At a glance Projection

Hold The Wire: Business Plan (v2, revised after red-team review)

A pentest of the payment process. We try to steal a mid-market company's money the way a real fraud crew would (a lookalike vendor, a synthetic voice, an urgent "bank change"), stop at the moment the payment would have left, and hand the CFO a scored, point-in-time result and a fix list.

Status and integrity note. This is a pre-launch plan. There are no customers, partners, carriers, brokers, MSSPs or revenue today. Every financial figure is a projection from the assumptions in model.json (built by model_build.py); market and comparable data are cited to their sources, and where data is thin the plan says so. Version 1 of this plan (plan_v1.md, model_v1.json) was stress-tested by an internal red team covering finance, sales/go-to-market and lean operations. This version adopts most of their critiques, pushes back on a few with evidence, and summarizes every point in Section 17.

What changed in v2, in one paragraph. The base-case Year-5 exit fell from ~$28M to ~$5.6M. Data licensing is out of the base case and out of ARR; the exit is priced on normalized EBITDA, not ARR; churn is 20% not 10%; delivery is costed per entity-drill with client-coordination hours; Year-1 volume is about half; the entry offer is a fixed-fee Baseline Wire Test with no executive cloning; broker revenue shares are replaced by flat fees; the "attestation letter" became a point-in-time Drill Results Statement; and every hire is gated. What survives: a modeled peak cash need of ~$289k, a hard-stopped downside of ~$330k, ~$1.1M of pre-tax owner cash flow in Year 5, and a ~$20M bull case.

1. Executive summary

The problem. Business email compromise cost US victims about $3.05 billion in 2025 across 24,768 complaints to the FBI; total reported internet-crime losses reached $20.9B, up 26% (IC3 2025 report; summaries: SpyCloud, The Register). In the AFP's 2025 survey, 79% of organizations were targets of attempted or actual payments fraud, BEC was the top channel (63%), and vendor-impostor fraud rose 11 points to 45% (AFP press release). Deepfakes make the classic callback beatable: Arup lost about US$25.6M after a finance employee joined a video call where the CFO and colleagues were AI-generated (CNN via ABC17).

The gap. Mid-market CFOs ($50M-$1B revenue) run real money through 2-8 person AP and treasury teams. Their IT provider may phish the staff; nobody tests the money-movement process end to end (vendor bank change, payment release, callback, escalation) and reports it in CFO terms.

The offer (revised).

1. Baseline Wire Test, $9,500 fixed fee. One full drill on one entity using a synthetic vendor voice and a lookalike vendor domain, so no executive likeness consent is needed. Ends with a board-ready readout. Fee credited against Year 1 if the client converts.
2. Annual plan. Per entity per year: one full drill, one retest after remediation, two light unannounced spot checks. Core $24k (1 entity); Growth $24k + $7k per additional entity (modeled at 4 entities, $45k); Enterprise from $90k (scoped, about 10 entities, operator-run video drill), sold only after SOC 2 Type II.
3. Executive voice cloning is an opt-in upsell once trust exists, under a BIPA-grade written release with the voice sample recorded at consent.
4. Output: a scorecard split into Observed in drill and Evidenced by client, and a point-in-time Drill Results Statement signed by the company. A "Remediated and Retested on [date]" statement is issued only after controls pass.

The model. Revenue comes from Baseline Wire Tests (paid entry, about 27% of Year-5 revenue), annual plans (about 71%) and post-incident retests for existing clients (about 2%). Brokers get a flat, non-contingent fee per qualified meeting, subject to counsel's review. MSSPs get 15% of first-year revenue and 10% in years 2-3, then nothing. Benchmark data licensing is not in the base case.

