# Sovereign Desk — Go-To-Market **Offering:** private on-premise AI workspace for Colorado law and CPA firms **Operator:** Mike Rodgers, solo, Denver CO **Motion:** founder-led, outbound-first, in-person close **Built:** 2026-02 · this document is the operating plan, not a pitch deck --- ## 1. ICP ### Primary ICP — "the 6-attorney firm with an institutional client" | Dimension | Definition | |---|---| | Entity | Colorado law firm, 4–15 attorneys, or a CPA firm 5–25 people with a tax-controversy or advisory desk | | Revenue | $1.5M – $8M | | Location | Denver metro, Boulder, Colorado Springs, Fort Collins (drivable — install is physical) | | Practice mix | Transactional, employment, real estate, tax controversy, trusts & estates, PE/M&A support | | Tech posture | Already pays for one cloud AI seat product (CoCounsel, Lexis+ AI, Clio Duo, Harvey-adjacent) **or** has explicitly banned AI use in writing | | Decision maker | Managing partner or firm administrator. One person, one signature, no procurement | | The tell | They have at least one client who sends a **vendor security questionnaire** — PE-backed portco, hospital system, bank, or a public company subsidiary | ### Secondary ICP — "the CPA practice with a 7216 problem" CPA firms are the softer sell and the faster close. IRC §7216 makes disclosure of taxpayer return information a criminal matter, and the profession is already trained to think about where data sits. Smaller ACV, shorter sales cycle, better referral velocity into their attorney relationships. ### Anti-ICP — do not pursue - Solos under 3 people with revenue under $600K — cannot carry the setup fee - Firms outside a 90-minute drive — install economics break, support is worse - Litigation-only shops whose AI need is research, not review — that is Westlaw's job, not this - Anyone who opens with "can you just do it in the cloud for cheaper" — wrong buyer, wrong product --- ## 2. Trigger events The product is not bought on a schedule. It is bought in the two weeks after one of these happens. Every outbound motion below is built to intercept one. | # | Trigger | How you detect it | Why it converts | |---|---|---|---| | 1 | **A client sends a vendor security questionnaire** | Ask directly in email #1. Also: firm announces a new institutional/PE client on LinkedIn | They cannot answer question 14 ("where is data processed and by whom") about their AI tool. This is the sharpest wedge that exists | | 2 | **Malpractice renewal** | Renewal season clusters in Q4 and at fiscal year ends. Carriers now ask about generative AI use | Filling out an AI-use question with "cloud vendor, cannot audit" is uncomfortable in a way that creates budget | | 3 | **Legal AI seat renewal / price increase** | Watch for LinkedIn complaints about renewal quotes; ask "what's your per-seat renewal come to?" | A 9-seat firm renewing at $225/seat sees $24K/yr for something they don't own | | 4 | **The firm hires its first IT or ops manager** | LinkedIn job change alerts: "Firm Administrator", "Director of Operations", small CO law firms | New ops hire needs a first win, has a mandate to fix things, and reads a security posture as a project | | 5 | **A public AI confidentiality incident** | News cycle. There will be another one | Two-week window where partners will take the meeting | | 6 | **The firm opens a second office** | CO SOS filings, LinkedIn posts | Data now moves between locations; someone has to think about it | | 7 | **A partner publicly posts about AI ethics** | LinkedIn engagement on ABA Formal Opinion 512 / Colo. RPC 1.6(c) content | They are already thinking about the problem. Comment first, DM second | | 8 | **A CPA firm picks up a §7216-sensitive engagement** | COCPA event attendance, practice announcements | Criminal-liability framing does the selling | --- ## 3. Cold email sequence Rules of engagement: plain text, no images, no tracking pixels, no attachments, sent from a real person to a real person. Sequence runs 15 business days. Stop on reply. Everything below is written to be sent as-is. > **Gate-D note:** nothing in this section has been sent. Instantly remains disarmed. Sending > requires typed approval per RIG doctrine. --- ### Email 1 — Day 0 · the question they can't answer **Subject:** question about your AI vendor's data path Hi {{first_name}}, Quick question rather than a pitch. When {{firm_name}} runs a client document through whatever AI tool you're using, can you tell a client — in writing — which company processed it and where the servers are? Most firms I talk to can't. Not because they're careless, but because the honest answer lives in their vendor's subprocessor list, which changes without notice. I build the alternative: an inference appliance that sits in your office. Document review, intake triage, and drafting all run on it, and the software logs every request so you can prove nothing left the building. Worth 20 minutes? Mike Rodgers Denver, CO --- ### Email 2 — Day 3 · the specific clause **Subject:** re: