The 90-second discovery dossier (when you don't have an SDR)
Share
Sales teams have an SDR. The SDR's job, among other things, is to spend twenty minutes researching every prospect before the discovery call so the rep walks in already knowing the company's revenue range, recent funding, recent hires, and the specific pain points that are most likely to map to the offering. By the time the rep is on the call, the prep is done.
Solo consultants do not have an SDR. They have a calendar that looks like this: a discovery call at 11 AM, an existing-client meeting at 12, a deliverable due at 5, and a fifteen-minute window between 10:45 and 11:00 to do the prep that the SDR would have done. So the prep doesn't happen. They walk into the call cold, ask generic discovery questions, and spend the first half of the call doing the research the prospect was hoping to skip.
This is a fixable problem. The fix is a 90-second pre-call dossier produced by AI, run on every discovery call without exception, and structured around three artifacts. The math works: three minutes total, including the read-through, before any call. You walk in already knowing more about the company than half their employees do.
This article is the workflow. If you want the broader frame for which consulting tasks should go to AI in the first place, the previous article covers that. Lead research is squarely in Category 3 — structured, doesn't require your voice, ~95% AI-tractable. The default should be: AI handles this unless you spot a reason it shouldn't.
The three artifacts
The dossier has three parts. Each one is a distinct artifact with a distinct purpose. Producing them in a single Claude or ChatGPT session is the trick — the model carries context between the three prompts, so by the third prompt it already knows the company, the person, and the situation.
Artifact 1: The company brief
A one-page summary of what the company does, how big it is, where it makes money, and what's happened in the last 90 days. The job of this artifact is to make sure you don't ask any question on the call that would have been embarrassing to not already know the answer to.
The prompt:
You are my research assistant. I have a discovery call with [PROSPECT NAME]
from [COMPANY]. Their public site is [URL]. Produce a 1-page company brief.
Sections:
1. **What they do** — 2 sentences. Include who their customers are.
2. **Size signals** — employee count (LinkedIn or public), revenue range
if discoverable, age of company, funding if VC-backed.
3. **Where they make money** — primary product/service line.
4. **Recent moves** — anything in the last 90 days: launches, hires,
layoffs, press, fundraises. Cite the source URL.
5. **Anything publicly weird** — technical, strategic, or cultural
signals worth knowing. If nothing, say "nothing weird."
6. **Three questions I should ask on the call** — specific to *this*
company, not generic discovery questions.
Keep it factual. If you can't verify something, say so — don't invent.
Cite URLs for non-obvious claims.
The "anything publicly weird" line is the one that earns its keep. It pushes the model to surface the thing you'd otherwise notice in week three of the engagement. The "three questions" line at the end gives you something specific to walk into the call with that no other consultant will be asking.
A worked output for a fictional B2B food-distribution startup might surface that the company's public site claims "200+ restaurants" but their LinkedIn job posts reference "growth from 80 to 250 SKUs." That's a SKU count, not a customer count — either the public number is aspirational, or it's stale, or "200+ restaurants" means something different than you'd assume. Three minutes of generation got you a question worth twenty minutes of careful manual research. The "publicly weird" callout is the kind of thing only a human researcher would normally spot, and only if they were looking carefully.
Artifact 2: The decision-maker brief
A one-page brief on the specific person on the call. Not the company — the human. Their path, what they probably care about, how they communicate, and one specific opener you could use that shows you did your homework.
The prompt:
The person on the call is [NAME], [TITLE] at [COMPANY]. Their LinkedIn
is [URL] (or copy-paste their profile here). Produce a 1-page brief.
Sections:
1. **Their path** — 2-3 sentences on how they got to this role. Note
any prior roles relevant to my offering.
2. **What they probably care about right now** — based on title +
industry context + any recent posts. Be specific.
3. **Communication style** — based on how they write (LinkedIn posts,
bio, comments). Direct? Hedged? Technical? Salesy?
4. **One opener** — a 1-2 sentence opener I could use in the first
minute that shows I did my homework. Not a flattering one — a
substantive one.