Headline numbers (base case, projection):

• Capital: $120k self-funded launch; peak founder cash in ~$289k (base). No outside capital in the base case.
• Break-even: sustained monthly EBITDA-positive from month 25; launch capital and losses paid back by month 40.
• Downside: kill gates at day 90, after the first 5 paid tests, month 12 and month 18. Modeled maximum loss before the month-18 gate is ~$330k; we plan for a $500k ceiling to cover E&O and legal costs that are not yet quoted.
• Exit (Year 5, base): 7x normalized EBITDA (EBITDA ex-data, minus $125k to replace a part-time founder with a full-time CEO) = ~$5.6M EV. Structured as 65% cash at close (~$3.7M) and 35% earnout. Bull: ~$19.9M (10x, strategic buyer). Bear: stopped by the gates; run to Year 5 it would need ~$718k and be worth little.
• Returns (base): MOIC ~24x and IRR ~108% on cash; ~7x and ~70% with the founder's unpaid time costed at $200/hour. Probability-weighted (50% stopped / 30% weak / 20% base) expected net value is ~$1.4M, or ~$0.9M with founder time costed.
• Founder profile this needs: an owner-operator with security-testing or risk-services experience who knows scoping, authorization letters, rules of engagement and selling to CFOs, and can personally sell the first 15-20 deals.

The honest pitch. This is a small, high-trust services business with a software-shaped upside. The base case is a very good return on a few hundred thousand dollars and a part-time founder, not a venture-scale outcome. The bull case, and any software multiple, depends on things we have to prove: retention above 85%, operator time per drill under one hour, and one carrier that values the result.

2. Problem and customer

Who exactly

Economic buyer: CFO (or VP Finance) at a US company with 100-2,500 employees and $50M-$1B revenue. Champion: Controller or Treasurer. Approvers we now plan for (red-team finding): general counsel, HR, IT (email allowlisting), and, for the cloning upsell only, each executive whose voice is used. Influencers: commercial insurance broker, external auditor, audit committee chair.

First verticals and why:

The pain, with evidence

• Frequency: 79% of organizations hit by attempted or actual payments fraud; BEC the top channel at 63%; vendor-impostor fraud up 11 points to 45%; executive impersonation at 49% (AFP 2025; Truist highlights PDF). Vendor impostors are the faster-growing vector, which is why the entry test impersonates a vendor, not an executive.
• Severity: $3.05B in reported BEC losses in 2025, about $123k per complaint on average (IC3 2025). That average includes very small victims; a mid-market AP run is larger. IRONSCALES' vendor survey reports about $280k per deepfake incident (IRONSCALES). Caveat: vendor-sponsored survey.
• Deepfake targeting: Ponemon (2025) found 41% of security professionals reported deepfake campaigns targeting executives; Medius reported 53% of finance professionals targeted (via Programs.com and DeepStrike). Caveat: secondary compilations of vendor surveys; directional only.
• Controls are manual and untested: a Trustpair survey found 69% still rely on manual fraud prevention such as callbacks (via Programs.com).
• Insurance raises the stakes, in both directions. Social-engineering coverage is often sublimited at $50k-$250k, and carriers increasingly make a callback to a previously known number a condition of coverage (Amwins; Insurance Business; Bank Director). Two consequences: the loss above the sublimit is uninsured and lands on the CFO, and a written finding that the callback failed can itself become a disclosure issue at renewal. The offer is designed around the second point (Section 3).

How they buy

• Budget line: risk, finance or audit, not IT. A $9.5k fixed fee is a signature, not a procurement project.
• Triggers: a near-miss or peer loss; insurance renewal (the questionnaire asks about callback procedures); an audit-committee question; an acquisition adding entities; new AP staff or a vendor-master migration.
• Cycle (projection, untested): Baseline Wire Test 30-60 days (MSA, rules of engagement, IT allowlisting, one sponsor). Annual plan: 60-120 days from first contact. Executive-cloning upsell: add 30-60 days for consent. Enterprise: 120-180 days with an RFP. First paid test is modeled in month 5, first annual plan in month 6.
• Who they trust: their broker, their auditor, their bank's treasury-management officer, and peers (CFO roundtables, AFP chapters).