question about your AI vendor's data path {{first_name}} — Adding one specific thing, because the general version of this is easy to ignore. Here is the data-processing language from a real vendor MSA I reviewed last month: > "Vendor may process Firm data on infrastructure operated by third-party subprocessors, including > cloud hosting providers located outside the State of Colorado. Vendor may use Firm data, in > de-identified form, to improve its products and models." That firm was two days from signing. The clause was section 5, page 3. If you want, send me the MSA from your current AI vendor and I'll mark it up for free — no meeting required. I do this to make a point, and the markup is useful either way. Mike --- ### Email 3 — Day 7 · the money **Subject:** what your AI seats renew at {{first_name}}, Different angle. Per-seat legal AI runs $180–$250 per user per month. At {{seat_count}} seats that's roughly ${{annual_seat_cost}} a year, and the renewal goes one direction. My pricing is $24,500 to install and $2,400 a month, hardware included and yours. Year one is roughly a wash for a nine-person firm. Year two you're ahead about $31,500 — and the appliance still works if you ever stop paying me, because it's your machine. I'm not going to argue that owning it is cheaper on day one. It isn't. It's cheaper on day 400, and it's the only version where you can answer a security questionnaire without forwarding it. Want the one-page math for your headcount? Mike --- ### Email 4 — Day 11 · the proof offer **Subject:** I'll pull the plug during the demo {{first_name}} — Standing offer, since talking about this is less useful than watching it: Send me one contract and one intake email. On a 20-minute call I'll run both through a live appliance and you'll see three things at once — the finding list with the exact quoted clauses, the Colorado limitations clock on the intake, and a request ledger showing zero bytes to the public internet. Then I unplug the WAN cable and run it again. It keeps working. That's usually where the conversation actually starts. If the first document doesn't surface something you care about, I'll say so and we're done. Mike Denver, CO --- ### Email 5 — Day 15 · the close-the-file **Subject:** closing the loop {{first_name}}, I'll stop here — you're clearly busy and I'd rather not be the fifth vendor in your inbox this week. One thing to keep, whether or not we ever talk: ABA Formal Opinion 512 puts the duty to understand where client data goes on the lawyer, not the vendor. If you're using a cloud AI tool, the two questions worth asking your rep in writing are (1) name every subprocessor that can touch our data and (2) what happens to it at contract termination. Keep the answers in the file. If you ever get a client security questionnaire you can't answer, that's the day to call me. I'll leave the door open. Mike Rodgers · Denver, CO · {{phone}} --- ## 4. LinkedIn posts Three posts, written to be published as-is. One per week. Comment replies are the actual channel — the post is the excuse. > Drafts were generated on the local fleet (blackwell, Qwen3-Coder-30B) and then rewritten by hand, > because the model versions invented client counts and war stories that never happened. Numbers > below are either arithmetic or cited. --- ### Post 1 — the duty is yours Your AI vendor's promise is a contract term. It is not a control. ABA Formal Opinion 512 is clear about where the duty sits: before a lawyer puts client information into a generative AI tool, the lawyer — not the vendor — has to understand where that information goes and who can access it. Colo. RPC 1.6(c) says the same thing in Colorado's own words: reasonable efforts to prevent unauthorized disclosure. "We don't train on your data" does not satisfy that. It is a sentence in an agreement you cannot audit, enforced by a company you have no visibility into, subject to a subprocessor list that can change without telling you. There is a version of this where the answer is boring: the machine that reads the document is in your office. You can point at it. The software logs every request it makes, and you can hand that log to your carrier. I build that version. Not because cloud AI is evil — because "I can prove it" is a materially different sentence than "they told me." If you have a client security questionnaire sitting in your inbox that you can't answer honestly, that's the conversation I'd want to have. --- ### Post 2 — the arithmetic A nine-person firm paying $225 per seat for cloud legal AI spends $24,300 a year. Over five years, assuming the renewal never rises — it will — that's $121,500 for software the firm does not own, cannot inspect, and cannot run if the vendor changes its terms or gets acquired. Here's the other column. An on-premise appliance for the same nine people: $24,500 to install and $2,400 a month. Year one, $53,300. Year two onward, $28,800. Year one you're behind. Year two you're ahead about $31,500 versus the seats