, but curating and validating such corpora by hand is prohibitively expensive. Without trustworthy data, evaluations and downstream research are unreliable.

How would they acquire their first 10, 50, and 100 customers

• First 10: Run high‑touch paid pilots with labs and collaborators already in network, delivering auditable evidence tables and a validated summary; capture written testimonials, short case studies, and referrals upon delivery.
• First 50: Package the pilot into a repeatable offer for pharma R&D, regulatory teams, and medical‑society panels; pursue conference talks/sponsorships and recruit a few CRO/consulting partners to resell or hand off review work, with standardized onboarding and reference commitments.
• First 100: Launch tiered self‑serve and enterprise/API (including on‑prem) options, integrate common literature sources/reference managers, and build inside‑sales plus reseller agreements with publishers, guideline bodies, and eval labs; support with a small case‑study library, onboarding checklist, and referral incentives.

What is the rough total addressable market

Top-down context:

Biomedical research now publishes roughly 80 systematic reviews per day (~29k/year as of 2019), illustrating large, ongoing demand for evidence synthesis (Journal of Clinical Epidemiology; PubMed). Related outsourcing categories such as medical/regulatory writing are multi‑billion‑dollar markets (~$4–5B mid‑2020s, growing ~10% CAGR) (Grand View Research; Mordor Intelligence).

Bottom-up calculation:

Pharma/regulatory: assume ~500 mid‑to‑large biopharma and device firms each run ~40 literature evidence reviews/year across programs at ~$30k average per review ≈ ~$600M. Academia/guidelines and APIs add a conservative ~$25–50M combined, implying a beachhead TAM of roughly ~$625–650M.

Assumptions:

• Focus is on health/biomed reviews where regulatory and clinical stakes are highest; adjacent fields add upside.
• Average price reflects audit‑ready, end‑to‑end reviews (screen→extract→synthesize→verify) versus lightweight tools.
• Only mid/large enterprises are counted in the pharma/device estimate; small firms and additional indications are excluded.

Who are some of their notable competitors

• DistillerSR: Cloud software for collecting, deduplicating, screening, extracting, and reporting study data with AI‑assisted screening and audit trails; strong on workflow management for regulatory/academic reviews, but oriented to managing human‑driven processes rather than end‑to‑end automated synthesis with validated reasoning traces (product; library guide).
• Covidence: Widely used platform for collaborative screening, full‑text review, and basic extraction (popular with Cochrane and universities); speeds manual steps and tracking but is built for human reviewers and structured workflows, not at‑scale automated extraction with provenance for regulated use (homepage; MSK guide).
• Elicit (Ought): AI assistant that finds papers, pulls out answers, and generates report‑style outputs inspired by systematic reviews; overlaps on automated extraction and summaries but positions as a researcher tool rather than an enterprise, audit‑ready synthesis service (product; university guide).
• Iris.ai: Tools for large‑scale literature/patent ingestion, semantic mapping, and automatic extraction to help R&D teams find and organize research; stronger on discovery and mapping than on fully auditable evidence tables and verified reasoning for regulatory or guideline decisions (homepage; overview).
• Scite: Provides smart citations and a Reference Check showing whether citations support, contrast, or mention a claim and flags editorial notices; competes on verification/trust but does not replace end‑to‑end synthesis and structured data extraction for full reviews (product; Reference Check).