3. Solution and offer

The drill

1. Intake and rules of engagement. Sponsor (CFO or Controller) named; client-designated test vendors only; standing allowlisting runbook run once a year by client IT (M365, Google, Proofpoint, Mimecast); third-party protocol signed (below).
2. Recon (AI, public information only) on vendor relationships and process, with no credential use and no intrusion. For cloning upsells, the executive's voice is recorded at consent (a 3-minute read script), not scraped.
3. Campaign plan (AI-drafted, operator-approved) from fixed playbooks, e.g. "Vendor bank change, Friday afternoon", "New AR contact, new cell, I'm traveling". Windows are spread across the week, and one operator monitors at most 3-4 live calls at once.
4. Execution: lookalike vendor-domain thread from a pool of pre-warmed domains; live adaptive voice agent with an operator listening and a kill switch; and any real-attacker tactic the client authorizes in the rules of engagement, including caller-ID simulation of the client's own numbers (verified with our provider) or a vendor number with that vendor's written consent. Default calls use a lookalike number. Every technique is approved by the client's board or authorizing executives first. Enterprise adds a scheduled, operator-run video segment, and annual engagements can run long-term social engineering across the term.
5. Stop point: the drill ends when money would have left (bank-detail change entered, payment released for approval, or callback performed or skipped). No payment rails are touched. A one-click sponsor event logger (mobile link) timestamps each stage; an optional read-only vendor-master audit-log export confirms it. Only stages with evidence are scored. Email blocked by the client's filter is scored as control passed.
6. Debrief within 48 hours with the CFO, and a welfare-debrief template for the client's HR to use with the targeted employees. Drill results are never to be used for discipline (written into the rules of engagement).

Scorecard: two halves, labeled honestly

Deliverables, redesigned for the buyer's legal position

• Findings go to the client's counsel where the client chooses. The MSA allows engagement through outside counsel. Caution: privilege over consultant work commissioned through counsel is not guaranteed; courts have compelled production of some security-consultant reports. Counsel decides; we do not promise privilege.
• Drill Results Statement (formerly "attestation letter"): point-in-time, "a test, not a warranty", no third-party reliance, signed by the company and not the founder personally.
• Remediated and Retested statement: issued only after a passing retest.
• Reliance letters for a carrier or broker only on request, with a liability cap and hold-harmless, and only after a carrier has agreed to accept one.
• Third-party protocol: the client pre-notifies each designated test vendor with a short "authorized test" letter; a 24/7 hotline number is on the rules of engagement; impersonated executives (upsell only) get a response script in advance.

Pricing

• Quote from exposure. The Wire Room calculator (Section 10) turns the client's average wire size and the failure point found into an expected-loss figure, and that figure anchors the quote.
• Renewals are priced on new staff, new entities and retest results, not on drill count. Triggers for an extra paid retest: AP turnover, vendor-master migrations, acquisitions.
• Design partners: the first 3 clients get Core at $15k for year 1, as modeled.
• Price pressure is real. Breacher.ai sells an AI vishing simulator from $199/month with MSP white-label (24-7 Press Release). Coalition gives policyholders 15 free Curricula licenses with phishing simulation (Coalition). Our price is justified by the process scope, the client-side coordination and the CFO-grade result. Those are the labor we deliberately do not automate away.

Premium add-on: logical + physical

A wire-fraud crew tests the phone and the inbox; a real one also walks in. For Growth and Enterprise clients we offer a combined logical + physical engagement, delivered with our affiliated physical-pentest team — the founder's firm, NexOasis, already runs physical pentests for banks and financial institutions. It pairs the logical drill with a physical test of the same payment process (pretext site visits to AP, treasury or reception; badge tailgating scoped to the finance floor; dropped-device and in-person "new vendor rep" scenarios), under a separate SOW and physical rules of engagement signed by an officer, every action authorized in advance by the board or sponsoring executives, priced per site. The board readout carries one combined timeline.

What we are explicitly not

• Not security-awareness training. We test one process, money movement, end to end.
• Not deepfake-detection software, not a payments product, not breach response (that goes to the insurer's panel firms).

4. Why now

1. Cost of attack collapsed. Real-time voice cloning and live avatars are commodity tools (SoftwareSeni).
2. The landmark loss is public. Arup turned "the CFO on the video call" into a board question (TechTarget).
3. Losses keep climbing: $20.9B reported in 2025, up 26%, with BEC at $3.05B (IC3).
4. Underwriters are using deepfake controls as a lens (Newfront). No carrier has said it will credit third-party testing; our searches found none.
5. Capital is validating the category. Adaptive Security has raised ~$146.5M (Biometric Update); Doppel raised $70M at $600M+ (SecurityWeek). They also make this a crowded field (Section 6).
6. Finance is automating. AP bots and invoice LLMs add an untested attack surface (an Enterprise SOW).

