Datopian: AI-Native, Run by Full-Cycle Operators

Internal strategy memo — leadership only. July 2026.

Executive summary

The problem. Datopian already sells the work, not a tool — the structural advantage every data-catalog SaaS vendor lacks. But we price that work by headcount and hours. Agents are collapsing our delivery cost while our price stays fixed to human time, so the value of every model release leaks to the client instead of accruing to us.

The proposal. Two changes, and they depend on each other:

Sell outcomes, not hours. A fixed-price catalogue — portal live in 30 days, catalogue migration per 1,000 datasets, DCAT compliance, quality monitoring — priced to client value rather than our cost. Every engagement lands on PortalJS Arc with a subscription, so services acquire and recurring revenue compounds.
One person owns the full cycle. Marketing through sales, design, delivery and renewal, run by a single accountable operator with an agent fleet. Agents absorb the labour at every stage; what's left is judgement, which transfers across stages far better than craft skill does. Why now. Migration is our wedge: well-scoped, pure intelligence work, already outsourced, verifiable, with an existing budget line — a vendor swap rather than a reorg. And we hold assets no one else does: CKAN co-stewardship for trust and distribution, PortalJS already agent-native, Arc for recurring revenue.

The ask. Freeze headcount for four quarters. Publish the fixed-price catalogue this quarter and run two full-cycle pilots with volunteers. Target by Q4: recurring ≥50% of revenue, revenue per FTE at 2x baseline, cash-flow positive.

The risk of waiting. Someone will sell "your data portal, live, guaranteed, $X" into our market within 18 months. The transition is optional today and won't be.

TODO

The thesis

Sequoia's Services: The New Software argues the next $1T company is "a software company masquerading as a services firm." Sell the tool and you race the model; sell the work and every model release makes you cheaper and harder to beat.

Datopian already sells the work. That is our structural advantage over every data-catalog SaaS vendor. But we price the work like a 2015 consultancy: staffed teams, time-and-materials, revenue capped by headcount. Our cost base falls with each model release; our price doesn't. That gap is the entire opportunity — and, if a competitor takes it first, the entire risk.

Split our delivery work the way the article does. Intelligence — harvesting and migrating catalogs, schema inference, metadata mapping to DCAT/Frictionless, pipeline plumbing, QA, frontend assembly, docs — is 70–80% of a typical engagement and is now largely machine-doable. Judgement — data strategy, governance design, stakeholder wrangling, procurement, architecture calls — is the remaining 20–30% and is where our reputation actually lives. Today we bill both at roughly the same rate. That is the mispricing.

The evidence on team size

Revenue is decoupling from headcount. The cleanest datapoint: Lovable crossed $400M ARR in February 2026 with 146 full-time employees — ~$2.7M per head, company-confirmed. Cursor passed $500M ARR in mid-2025 with a team reported around 60. Midjourney has taken no outside funding at all and is credibly estimated in the low hundreds of millions of ARR on well under 100 people, though it discloses nothing. For comparison, median private SaaS runs ~$175K ARR per employee — an order of magnitude lower. Academic work (Kim/INSEAD, Koning/HBS, ~2,900 YC startups plus ~50k PitchBook companies) finds AI-native firms run roughly 25% leaner than comparable peers, with ~15% fewer entry-level staff and ~15% fewer managers.

The lesson is not "fire people." It is that headcount has stopped being the unit of capacity, so any business whose price, plan and org chart are denominated in headcount is structurally disadvantaged.

The org model: the full-cycle operator

The core organisational bet: one person owns the entire cycle. Content and marketing → outbound and qualification → discovery and solution design → scoping and proposal → build and delivery → launch → account expansion and renewal. Not a specialist in a relay race — a single accountable operator running the whole loop with agents doing the labour at every stage.

Why this is now possible: each of those stages used to require a distinct professional because each was labour-heavy. Agents absorb the labour. What remains at each stage is judgement — which client to pursue, what to promise, what architecture fits, what "done" means, whether the output is good enough to put our name on. Judgement transfers across stages far better than craft skill does. One person with taste plus a fleet beats five specialists plus a coordination tax.

Why it wins commercially: handoffs are where margin, context and trust leak.

The person who wrote the proposal is the person who delivers it, so scoping gets honest fast — a full-cycle operator who mis-scopes eats their own mistake, which is the tightest feedback loop in a fixed-price business.
Clients get one accountable name instead of an account manager relaying to a delivery lead.

