From Discovery Call to a Demo the Buying Group Remembers

Five moves, in the order the information arrives. Each decision is yours; each tool only saves drafting. The debrief skill earns the most. The buyer group map is your own list until the champion corrects it, so a blank org chart and one conversation can replace it. Keep the script builder for its clock.

Step 1. Map who decides before the first question

About this example. Lyzr is a real company. Lyzr has not reviewed this guide. This is a worked example on public facts and illustrative inputs, not a client result. Facts about Lyzr come from public sources, linked alongside with the date I checked them. The bank, and every number, capability, customer type or product detail marked illustrative, is an input I chose to show the method; it is not Lyzr's data, and the tool outputs are exercise results, not Lyzr's real strategy.

What is public: Finovate describes New Jersey-based Lyzr as an agentic AI infrastructure platform and names workflows such as loan servicing and origination. Lyzr says it "supports both cloud and on-premise deployments". So ask early whether a data or AI team owns this, and where it would run: on-premise, borrower documents may never leave the bank.

So the exercise: if you ran go-to-market at a company like Lyzr, how would you take one US regional bank from a discovery call to a demo? The bank is illustrative: about 2,000 employees, about $15 billion in assets, commercial lending, vendor-hosted, a deal of $150,000 to $250,000 a year.

Input to the buyer group analyzer: product "Fintech: AI agent platform for banking operations: loan onboarding, document checks and servicing workflows"; company Lyzr; business_model saas; deal size "$150K-250K"; company size "US regional bank, 1,000-3,000 employees"; champion VP Lending Operations; stakeholders, in order: Chief Operating Officer, Chief Information Officer, Chief Information Security Officer, Head of Model Risk Management, Loan Operations Manager.

How to read it: the map starts from your titles and list order, and it cannot read who really holds the budget. So the COO as economic buyer is my input, not a finding. The CIO stayed under "Also in the group" while "Head of IT" was added as technical evaluator. The CIO's team owns the systems the agent touches, so move the CIO, or a delegate, there. Model risk sits in that group as "review", so treat it and the CISO as reviewers.

The missed roles sit under the map, including "Chief Financial Officer (budget or sign-off)" and "Risk and Compliance Lead (review)", with "Ask your champion whether they are involved". Add a head of data and AI yourself, since Lyzr is infrastructure. All three go into Step 2 as a gap.

Step 3. Debrief the call the same day, then plan the reviews

I ran the Meeting Prep and Debrief skill from the B2B Sales Enablement plugin on illustrative notes from a call with the VP and a Loan Operations Manager, and asked for the debrief and the demo prep. The notes carried the Step 2 pains and numbers and added: the COO owns the automation budget and wants a cost per loan figure; the CIO's team must approve any write access to the loan origination system; model risk took 14 weeks to review a chatbot last year; an OCR tool failed two years ago because exceptions were not routed. Agreed next step: a 45 minute demo in two weeks with the VP, the Loan Operations Manager, an Enterprise Architecture lead and someone from model risk. The COO was not confirmed. The 14 weeks is one illustrative review, not a benchmark.

What came back: a debrief, seven action items, a follow-up email and a demo brief. Keep this output nearly whole, because it says why each point matters to each person: "An exception file first. Why it matters to the Loan Operations Manager: the OCR tool failed on exceptions; show the routing to a person before the happy path." Fill in its "[Verify before meeting]" and "[User to add: ...]" flags.

The COO as budget owner is the VP's report, not a fact. The skill calls the VP a "champion candidate" and says "test it by asking the VP to bring the COO". Run that test. If the VP will not, do not plan on that budget.

Two lines disagree. The email asks for the COO "for the last 15 minutes"; the action item says "after the demo". Prefer the separate slot, with the VP presenting, because minute 30 is mid-demo. If that window is all you get, the cost step still lands inside it. The email also says "today" but is due "Mon 2 Nov 2026", so send it the same day.

The action item asks for "10 redacted sample loan files". My reading: borrower documents leaving a bank before a contract will usually need a yes from information security or legal. Ask the VP who approves it first, and build a synthetic file to the manager's field list by Wed 11 Nov.