5. Market size (bottom-up)

Data is thin at the exact segment boundary. Each step is labeled sourced or assumed.

The crime-insurance market size for the mid-market is not reliably published; we do not use a figure. Willingness to pay: generic social-engineering pentests sell for roughly $4k-$15k per engagement (e.g., a fixed $7,950 phishing+vishing package from RSI Security; general pentests mostly $10k-$30k per DeepStrike). The $9.5k Baseline Wire Test sits inside that range. Price acceptance is untested.

6. Competition and positioning

Honest assessment. Funded incumbents and boutiques already sell deepfake simulations, and any of them could add a "payment drill" SKU. Our edges are narrow: (1) the payment-process scope and CFO buyer, (2) the client-side coordination, evidence model and scorecard that make results defensible, (3) an owner-operator with security-testing credibility selling to finance. The benchmark dataset is now a marketing asset (a "State of the Callback" report), not a revenue line or moat claim.

Positioning line: "Awareness platforms test your people. We test whether your money can leave."

7. Business model and pricing

• Recurring share is about 71% of Year-5 revenue. That matters for valuation: one adviser notes that firms below about 50% recurring revenue get repriced toward project-services multiples (CT Acquisitions).
• Channel economics (revised):
  • Brokers: a flat education or referral fee of $1,000 per qualified meeting, paid whether or not anything is bought (modeled as ~$4k per broker-sourced logo). No revenue share. Not tied to any premium term. Subject to counsel's state-by-state review, since producer compensation, disclosure and anti-rebating rules vary (Mayer Brown; NC DOI FAQ; NY DFS opinion). Brokers also get a free Wire Readiness Review (scorecard questionnaire, no drill) to offer at renewal.
  • MSSPs: 15% of first-year revenue, 10% in years 2-3, 0% after. Or a "CFO edition" playbook and scorecard licensed to run on whatever simulation engine they already use.
  • Channel cap: channel-sourced logos are 10% of new logos in Year 1, rising to 30% by Year 5 (v1: 30-55%).
• Contract terms: annual, paid upfront or quarterly in advance; auto-renew; renewal date aligned with the client's insurance renewal where possible, so the retest result is fresh for the application.

8. Go-to-market

Sequence

1. Days 1-30, before building any platform: hand-run 5 paid Baseline Wire Tests ($7.5k-$9.5k) with off-the-shelf voice tooling and one contractor operator. Measure the real cycle time, the number of approvers, the coordination hours and conversion to annual.
2. Ask 3 commercial crime underwriters directly: "Would a dated remediated-and-retested statement change your terms or pricing?" If all three say no, the insurer story comes out of the pitch entirely.
3. The founder sells the first 15-20 deals. A closer is hired only when ARR is at least $400k and the founder has closed at least 4 annual plans in each of two consecutive quarters (base model: month 31).
4. Carrier goal for Year 1: get onto a carrier's vetted-vendor or risk-services list (e.g., Corvus offers policyholders a network of preferred security vendors (Corvus Signal)). This is distribution, not a data deal. No carrier has agreed to anything.

First 10 customers by title (targets, not commitments)

Funnel and unit economics (base, projection)

Year-3 unit economics, direct vs channel, reported separately:

The go-to-market reviewer expects $25-35k blended CAC. If that is right, payback is ~15-18 months and LTV:CAC ~3-4x. That is still viable, but it is materially worse. Our lower figure depends on the paid test covering its own cost of sale; the first 5 tests will tell us which number is closer.

9. Operations with minimal headcount

Where the labor really is

The red team was right that the attack is not where the work sits. The work is client coordination, side-effect cleanup and liability. v2 budgets these explicitly.

Delivery hours per customer per year (projection; operator at $150/h, CS at $90/h):

For comparison, v1 costed Core at $3.6k, Growth at $7.8k and Enterprise at $16k. Additional entities are costed at 60% of the first because recon and reporting are shared.

