Ask a simple question about almost any private company — what does it actually sell? — and watch how hard it is to answer well. Product intelligence is the least developed layer of company data, and it shows. Firmographic databases will tell you where a company is headquartered, when it was founded, and who invested. Then they hand you a one-line description written years ago and a category tag, and the most commercially important fact about the company — its actual products and services — is something you still have to reconstruct by hand.
We think that is backwards. What a company sells is the center of gravity for nearly every serious research task: account research before a sales conversation, battlecards before a competitive deal, market scans before an entry decision. So on aVenture, products and services are not an afterthought on the company profile. They are first-class records in the research graph, built and maintained by an agent-powered research pipeline.
The gap: firmographics tell you who, not what
The standard company record answers identity questions. It rarely answers offering questions:
- What are this company's distinct products and services — not its tagline, its actual lines of business?
- How does each one position itself, and in what category does it compete?
- Which other companies sell against each of those offerings?
Teams close this gap manually today. A seller preparing for an account skims the prospect's website and guesses at the product line. An analyst building a battlecard copies positioning language into a slide, where it starts going stale immediately. A strategy team scoping a market pays for a one-off scan that cannot be refreshed without paying again.
The information is public. The problem is that it lives in marketing copy — inconsistent, self-described, scattered across pages — and nothing turns it into structured, comparable, current records. That transformation is exactly what we built.
Products and services as first-class records
On aVenture, each product or service is its own record, tied to the company that provides it, with a name, a description, and a category. That sounds mundane. It is not. Giving offerings their own identity in the graph is what makes every downstream use possible:
- Offerings can be compared across companies, because they exist as parallel structured records instead of free text.
- Offerings can be linked — most importantly to competing offerings from other providers, which places each product in its market context.
- Offerings can be maintained — updated, corrected, deduplicated — without rewriting a company-level description blob.
Our research agents build these records from public evidence, company by company. On company profiles, analysis sections summarize what the company offers alongside the rest of its research record — funding, people, news, relationships — so the offering layer is part of the same picture, not a separate export.
We are also explicit about the boundary between description and inference. An offering record reflects what the evidence supports about a real product a real company sells — it is not a paraphrased tagline, and not a guess promoted to a fact.
How the catalog stays clean
Anyone can scrape product pages. The hard part — the part we spend most of our engineering effort on — is keeping a catalog of many companies' offerings trustworthy. Three mechanisms do the work.
Duplicate detection, continuously
The same product surfaces again and again under different names: a rebrand, a shortened name on a pricing page, a differently worded description in a press mention. If each surface becomes a new record, the catalog inflates and every count built on it is wrong. We run dedicated duplicate sweeps across the offering catalog to collapse these into single records, so one product is one record regardless of how many research paths discovered it.
Attribution on every write
Every offering record carries provenance: where the fact came from, when it was recorded, and — because our enrichment is agent-powered — which research agent and which model produced it. That attribution is recorded at write time, not reconstructed later. If a record looks off, we can trace exactly how it was produced. If a model or pipeline is upgraded, we know which records it touched. Research you cannot audit is research you eventually stop trusting; we built the audit trail in from the start.
Validation before storage
Structured research is only structured if the structure is enforced. Writes into our research graph pass through governed contracts — the shape, the allowed values, the required fields are checked when data enters, not cleaned up after. A catalog that accepts anything degrades into the free-text soup it was meant to replace.
Market context: competing-product links
An offering record in isolation answers what does this company sell. The more valuable question is usually against whom.
Because offerings are first-class records, competition can be recorded as an explicit link between a company's product and competing products from other providers. Our research agents sweep the market for these competing offerings deliberately, starting from a specific product rather than a category label — which is how you catch the rival that describes the same capability in completely different words.
On the company profile, each offering carries its researched competing products.
For anyone doing sales intelligence research, this is the difference between a static battlecard and a living one: the competitive set attached to a product updates as research updates, rather than freezing at the moment an analyst last edited a slide.
What this unlocks for GTM and research teams
Concretely, a governed product and service layer changes several everyday workflows:
- Account research. Before a first conversation, see a prospect's actual lines of business as structured records — with the company's funding, people, and news one step away in the same graph — instead of skimming a website and hoping you inferred the product line correctly.
- Battlecards. Build competitive one-pagers from explicit product-versus-product links, each backed by recorded sources, and come back to a card that has kept itself current.
- Market entry scans. Enumerate who sells a type of offering today, across category labels, and see the providers behind each product.
- Portfolio and coverage work. For investors, product records make "what does this company actually do" a queryable fact across an entire watchlist rather than a memo-by-memo rediscovery.
The common thread: when offerings are records instead of prose, questions about them become queries instead of projects.
Built for correctness first
We want to be honest about the trade we are making. Extracting product claims from the public web with agents is the easy half; we have invested most heavily in the unglamorous half — validation contracts, duplicate sweeps, and write-time attribution — because a product catalog that is fast but silently wrong is worse than no catalog at all. Depth first, scale second is the order we build in, and the offering layer is where that discipline matters most.
Try it early
aVenture has not commercially launched yet. We are building an agent-powered research system for private company intelligence, and the product and service layer described here is one of the pieces we most want early users to push on: tell us where the catalog is right, where it is wrong, and what your team would need it to answer.
If knowing what companies actually sell is a problem you work on, join the free research preview waitlist at https://aventure.vc/free-research.