plus the overflow document-review spend most firms of that size are also carrying. And the hardware has a seven-year life, so year six is the same $28,800 while the seat price has gone up twice. I'm not claiming this is free money. Owning infrastructure never is. I'm claiming that when you spend $121,500 over five years, you should get an asset at the end of it, and the ability to answer where your client's file was processed. Happy to run the math for your actual headcount. It takes about four minutes and I'll send it whether or not you buy anything. --- ### Post 3 — the clause I reviewed a vendor MSA for a small Colorado firm last month. Section 5, page 3: "Vendor may process Firm data on infrastructure operated by third-party subprocessors, including cloud hosting providers located outside the State of Colorado. Vendor may use Firm data, in de-identified form, to improve its products and models." Section 6 required the firm to indemnify the vendor, and there was no limitation of liability anywhere in the agreement. Section 9 put governing law in Delaware. The insurance limit was a blank line — literally $[____]. Nobody was being sneaky. The clause was in plain English and the demo had been good, and the document was going to get signed because reading twelve pages of vendor paper is the least billable thing a partner does all week. That is the actual failure mode. Not malice — a page-3 clause and a busy Tuesday. The first-pass version of this review is not hard to automate. Governing law, uncapped indemnity, one-way indemnity, auto-renewal notice windows, blank placeholders — those are rules, not judgment, and a machine can flag them in about 15 milliseconds with the exact sentence quoted. Then a human decides. That ordering matters. If you want the checklist I use, comment and I'll send it. No opt-in. --- ## 5. Pricing and the ROI math Full detail in `PRICING.md`. Summary for the sales conversation: | Tier | Setup | Monthly | Seats | Fits | |---|---|---|---|---| | Solo | $9,500 | $850 | 5 | 1–3 attorneys/CPAs | | **Firm** | **$24,500** | **$2,400** | **15** | **4–12 attorneys — the target** | | Vault | $48,000 | $4,800 | unlimited, 2 offices | multi-office, security mandates | ### The ROI conversation, nine-person firm **What they spend today (verifiable with their own invoices):** | Line | Annual | |---|---| | Cloud legal AI, 9 seats × $225/mo | $24,300 | | Overflow document review, 2 engagements/qtr × $4,500 | $36,000 | | **Hard cost in scope** | **$60,300** | **What Sovereign Desk costs:** | | Year 1 | Year 2 | Year 3 | |---|---|---|---| | Setup | $24,500 | — | — | | Monthly × 12 | $28,800 | $28,800 | $28,800 | | **Total** | **$53,300** | **$28,800** | **$28,800** | | **Net vs. $60,300** | **+$7,000** | **+$31,500** | **+$31,500** | Three-year swing: **+$70,000**, before counting a single recovered hour. **The capacity argument — use second, never first.** A partner who spends four hours a week on first-pass contract review and recovers half of that recovers ~92 hours a year. At a $385 realized rate that is ~$35,000 of capacity per partner. Say it once, do not lean on it — the buyer discounts soft numbers and the hard numbers are already sufficient. **The argument that actually closes:** one institutional client retained because the firm could answer the security questionnaire. One mid-size PE portco engagement is worth more than five years of this product. --- ## 6. Top 3 objections and the answers ### Objection 1 — "A local model can't be as good as GPT-5 / Claude / CoCounsel." **Answer:** Correct, and it doesn't need to be for this job. Two things are happening in the product. The first is a deterministic rules engine — governing law, auto-renewal notice windows, uncapped indemnity, blank placeholders, missed deadlines. That is regular code with the quoted sentence attached, it runs in about 15 milliseconds, and it is not a model at all. It doesn't hallucinate because it doesn't generate. The second is a 30-billion-parameter model doing a second read. For issue-spotting on a lease or an MSA, it is genuinely good. For novel appellate reasoning it is not, and I will not pretend otherwise. Keep Westlaw. This replaces the AI add-on you pay for on top of Westlaw, not Westlaw. And you can test it before you buy — bring your own document to the walkthrough. I would rather lose on a real document than win on a slide. --- ### Objection 2 — "We can't afford $24,500." **Answer:** Then you probably shouldn't buy the Firm tier, and I'll tell you that on the call. But run the number against what you're already spending. Nine seats of cloud AI is $24,300 a year, every year, forever, and you own nothing at the end. The setup fee is roughly one year of a bill you are already paying — the difference is that year two costs you $28,800 instead of $24,300 plus whatever the renewal increase is, and you have an asset. If the cash timing is the real problem, say so. Solo tier is $9,500 and can be upgraded by swapping the appliance, and I