What is automated: recon drafts, scripting, synthetic vendor voices, call execution under monitoring, scoring drafts, report drafts, domain-pool warming. What stays human: campaign approval, live monitoring (no more than 3-4 concurrent calls per operator), sponsor coordination, vendor-master rollback, third-party notices, QA, debrief. Year 1 is run by hand on hosted tools: a hosted voice-agent platform, Google Workspace domains, DocuSign, and a scorecard in a spreadsheet. The platform is built only after 15-20 real drills show which steps repeat.

Vendor terms. ElevenLabs' use policy bars cloning or impersonating a living person "without their explicit and informed consent", bars facilitating "fraud, scams, or other financial deception", and allows testing of its own services only with written authorization (ElevenLabs use policy). Consent covers the first clause. Whether an authorized drill falls under the second is a judgment the vendor makes, so before launch we obtain written confirmation from two voice vendors (or an enterprise agreement naming consented security simulation) and keep a self-hosted fallback. The entry test uses synthetic, non-cloned voices, which avoids the likeness question entirely.

Telephony. Twilio rejects unverified caller IDs (error 21210), and unsigned or low-attestation calls get spam-labeled under STIR/SHAKEN (Twilio). Caller-ID simulation is a verified, consent-gated feature, not open spoofing. The client authorizes its own numbers in the rules of engagement and we verify ownership with the provider before display; a vendor number is displayed only with that vendor's written consent; otherwise calls use a lookalike number. The Truth in Caller ID Act bars display intended to defraud or cause harm, which an authorized assessment under signed RoE with the number owner's consent is not, and counsel confirms each engagement. Simulated calls may carry lower attestation or a spam label, which is realistic.

Org chart (gated, projection; base-case hire month in brackets)

Year-5 base: 5 W-2 people including the founder, plus about 2,100 contractor operator hours (roughly 1.3 FTE). The bull case reaches 8 W-2. Hiring is gated, so if the business underperforms, the hire does not happen.

Founder hours (projection, revised)

Unpaid founder hours (about 3,250 over months 1-34) are costed at $200/hour, $650k, in the time-costed returns (Section 12).

Compliance operations

• BIPA-grade consent kit for any cloned voice: written release, retention and destruction schedule, automated certificate of destruction per drill. Nine BIPA-related voice-AI suits were reported in May 2026 against companies including Google and ElevenLabs over scraped voiceprints (Biometric Update); whether a clone is a "voiceprint" is unsettled (ABA). We collect only consented samples, which is the path BIPA requires.
• No call recording by default; structured transcripts only, unless counsel approves recording for that state.
• SOC 2 via a compliance-automation platform (Vanta or Drata class) from month 6; Type I in Year 2, Type II in Year 3.

10. Technology and the prototype

Platform (built in Year 2 from observed repetition, not in month 1)

1. Campaign orchestrator: playbooks, consent windows, rules-of-engagement enforcement (refuses any number not on the approved list), week-spread scheduling and an operator concurrency cap.
2. Voice pipeline: licensed vendors with written authorization plus a self-hosted fallback; synthetic vendor voices by default; consented clones only for upsells; automated destruction.
3. Adaptive voice agent with guardrails and an operator console (listen-in, kill switch). Path to the software multiple: AI-only calls with operator review of exceptions, targeting under 1 operator hour per drill by Year 3. This is not assumed in the base model.
4. Sponsor event logger and evidence store: stage timestamps, audit-log uploads, "Evidenced by client" questionnaire.
5. Report generator: scorecard, Drill Results Statement, Remediated and Retested statement.
6. Benchmark store: anonymized, minimum cohort sizes, used for the published "State of the Callback" report.

The prototype: "Wire Room" (browser-only sales demo)

A playable AP-specialist drill (lookalike vendor thread, synthetic voicemail, chat-style adaptive caller), an instant debrief, a scorecard questionnaire with expected-loss math that becomes the quote anchor, and a campaign builder that generates rules of engagement, a consent checklist and a state recording-law lookup. Sample outputs are labeled samples. It needs no backend and touches no real data.

