← Back to all demos

Sales preparation

A new inquiry. What should you know before replying?

An inquiry arrives. Your team needs the facts, relevant context and a clear next step before the first conversation. See how the Lead Agent prepares a source-linked brief while keeping missing or conflicting information visible.

Synthetic inquiries · Recorded examples. Switching cases shows saved results and makes no model calls. Drafts await human review.

Explore three inquiries

7 facts · 3 unknowns

3 questions prepare the next conversation.

Complete synthetic inquiry, recorded brief. Draft awaits human review.

Source: lead-demo/cases.json · complete inquiry

View data table
7 facts · 3 unknowns
EvidenceValue / state
Supplied facts7
Unknowns3
Discovery questions3

Explore the outcome

One inquiry. A clearer next step.

Choose a recorded case above to see its input, decision and actual draft below.

Inquiry facts become a reviewable brief

Trace each item to its recorded input or draft. No reply is sent.

Before · The inquiry

Company, tools and budget are supplied. Saved company context matches the inquiry.

Frequency
weekly
Client-stated budget
€5,000 pilot
Compared subscriptions
3,000
+3 more inquiry facts
Review workload
2 people × 4 hours
Reported exceptions
240
Estimated staff cost
€45/hour

This is the prospect's supplied budget, not a quote from us. Sources: inquiry, saved company snapshot and service guidance.

Source records

Original inquiry · verbatim

“We reconcile 3,000 subscriptions every week using 3,000 HubSpot CRM records and 3,000 Stripe billing records. Two staff members each spend 4 hours per week on this manual work. Last week 240 subscriptions had discrepancies. Our estimated fully loaded staff cost is EUR 45 per hour. We need a read-only discrepancy report reviewed by our operations lead.”
  • inquiry-company

    Company name supplied as the prospect.

    inquiry-company · “the prospect”

  • inquiry-url

    Company website supplied.

    inquiry-url · “https://company.example”

  • inquiry-problem

    The team reconciles 3,000 subscriptions weekly using HubSpot CRM records and Stripe billing records; two staff members each spend 4 hours per week on this manual work. Last week 240 subscriptions had discrepancies. Estimated fully loaded staff cost is €45/hour. They need a read-only discrepancy report reviewed by their operations lead.

    inquiry-problem · “We reconcile 3,000 subscriptions every week using 3,000 HubSpot CRM records and 3,000 Stripe billing records. Two staff members each spend 4 hours per week on this manual work. Last week 240 subscriptions had discrepancies. Our estimated fully loaded staff cost is EUR 45 per hour. We need a read-only discrepancy report reviewed by our operations lead.”

  • inquiry-tools

    Tools supplied: HubSpot and Stripe.

    inquiry-tools · “["HubSpot","Stripe"]”

  • inquiry-frequency

    The workflow frequency supplied is weekly.

    inquiry-frequency · “weekly”

  • inquiry-budget

    A pilot budget of €5,000 is supplied; confirmation of whether this is the available or approved budget remains open.

    inquiry-budget · “EUR 5000 pilot”

  • inquiry-workload

    Workload record: 3,000 CRM records, 3,000 billing records, 3,000 compared subscriptions, 240 exceptions, two reviewers, 4 hours per reviewer, and hourly cost of €45/hour.

    inquiry-workload · “{"crmRecords":3000,"billingRecords":3000,"comparedSubscriptions":3000,"exceptions":240,"reviewers":2,"hoursPerReviewer":4,"hourlyCostCents":4500}”

After · The next step

Bring the workflow, service fit and open questions into one brief for the first call.

Company context
1 company finding
Still unknown
3 open items
Discovery questions
3 discovery questions

Review required

Inspect this brief
Source records
  • company-product

    The company source describes the prospect as a subscription analytics software provider.

    company-product · “the prospect provides subscription analytics software.”

  • service-fit

    The supplied weekly manual reconciliation is a repetitive workflow with a stated comparison volume and exception count, which could support discovery of a bounded pilot. The service guidance requires human oversight and makes no ROI or SLA guarantees; the requested read-only report reviewed by the operations lead is consistent with that oversight requirement.

    service-fit · “Start with one repetitive workflow, a clear success condition and a small paid pilot. Human oversight is required. Timelines follow discovery; no ROI or SLA guarantees.”