Now take the reviewers out of the demo's job. They come to find out whether to say no, and they work from documents. Make the first look for the VP, the manager and architecture. The model risk reviewer can sit in, but book a separate evidence session on accuracy, failure cases and model changes. The audit log supports that case; it validates nothing. The agreed attendees leave out the CISO and third-party risk, so start those conversations before the demo, through the VP.

Then ask for less. My suggestion, not a rule in the guidance: replace the brief's pilot with a read-only parallel run on past files of one loan type. The agent proposes, analysts key as they do today, and you compare. Write access leaves the first decision, and Model Risk gets accuracy on the bank's own files. It still needs security and third-party review, and it saves nothing while it runs.

Ask what the bank's own policy needs. The April 2026 interagency model risk guidance (SR 26-2 and the OCC's Bulletin 2026-13) says generative and agentic AI models are "not within the scope of this guidance" and is "expected to be most relevant" to banks over $30 billion in assets. The guidance generally excludes banks at or below $30 billion, but it can still be relevant where model risk is significant. For this illustrative $15 billion bank, ask the risk team which parts apply. Footnote 3 applies the principles to "non-generative, non-agentic AI models", so ask which kind each step uses. Expect a review under the bank's own policy. Last year's chatbot checklist is your best clue.

For third-party review, only the 2023 interagency guidance is final. A proposal of 15 September 2026 would replace it, with comments to 16 Nov 2026, so plan on the 2023 text and recheck then. Checked 1 Nov 2026.

The notes do not settle compliance, the CFO or a data and AI owner, so ask again.

This is the point where I would want a second pair of eyes: the first regulated-buyer reviewer meeting is coming up. Bring me your seat map, discovery gaps, reviewer criteria and proof, and we will work it through in 45 minutes.

Step 4. Script one file for the room, not the product

Follow one file through its life: the exception first, then extraction, the approval, the log of that file, and cost per loan last. "Read versus write access with approval before each write" is my wording for a control to confirm with your product team, not a claim about Lyzr.

Demo script builder inputs (illustrative)

What came back: 6 minutes opening, 6 to confirm, 25 of demo in 5 steps, 6 of discussion and 2 to close. Keep that clock.

How to read it: rewrite "Who sees what" from the brief. It lists the failed OCR tool as what every attendee "came for", and the Confirm part has the same fault. Each person's concern is in the brief's discussion points, so the tables alongside start there.

The script's "one real case" also clashes with the synthetic file. Unless sample files are approved, say instead: "This is a synthetic file built to your team's field list. Tell me where it differs from your real ones."

Replace two objection answers. On write access the script quotes your outcome as "The value point to judge it against"; use the brief's "Name the fields it writes, whether a person approves each write, where every action is logged and who can revoke access." On the 14 week review, ask what last year's review required and offer to map your evidence to it in the separate session.

The cost step came back generic, so build it with the VP beforehand. The 9 days is elapsed time, not hands-on work, so show it beside the cost, not inside the sum, and ask the VP for hands-on hours and the loaded hourly cost.

All illustrative, alongside: a file costs $600 today and $390 if it goes straight through. Exceptions killed the OCR tool, so assume 20% stay manual and save nothing. That leaves $168 a loan, $100,800 a quarter, or 1,680 hours. That is capacity, not a saving, until the bank redeploys those hours.

Review and the parallel run save nothing, so assume two quarters to go-live. Year 1 lands between $51,600 ahead of the fee and $48,400 behind it, before set-up. The soft spot is rework: if touches stay at 3, each quarter runs from $5,700 ahead to $19,300 behind. So make the parallel run measure exceptions and rework, and let the VP replace every input.

These are whole-book sums. On one loan type of an illustrative 150 loans a quarter, the same inputs give $25,200 a quarter, less than a quarter of the deal. Sell the first stage as evidence, not as payback.

1. Demoing to the champion's problem only. Name what each attendee came for.
2. Meeting the reviewer at the end. Model risk and security can set the length of the deal.
3. Asking for write access first. A read-only run takes it out of the first decision.
4. Showing a saving with no exception rate or go-live date.
5. Trusting the first role map. Check every seat with the champion.

• One B2B Deal From MEDDPICC Discovery to a CFO-Ready ROI Case and a Dated Close Plan
• How to Hold Price in Late-Stage B2B Negotiation Without Losing the Champion
• Trust-Led Enterprise Selling in Fintech: Selling Payments Infrastructure to Banks Through Every Review
• All 30 guides