11. Financial plan (projection)

All figures come from model.json (v2), which is a monthly cohort model. Key assumptions: first paid test in month 5; 50% test-to-annual conversion; 20% annual logo churn (applied monthly); new-logo mix of 78/22/0 in Years 1-2 and 70/22/8 from Year 3; 10% of Core customers upgrade to Growth at each renewal; delivery costed per entity-year as in Section 9; W-2 operators supply 1,200-1,400 delivery hours a year, with the rest at contractor rates; hires gated as in Section 9; insurance and legal roughly doubled versus v1 in Years 1-2. No data licensing in the base case.

Five-year projection (base case)

Year-5 opex detail (base): founder $200k; closer $150k; commissions $107k; broker flat fees $32k; MSSP share $34k; platform and dev $200k; marketing $180k; G&A incl. SOC 2 $105k; insurance $90k; legal $65k. Year-5 COGS detail: W-2 delivery staff $485k; contractor hours $329k; tech $183k; hosting $84k.

Sensitivity (base volumes; Year-5 EBITDA / EV at 7x normalized EBITDA):

Scenarios

Use of funds ($120k launch, revised)

12. Returns (projection)

Assumes 100% founder ownership, no outside capital, pre-tax, an exit at the end of Year 5, and an exit priced on normalized EBITDA: Year-5 EBITDA excluding data licensing, less the extra cost of replacing a part-time founder with a full-time CEO (a $325k loaded CEO minus the modeled founder salary). Consideration is 65% cash at close and 35% earnout over 2 years, with the earnout counted at 75% expected achievement.

Bull EV excludes the data-licensing EBITDA ($250k in Year 5), which is treated as a separate option. If a carrier relationship is real at exit, a strategic buyer may pay for it; we do not count it.

Probability-weighted view. We used the finance reviewer's rough weights: 50% stopped by a kill gate, 30% weak, 20% base. "Stopped" averages a day-90 stop (-$161k) and a month-12 stop (-$269k). The 30% weak weight goes to the mid-case run to Year 5, because the true bear case is stopped by the gates. Result: expected net value ~$1.4M, or ~$0.9M with founder time costed. That is positive but modest; the bet is attractive because the loss is capped, not because the average is huge.

The valuation is a services valuation. Base EV is about 1.8x Year-5 revenue, in line with the 1.9x median EV/revenue for cybersecurity services deals (Aventis, below). A higher multiple has to be earned with retention above 85%, operator time per drill under 1 hour, and a strategic buyer (an awareness platform, an email-security vendor, or an insurtech that wants a pre-bind controls test).

Comparable transactions and multiples (cited; re-checked for v2)

The large deals show the buyer universe, not our multiple. We use 5x / 7x / 10x normalized EBITDA for bear / base / bull. 7x sits at the top of the $2-20M IT-services band, inside the 5-9x services range, and inside CT's band for a $3M-revenue firm with about 71% recurring revenue. 10x requires a strategic buyer.

13. Risks and mitigations

14. Legal and regulatory checklist (factual, not legal advice)

Have outside counsel confirm each item before launch.

• Authorization: MSA plus rules of engagement signed by an officer, naming targets, channels, windows, stop points, client-designated test vendors and a 24/7 hotline.
• Deliverable language: point-in-time Drill Results Statement, no third-party reliance, company signatory; capped reliance-letter template; option to engage through the client's counsel (privilege is not guaranteed).
• Executive Likeness Consent Kit (upsell only): BIPA-grade written release, voice recorded at consent, retention and destruction schedule, certificate of destruction, annual renewal when executives change. Context: Tennessee ELVIS Act (Manatt); state voice-cloning privacy activity (Bloomberg Government); BIPA voice suits (Section 9).
• Call recording: roughly 11-12 all-party-consent states including California and Florida (Rev; Kilpatrick Townsend). Default: do not record.
• AI-voice calls and TCPA: the FCC's 2024 ruling treats AI voices as "artificial" (Cooley; Orrick). Confirm applicability to consented business-line calls.
• Caller ID: consent-gated simulation only. Client-owned numbers authorized in the RoE and verified with the provider; a vendor number only with that vendor's written consent; otherwise a lookalike number (Twilio STIR/SHAKEN). Truth in Caller ID Act reviewed per engagement; counsel sign-off required.
• Lookalike domains: registered in our name; client-designated vendors only; trademark review.
• Employment law: no discipline from drill results; employee-monitoring notice statutes (e.g., NY, CT, DE); HR welfare debrief.
• Broker and MSSP payments: state-by-state opinion on flat referral or education fees paid to licensed producers, compensation disclosure and anti-rebating (Mayer Brown; NC DOI; NY DFS. The NY opinion addresses producers paying unlicensed referrers, the reverse of our flow, so it is context, not an answer).
• Vendor terms: written authorization from voice and avatar vendors for consented security simulation.
• Insurance for us: a real tech E&O, cyber and GL quote with the statement wording attached, before day 30.
• SOC 2: Type I in Year 2, Type II in Year 3 (a prerequisite for Enterprise).