will do install in two payments across two quarters for firms I want to work with. What I will not do is a per-seat subscription, because the entire premise is that you own it. --- ### Objection 3 — "What happens when you get hit by a bus?" **Answer:** The fairest question a solo operator gets, and the reason the architecture is what it is. Three concrete protections. First, there is no license server and no kill switch — the appliance runs the last installed build indefinitely with zero contact with me or anyone else. If I disappear tomorrow, your software works Monday. Second, the stack is open: standard hardware, open-weight models, and a plain HTML/JS application with no build step. Any competent IT contractor can maintain it, and I hand over the runbook and the source at install, not at some future exit. Third, the escrow clause is in my standard agreement — full source, model weights, and documentation, released on 30 days of unresponsiveness. Compare that to the cloud vendor: if they get acquired or sunset the product, your workflow stops that day and your data is in someone else's migration plan. --- ## 7. First 10 prospect types, and exactly where to find them Apollo's key on this account is on a Free plan — `mixed_companies/search` and `organizations/enrich` both return 403/422, so no list was pulled programmatically. Every source below was checked live and returns HTTP 200. Sourcing is manual, which for 40 accounts is fine. | # | Prospect type | Why they buy | Where to find them (verified live) | |---|---|---|---| | 1 | **Denver mid-size transactional firm, 6–15 attorneys, PE/M&A support work** | Their clients send security questionnaires; they review vendor paper all day | `cobar.org` → Business Law Section member listings; CO SOS entity search at `sos.state.co.us/biz/BusinessEntityCriteriaExt.do` for "LLP" + Denver | | 2 | **CPA firm with a tax-controversy desk, 5–25 people** | IRC §7216 criminal exposure on return information; CP2000/audit volume is seasonal and crushing | `cocpa.org` member directory + COCPA chapter events | | 3 | **Employment-side law firm, 3–10 attorneys** | Intake volume with hard 300-day CADA/EEOC windows; missed-date malpractice is their nightmare | `cobar.org` Labor & Employment Section; Colorado attorney search at `coloradolegalregulation.com/attorney-search/` | | 4 | **Trusts & estates practice, 2–8 attorneys** | Highest-sensitivity documents in the profession; wealthy clients ask where files live | `denbar.org` Trust & Estate Section; Boulder and Cherry Creek concentration | | 5 | **Firm that just hired a Firm Administrator or Director of Operations** | New ops hire with a mandate and a budget line | LinkedIn Sales Navigator: title = Firm Administrator / Legal Administrator, industry = Law Practice, geography = Colorado, "changed jobs in last 90 days" | | 6 | **ALA Mile High chapter members** | Legal administrators are the real economic buyer in a 10-attorney firm | `milehigh.alanet.org` — chapter meetings, sponsorship is a legitimate entry | | 7 | **Boulder / Fort Collins IP and tech-transaction boutiques** | Client base is technical, will understand and respect the architecture immediately | `cobar.org` IP Section; Boulder Chamber tech listings | | 8 | **Real estate and construction firms, 4–12 attorneys** | Lease and AIA-contract review is high-volume and rules-heavy — the rules engine demos itself | CO SOS entity search; Denver Metro Commercial Association of REALTORS® vendor lists | | 9 | **Firms serving healthcare practices (dental/ortho/vet DSO roll-ups)** | BAA obligations flow downhill; adjacent to Mike's existing dental practice book | Existing RIG dental relationships → warm intro to their transaction counsel | | 10 | **Any firm whose partner engaged with an AI-ethics post in the last 30 days** | Already thinking about it; comment-first outreach converts 5–10× cold | LinkedIn search: posts mentioning "Formal Opinion 512" or "RPC 1.6" filtered to Colorado; engage in comments before any DM | ### Sourcing sequence (week one) 1. Build a 40-account list from sources 1–4 above. Firm name, domain, managing partner, headcount, practice mix, and the single trigger you believe applies. 2. Verify each managing partner is currently licensed at `coloradolegalregulation.com/attorney-search/`. Costs nothing, prevents an embarrassing first email. 3. Send Email 1 in batches of 10/day from a real mailbox. Never blast. **Requires Gate-D approval.** 4. Every reply gets a same-day answer, and the ask is always the same: one contract, one intake email, 20 minutes. ### Targets for the first 90 days - 40 accounts sourced, 40 sequences started - 8–12 replies (20–30% is realistic for founder-sent, hyper-specific plain text at this volume) - 5 walkthroughs booked - **2 installs closed** — $49,000 setup + $4,800 MRR That is the whole plan. Two closes in 90 days makes this a real line of business; zero replies in 40 accounts means the trigger thesis is wrong and the ICP needs to move, not the copy.