Steno's $49M and the data-moat trade in legal AI

Steno closed a $49M Series C this week, bringing total funding to $150M, on a thesis that is doing a lot of work: that running a court-reporting services business is the right way to build a legal AI product. Savano Capital led, with First Round and The Legal Tech Fund following on. The pitch is vertical integration (one company owning the data origination layer and the software layer that sits on top), and it sits at an interesting angle to where the rest of the legal AI market is going.

What was actually funded. Steno is two businesses stapled together. One is a court-reporting services firm: stenographers and videographers attend depositions, produce the official transcript, and bill the firm. The other is Transcript Genius, a generative AI product that lets litigators search, summarise and interrogate those transcripts. The services arm originates the data. The software arm monetises it a second time.

The investor deck reading is that the services business is the moat and the software is the product. Pure-software competitors, Harvey, Legora, the long tail of deposition-analysis tools, have to either license transcripts, scrape them from court records, or wait for the firm to upload its own archive. Steno already has them, because Steno made them.

The frame. Two lenses fit this one. The first is vertical integration as data strategy: owning an upstream service business to feed a downstream AI product, on the bet that proprietary data beats model quality at the application layer. The second is FDE market structure (forward-deployed engineering, the embedded-into-customer-workflow go-to-market pattern): Steno deploys AI inside a relationship law firms already pay for, dodging the enterprise procurement battle that pure-software vendors fight every quarter.

Both frames flatter the deal. Both are also testable.

Where the frame holds. The FDE read is the stronger one. AmLaw 200 firms (the 200 largest US firms by revenue, who do most of the high-stakes litigation work) already have Steno on a vendor list, already process its invoices, already have its stenographers in their conference rooms. Selling them Transcript Genius is a line-item expansion, not a new procurement cycle. Pure-software vendors selling into the same firms run a six-to-twelve month enterprise sale, security review, and pilot. Steno runs an upsell.

This is the same structural move that has worked for embedded-engineer go-to-market at the frontier labs: capability gets deployed where the customer is already paying, not where procurement has to approve a new SKU. It is replicable in other regulated verticals where a services relationship already exists — medical transcription, audit working papers, regulatory filings.

Where the frame breaks. The data moat is thinner than the pitch suggests. Court-reporting transcripts have a regulated, standardised format. The content is whatever the deposition contained. Steno's corpus is large, but it is not structurally different from the transcript a competing reporter would produce of the same deposition, and AmLaw 200 firms already own copies of every transcript Steno has ever delivered to them. Nothing stops a firm from exporting its archive into Harvey or Legora and getting most of the retrieval benefit.

What Steno actually owns is the workflow — the live pipeline, the metadata, the indexing, the linkage between transcript and case. That is a real edge for retrieval quality and for product features (real-time deposition assistance, cross-case search inside Steno's own UI). It is not an edge on model capability. If Harvey trains a better legal reasoning model, Steno's transcript advantage compresses to "we have a slightly nicer search box".

The bifurcation. Step back and the legal AI funding map has three altitudes. Legora at $550M and Harvey at a cumulative $300M-plus are operating at frontier-application scale, buying distribution and model spend that mid-market vendors cannot match. Steno at $150M total is in the niche-data tier, alongside other vendors who own a specific corpus or workflow. The middle, pure-software legal AI without either frontier capital or proprietary data, is the squeezed segment.

Steno's round is consistent with that map. Investors are funding either the top of the stack or the proprietary-data plays underneath, and bypassing the middle. It is a coherent allocation if you think model quality at the top will commoditise and data ownership underneath will not.

The counter-case is five days old. On 21 May, Osborne Clarke spun off Justima, its internal legal AI product, into a standalone company. Osborne Clarke owned the workflow (it is the law firm) and chose to externalise the product. Steno owns the workflow (it is the reporter) and is building the product in-house.

These are opposite bets on where the value sits. OC's read is that the workflow firm is the wrong corporate vehicle to commercialise software — different margin profile, different talent pool, different investor base. Steno's read is that the workflow firm is exactly the right vehicle, because the services relationship is the distribution. Both can be right for their respective businesses. They cannot both be right as a general theory of the legal AI market.

The execution-risk question I would press. A court-reporting services business is operationally heavy in ways a SaaS business is not — stenographers, scheduling, regional coverage, accounts receivable from law firms (who pay slowly). The $49M is presumably being split between scaling services capacity and scaling the AI product. The investor pitch wants software multiples; the cost structure delivers services margins on a chunk of revenue. That tension does not resolve itself with one Series C.

I notice the disclosure does not break out what fraction of Steno's revenue is services versus Transcript Genius. That is the number that would tell you whether this is a software business with a services moat, or a services business with an AI feature. The two get valued very differently.

What to watch.

• ARR (annual recurring revenue) disclosure for Transcript Genius specifically, separated from court-reporting services revenue.
• Whether Harvey or Legora announces transcript-ingestion partnerships with court-reporting firms — the obvious counter-move.
• Net dollar retention (revenue kept from existing customers after expansion and churn) on Transcript Genius inside AmLaw 200 accounts. If the upsell thesis works, this number will be high.
• Further services-plus-AI rounds in adjacent regulated verticals. The pattern is replicable; if it spreads, the frame is doing real work.

Footnotes

¶