15. 90-day launch plan and kill criteria

Kill criteria

• Day 90: fewer than 2 paid tests booked, or counsel cannot produce a workable consent and recording framework, or E&O cannot be placed on workable terms: stop (modeled loss ~$161k).
• First 5 paid tests: fewer than 2 convert to an annual plan: stop or pivot to a one-off test business.
• Underwriter check: if all 3 underwriters say a retest statement would not change terms, remove the insurer story from the pitch (not a stop).
• Month 12: ARR below $100k or fewer than 4 annual logos: stop or sell the IP (bear-case loss ~$269k; the mid-case is also caught here).
• Month 18: ARR below $250k: no further hires; pursue a sale or white-label deal.
• Hiring gates (Section 9) mean no fixed salary is committed before the business earns it. Modeled maximum loss before the month-18 gate: ~$330k; planning ceiling $500k.
• Any time: a drill causes a real payment or a serious legal claim without a clear fix: pause new sales until it is resolved.

16. Domain names

1. holdthewire.com
2. wirehold.com
3. holdwire.com
4. wiredrill.com
5. callbackcontrol.com
6. stopthewire.com

17. Red team and response

Three internal reviewers (finance; sales and go-to-market; lean operations) stress-tested v1. The finance reviewer confirmed that v1's internal arithmetic tied out (ARR, founder cash-in of $237.9k, $28.2M EV, 143x MOIC) and that the IC3 2025 figures were accurate. The problems were in the assumptions. Net effect: base-case Year-5 EV fell from ~$28.2M to ~$5.6M. That is lower than any single reviewer's estimate ($10-14M from finance, $15-25M from go-to-market), because v2 applies all three reviews together rather than one at a time. The thesis that survives: a modeled peak cash need of ~$289k, a gated downside of ~$330k, ~$1.1M of owner cash flow in Year 5, and a ~$20M bull case.

v1 vs v2, base case, Year 5 (projection):

Finance review

Sales and go-to-market review

Lean-operations review

What would change our mind

• Upward: if the first 5 paid tests convert at 60% or more and Year-1 renewal intent is above 85%, the sensitivity table puts Year-5 EV at $7-9M on base volumes, before any software-multiple re-rating.
• Downward: if conversion falls below 35% or E&O is unplaceable, stop at the gate. That is a ~$160-270k lesson, not a $700k one.