If I haven't given you their LinkedIn, say so and ask for it.
The opener field is what changes the call. Most discovery calls open with a generic "thanks for taking the time, tell me about your business." Replace that with "I noticed you wrote a piece on X six weeks ago — is that the same thing you're trying to solve here?" and the conversion lift is real. The prospect immediately registers that this is a different kind of consultant than the last six they talked to.
The substantive-not-flattering instruction matters. AI defaults to "I love what you're doing at..." which sounds like every cold pitch. Tell it to find an actual hook from their public output, not a compliment.
Artifact 3: The fit flag
This is the one most consultants skip, and the one with the highest leverage. Before the call happens, decide whether this is a deal worth proposing for at all. The flag's job is to save you the 30 minutes plus 60 minutes of follow-up on calls you shouldn't be taking.
The prompt:
Based on the company brief and decision-maker brief above, give me a
fit flag for whether to do this discovery call.
My ICP:
[PASTE YOUR ICP — see Proposal-Closer Prompt Pack Prompt 02 for a
template, or paste the criteria you already use]
Output:
- **Verdict:** GREEN / YELLOW / RED
- **Confidence:** 0-100%
- **Reasoning** — 3 bullets, each citing the specific evidence
- **If YELLOW:** the 1-2 questions to ask in a follow-up email before
the call, to upgrade or downgrade the verdict
- **If RED:** a 3-sentence polite decline I can send
Be honest. RED on this is not failure — it's the system working.
A GREEN verdict doesn't mean the deal will close. It means it's worth the 30 minutes. RED on 25-35% of inbound is healthy — that's the system filtering out the calls you'd otherwise spend two hours on (prep + call + follow-up + thinking-about-it-afterward) for a 0% close rate.
The "polite decline" output is a real artifact. AI is bad at writing the cold-email-decline that doesn't sound dismissive; this prompt is the exception, because it has the brief context to write something specific.
Running the three together
The whole sequence is meant to happen in a single Claude or ChatGPT session. Same tab, three prompts, top to bottom. The model maintains context between them — by the time you run the fit-flag prompt, it already has the company brief and the decision-maker brief in working memory and is reasoning across both.
Total wall-clock time, after you've done it twice and have the muscle memory: about three minutes. Two minutes of generation + one minute of read-through. You walk into the call with the dossier open in a second tab, glance at the "three questions" list, and you're prepared in a way the prospect won't expect.
If your process is to take the dossier output and save it to a clients/<prospect-name>/dossier.md file, you also have the source material for the proposal you'll write later — the same facts you used in the call inform the proposal that follows it.
A pitfall worth flagging
AI will confidently report that the company has "$12M in ARR" when no public source has ever said so. Numerical claims are the highest-risk hallucination zone for this workflow. Always treat any number AI gives you as suspect unless it cites a verifiable source URL. The "cite the source URL" line in the prompt helps but doesn't eliminate the risk; spot-check the numbers that would change your behavior on the call.
This is also the workflow where ChatGPT with browsing edges out Claude — the live-search advantage matters more for fresh research than it does for writing. If you're on Claude without web access, ask the model to flag any claim it can't verify; it will, and you can fill the gaps from a manual five-minute search.
What to do this week
You probably have a discovery call on the calendar. Run the three prompts before it. Take the dossier into the call. Notice what you'd have missed.
Most consultants come back from this with one observation: the prep made the call qualitatively different — the prospect noticed, and the conversation went somewhere the cold version wouldn't have. That's the unlock. Ten or fifteen of these in a quarter, and your conversion rate moves visibly without anything else changing.
Going deeper
This article distills the pre-call workflow from Chapter 2 of The Solo Consultant's AI Playbook, which covers the full lead-research workflow including the worked example with a fictional engagement. The chapter also pairs with the Proposal-Closer Prompt Pack, which picks up where this article ends — turning the discovery call into a signed proposal.
The frame for which consulting tasks should go to AI at all (Category 1 / 2 / 3) is in the first article. Lead research is squarely Category 3, which is why the workflow above is so AI-tractable in the first place.
— Digital Kreative