And the marginal cost of pursuing a small deal collapses, which opens the long tail of city, agency and NGO clients we currently can't afford to chase.

What it demands of us:

One title, one track. Collapse role titles into a single technical track. Career progression = scope owned end-to-end, not reports managed.
Everyone sells; everyone ships. No pure-delivery and no pure-sales roles. Uncomfortable, and non-negotiable — the model breaks if some people opt out of half the cycle.
Shared spine, not shared work. Full-cycle only scales if the repeatable parts are institutional: the fixed-price catalogue, PortalJS skills, proposal templates, pricing guardrails, Arc deployment. Individual autonomy on top of a strong shared spine — not everyone inventing their own process.
Hard guardrails. Pricing floors, scope templates, a second pair of eyes on anything above a threshold, and shared clients-of-record. Autonomy without underwriting discipline is how a fixed-price business dies.
A different hire. Recruit for range and ownership — people who will write the blog post, run the call and ship the portal. Rarer than specialists, and far more valuable. A handful of central roles (finance, legal, contracts) stay specialised for now.

The scarce human skill shifts from doing the work to specifying, reviewing and standing behind the work — because what the client actually buys is our accountability, the one thing a model cannot sell them.

What we become

The autopilot for public and organisational data. Not "we build you a portal" but "your data is published, catalogued, quality-checked and kept current — priced per outcome, guaranteed."

We are unusually well positioned: CKAN co-stewardship gives us the incumbent standard and the client relationships; PortalJS is already agent-native (architect / new-portal / add-dataset / migrate / deploy skills); and PortalJS Cloud/Arc gives us managed hosting — the recurring-revenue substrate.

Five moves.

Reprice. Kill time-and-materials for intelligence work. Ship a fixed-price catalogue: Portal Live in 30 Days, Catalog Migration (per 1,000 datasets), DCAT Compliance, Data Quality Monitoring. Fix the price to client value, not our cost. Margin accrues to us as models improve.
Productise the wedge. Migration is our Crosby-NDA: well-scoped, pure intelligence, already outsourced, verifiable output, existing budget line, frictionless vendor swap. Win on migration; expand into governance and strategy where judgement still commands a premium.
Rebuild delivery agent-first. Every repeated delivery step becomes a PortalJS skill within one quarter of being done twice. Publishing the skills openly is the moat, not the leak — it feeds the data and the funnel.
Flatten and go full-cycle. One technical track, no net hires for four quarters, and every operator running marketing-to-renewal on their own accounts. Redeploy freed capacity into skills and product.

Scoreboard: revenue per FTE · gross margin per fixed-price engagement · recurring share of revenue · human-hours per portal delivered · % of delivery steps covered by a skill · % of engagements owned end-to-end by one operator.

What we stop doing

Hourly billing for intelligence work. Staffing projects by seniority pyramid. Handoffs between sales and delivery. Bespoke one-off builds that produce no reusable skill. Hiring to win capacity-constrained deals.

Risks

Public-sector procurement is built for day rates and may resist outcome pricing — expect a sales-cycle drag; pilot with two friendly clients first. Fixed price transfers delivery risk to us, so underwriting discipline matters more than sales volume.

Full-cycle is the hardest part: some strong specialists will not want it or will not be good at it, and forcing the model on everyone at once would cost us people we need — hence volunteers first, default later. Existing revenue must fund the transition, so sequence it rather than cut over. And the model improvements that lower our costs also lower a new entrant's barrier: our defensibility is CKAN trust, migration data and accountability — not code.

The asymmetry: the transition is optional for us right now, and won't be for long. Someone will sell "your data portal, live, guaranteed, $X" into our market. It should be us.

Sources: Sequoia Capital, "Services: The New Software," Julien Bek (Mar 2026) — note Crosby is a Sequoia portfolio company and the $1:$6 ratio is directional framing, not a footnoted statistic. Lovable figures: TechCrunch (Mar 2026). Cursor: Fortune / Bloomberg via Wikipedia. Midjourney: Sacra equity research. Leanness and manager-ratio findings: Kim (INSEAD) & Koning (HBS) via Forbes (Jun 2026). SaaS benchmark: Benchmarkit 2026. Titles: OpenAI job conventions and anthropic.com/jobs. PortalJS: datopian/portaljs.