Sources

• FBI IC3 2025: https://www.ic3.gov/AnnualReport/Reports/2025_IC3Report.pdf ; https://spycloud.com/blog/fbi-internet-crime-report-2025/ ; https://www.theregister.com/2026/04/07/cybercrime_losses_reach_alltime_high/ ; https://abnormal.ai/blog/ai-cybercrime-ic3-report-2025
• AFP 2025: https://www.financialprofessionals.org/about/learn-more/press-releases/Details/survey-79-percent-of-organizations-were-victims-of-attempted-or-actual-payments-fraud-activity-in-2024 ; https://truist.com/content/dam/truist-bank/us/en/documents/info/cci/2025-afp-payments-fraud-control-survey-report-key-highlights.pdf
• Arup: https://abc17news.com/money/cnn-business-consumer/2024/05/16/british-engineering-giant-arup-revealed-as-25-million-deepfake-scam-victim/ ; https://www.techtarget.com/cybersecurity/news/366651907/Beyond-Arup-CISOs-combat-escalating-deepfake-fraud
• Deepfake statistics: https://ironscales.com/news/ironscales-report-finds-over-half-of-organizations-reported-significant-revenue-losses-attributed-to-deepfakes-in-the-past-year-with-average-losses-of-over-280k ; https://programs.com/resources/deepfake-stats/ ; https://deepstrike.io/blogs/deepfake-statistics-2025 ; https://www.softwareseni.com/how-deepfake-fraud-became-a-five-dollar-subscription-service/
• Insurance: https://www.amwins.com/resources-and-insights/market-insights/article/how-cyber-and-crime-insurance-policies-respond-to-social-engineering ; https://www.insurancebusinessmag.com/us/news/claims/wire-fraud-exposes-the-gaps-between-cyber-and-crime-coverage-595126.aspx ; https://www.bankdirector.com/?p=5427 ; https://www.newfront.com/blog/deepfake-claims-create-a-new-lens-for-underwriting ; https://www.corvusinsurance.com/corvus-signal ; https://www.coalitioninc.com/blog/curricula-security-awareness-training/
• Competitors: https://www.biometricupdate.com/202512/new-series-b-round-brings-adaptives-total-capital-raised-to-146-5-million ; https://www.adaptivesecurity.com/blog/deepfake-simulation-exercises ; https://versprite.com/cybersecurity-listings/offensive-security-services/deepfake-as-a-service/ ; https://www.securityweek.com/doppel-raises-70-million-at-600-million-valuation/ ; https://www.24-7pressrelease.com/press-release/539997/press-release-oses-ai-vishing-simulator-launch ; https://www.24-7pressrelease.com/press-release/532496/breacherai-adds-live-deepfake-video-avatars-to-its-ai-social-engineering-simulation-platform ; https://www.vistaequitypartners.com/news/knowbe4-to-be-acquired-by-vista-equity-partners-for-4-6-billion/
• Pricing and market: https://store.rsisecurity.com/?p=4126 ; https://deepstrike.io/blogs/pen-testing-faqs-2025 ; https://apolloacademy.com/wp-content/uploads/2024/07/070724-Chart.pdf ; https://census.gov/data/tables/2022/econ/susb/2022-susb-annual.html
• Multiples: https://aventis-advisors.com/cybersecurity-valuation-multiples/ ; https://www.feinternational.com/blog/cybersecurity-business-valuation ; https://www.ctacquisitions.com/guides/mssp-ma-multiples-2026/ ; https://soferadvisors.com/insights/blog/how-to-value-a-cybersecurity-company-arr-nrr-and-2026-comps/ (not independently confirmed) ; https://www.securityweek.com/travelers-to-acquire-cyberinsurance-firm-corvus-for-435-million/ ; https://www.bankinfosecurity.com/proofpoint-to-buy-tessian-to-infuse-email-protection-ai-a-23432
• Vendors and telephony: https://elevenlabs.io/nl/use-policy ; https://www.twilio.com/docs/voice/trusted-calling-with-shakenstir ; https://www.twilio.com/en-us/blog/shaken-stir-call-flows
• Legal: https://www.biometricupdate.com/202605/tech-giants-sued-under-bipa-over-voiceprints-used-to-train-ai ; https://www.americanbar.org/groups/litigation/resources/newsletters/class-actions-derivative-suits/voiceprints-ai-bipa-new-trends-biometric-privacy-litigation/ ; https://news.bgov.com/in-house-counsel/state-privacy-laws-poised-to-fight-unauthorized-voice-cloning ; https://www.manatt.com/insights/newsletters/client-alert/tennessees-elvis-act-expands-publicity-rights-for ; https://www.cooley.com/news/insight/2024/2024-02-15-fcc-ai-generated-robocalls-illegal-under-the-tcpa ; https://www.orrick.com/insights/2024/02/Did-the-FCC-Really-Ban-AI-Generated-Voice-Calls-No-But-Restrictions-Apply ; https://www.rev.com/blog/productivity/phone-call-recording-laws-state ; https://ktslaw.com/en/Blog/GlobalPrivacy-and-CybersecurityLaw/2024/7/Wiretap-Laws-in-the-United-States ; https://www.mayerbrown.com/de/insights/resource-centers/insurtech/anti-rebating-laws ; https://ncdoi.gov/documents/agent-services/referral-fees-faqs/open ; https://www.dfs.ny.gov/insurance/ogco2005/rg050416.htm