Trueness · plan

Trueness.ai business and pilot plan

Trueness.ai should start as a managed audience research business: audit an existing synthetic audience against fresh human evidence, build a bespoke audience where needed, and maintain only the uses that the evidence supports. The commercial proposition is a useful audience with a visible, current account of its limits.

This planning package covers the competitive set, positioning, operating model, three financial cases, written brand platform, pilot research method, customer and operator workflows, AI automation and assistant integrations. It is dated 25 October 2026. All Trueness financial assumptions and outputs are in USD. No visual identity or application has been built, no customer has been contacted, and no business has been launched by this work.

The decision

Proceed to paid market validation and exact-scope research quotes before committing to the full platform. Begin with three design partners, one Audit and two Audience builds, in an accessible consumer category. Agencies and brand teams with existing research are the working target; travel and hospitality provide a possible entry network, not an established exclusive market.

Human calibration is already part of the competitive landscape. Electric Twin publicly describes survey-grounded audiences, hidden-response tests and later revalidation; Ipsos publishes work on measuring synthetic uncertainty. Trueness therefore needs to earn a more specific position: provider-neutral assessment, portable customer records, explicit failed and inconclusive findings, and continuing checks for a named use. That combination is a positioning hypothesis, not evidence of an uncontested category. See Electric Twin's method and Ipsos's confidence paper overview.

The commercial advantage to test is whether customers pay for that accountable service and use it repeatedly. The defensible assets would be permissioned validation data, documented methods, delivery relationships, customer outcomes and reputation. Two frontier models, a persona interface or an MCP connection are not sufficient advantages by themselves.

Read the plan

Question Document
How should the venture operate and win customers? Business plan
Who competes, what do they claim, and how do they sell? Market landscape and pricing evidence
What should the brand stand for and sound like? Written brand platform, following the reference brand-page template
What does the earlier reference synthetic panel work contribute? Implementation inheritance and construction-method lessons
How is faithfulness tested? Calibration protocol and primary evidence register
What could the business earn and what cash does it need? Financial case and editable Excel model
What happens during the first three paid engagements? 90-day pilot
How do delivery, administration and AI automation work? Operating procedures and record templates
What should customer and back-office screens do? Product and UX specification
How should Claude, ChatGPT, APIs and MCP connect? Models and integrations

The proposed offers

Offer Mid-case price hypothesis Deliverable Direct cost allowance
Audit $15,000 Test one existing audience and bounded task against 240 fresh human respondents; deliver a Calibration Record $10,325
Audience $35,000 One audience/market/language, evidence-grounded build, 12 discovery interviews, 160 development and 240 sealed validation completes $26,325
Maintained $4,000/month Four bounded studies, six hours of research/service review, quarterly human pulse and controlled access $1,795/month

These are proposed prices and prudential cost allowances, not observed demand or supplier quotes. Payment for an Audit or build buys the agreed investigation and documented finding, including a negative or inconclusive outcome. Maintained access requires an appropriate supported scope. A full rebuild or expanded validation is separately quoted; no annual revalidation revenue is assumed in the base model.

The pilot runs new stimuli and new human respondents after the candidate is frozen. The Calibration Record gives metrics, uncertainty, supported tasks, exclusions, dates, reviewer and expiry. It is not a universal truth score or accredited certification. Model agreement and large numbers of generated responses do not replace human evidence. A later synthetic-only study shows its actual fresh-human base, often zero, separately from the historical validation sample.

Financial cases

The model includes paid founder time, research labour, recruitment, AI, engineering, commercial acquisition and staffing, operations, setup delays, conversion, churn, receivables, deposits and research work in progress. Development is expensed. EBITDA and cash are shown separately. Years are modeled operating years from the conditional launch; startup costs include a prelaunch period.

Case Year 1 revenue Year 1 EBITDA Year 2 EBITDA Year 3 revenue Year 3 EBITDA Modeled funding with 20% contingency
Low $190,701 −$227,025 −$170,912 $713,557 −$142,258 $726,925
Mid $559,230 −$70,780 $257,092 $2,961,318 $702,295 $172,137
High $1,475,444 $338,167 $1,662,654 $8,213,746 $3,570,846 $150,720

The low case is a stop-or-reprice case: its $25,000 Audience fee does not cover the $26,325 direct cost. The mid case assumes 12/24/36 new Audience builds and 6/12/18 Audits across years one to three, 75% conversion from delivered builds to Maintained, and 1% monthly churn. The high case requires around eleven research delivery FTE at its peak, plus commercial and operational support. It can be light in founder delivery time, but it is not a business without human staff or contractors.

The specified three-partner pilot has a mid-case cash outlay of $167,875, including prelaunch and three postlaunch months. With milestone receipts and a 20% contingency, it requires $141,450 of funding. The full mid growth case requires $172,137 under its assumed timing, rising to $341,337 if sales slip six months. These are scenario funding needs before tax, interest and financing; a smaller concierge-only entry would need its own budget.

Premium pricing is a central uncertainty. At the separately tested $20,000 build / $2,250 monthly price, every case is loss-making in year three at the modeled scope. Validate willingness to pay and supplier costs before treating the attractive cases as a hiring or investment plan. The workbook also tests lower sales, higher human and AI costs, doubled review hours, procurement stress and churn.

Pilot and automation priorities

The 90-day pilot covers 24 discovery interviews and 1,040 quantitative human completes across three separate customers, with 720 in sealed evaluations. These cannot be pooled into one audience's sample size. The baseline delivery allowance is 200 research hours before maintained use.

Automate source indexing, evidence lineage, approved job execution, deterministic metrics, export assembly, cost logs, invoice reconciliation, reminders and exception routing first. Keep method approval, rights decisions, interpretation, release and material incidents accountable to named people. Measure total minutes per delivered outcome, including exceptions, rather than advertising an untested automation percentage.

Start with a thin web workspace and portable exports. Test a private read-only MCP connector with a willing design partner; add study creation only after permissions, spend confirmation, scope checks and idempotency work. Customer value must survive outside any one assistant or model vendor.

Evidence and completion status

The earlier workbench was inspected for architecture, corpus inventory, workflow and failure modes. Its document corpus and synthetic discussion workflow inform the lessons; confidential research and participant content have not been copied into this venture as a training corpus. Ownership and reuse permissions remain separate from the contribution of development effort.

Claude led the competitive and brand/business strategy work through the shared NTFY channel. Codex led the operating/product plan and coordinated frontier-model research and financial work. Cross-review changed the differentiation, removed a universal score, increased the research budget, corrected monetary units and caught two editable-model edge cases.

The financial package passes eight tests, including independent recalculation of its Excel formulas, cohort timing, cash/receivable/deferred-revenue/WIP identities and input validation. A separate reviewer recomputed the pilot budget and checked the main cash identities. These checks establish arithmetic consistency, not market demand or research validity.

The next evidence needed is concrete: buyer interviews and paid commitments, rights-cleared inputs, written recruitment quotes, a qualified research lead and external reviewer, and a frozen first study protocol. The planning package is complete; those are launch activities, not results already achieved.

All documents

brand

finance

operations

product

research

review

strategy