Source: Recorded synthetic inquiries and briefs (JSON).

The prepared brief

Prepare the discovery brief

Recorded case: Complete inquiry. Every reply remains a draft for human review.

Provisional hypothesis

Provisional workflow hypothesis: a pilot could compare the supplied HubSpot and Stripe records and produce a read-only discrepancy report for the operations lead to review. This is a hypothesis for discovery, not a claim that reconciliation is already automated or that the workflow is approved.

Reply draft · pending human review

Hello the prospect team, Thanks for sharing the reconciliation context. I understand you reconcile 3,000 subscriptions weekly across HubSpot and Stripe, with a read-only discrepancy report for your operations lead to review. A possible first step for discussion is a bounded pilot that compares the records and surfaces discrepancies for human review. Could you clarify whether €5,000 is the available or approved pilot budget? How do you define discrepancies and choose matching fields? What report format and delivery method would suit the operations lead’s review? This is a draft for local review; scope and timing would follow discovery.

From unknowns to discovery questions

Ask about the gaps before agreeing scope or using company context.

Numbers link related items in the recorded draft. Answers and human verification are still pending; a question does not settle an unknown.

Still unknown

  1. Whether €5,000 is the available or approved pilot budget.

  2. How discrepancies are defined and which fields or matching rules the comparison should use.

  3. How the operations lead wants to review and receive the read-only report.

Discovery questions

  1. Is €5,000 the available or approved budget for the pilot?

  2. How do you define a discrepancy, and which fields or matching rules should the report use?

  3. What format and delivery method would work for the operations lead’s read-only review?

Source: Recorded synthetic inquiries and briefs (cases.json). Numbered links are editorial, not recorded model decisions.

View unknowns and questions as a table
Recorded unknowns and related discovery questions
ItemStill unknownRelated question
01Whether €5,000 is the available or approved pilot budget.Is €5,000 the available or approved budget for the pilot?
02How discrepancies are defined and which fields or matching rules the comparison should use.How do you define a discrepancy, and which fields or matching rules should the report use?
03How the operations lead wants to review and receive the read-only report.What format and delivery method would work for the operations lead’s read-only review?

Review boundary

A clear next step, grounded in the inquiry. A brief enters review. It does not send itself.

Review the evidence and unknowns before choosing how to respond. The brief does not approve scope, budget or commitments.

Your team checks the facts, resolves unknowns and accepts, corrects or rejects the draft before responding. Current selected case: complete. Review status: pending.

This page displays recorded outputs; it does not operate the local queue. Production inbox and CRM connections are not part of this demo.

Facts, uncertainty and reviewer usefulness

Preserve the facts

Keep the stated problem, tools, frequency and workload linked to their sources. Technical checks verify source membership and excerpts; a reviewer checks their meaning.

Make uncertainty visible

Withhold company findings when identity is missing or conflicting. Separate what is known from provisional hypotheses.

Measure reviewer usefulness

Assess omitted facts, necessary corrections and time until the team accepts the brief. These human measurements remain pending.

The reconciliation workload in the inquiry is context for discovery. This Lead Agent does not reconcile customer relationship management (CRM) or billing records or establish that the prospect's manual workload has been reduced.

At a glance

Business question
What do we know about this inquiry, and what should we ask next?
What the workflow did
Selected permitted company context and prepared a source-linked brief, questions and a reply draft.
Recorded complete inquiry
3 discovery questions · 1 company finding. Usefulness and time savings remain unmeasured.
Human decision
Check meaning and fit, resolve unknowns, then approve or revise the reply before responding.

From inquiry to human review

Complete inquiry: company context included; review open questions.

  1. 01

    Inquiry

    7 supplied facts

  2. 02

    Evidence

    1 company finding · 3 unknowns

    Context found
  3. 03

    Questions

    3 discovery questions

  4. 04

    Reply draft

    Source checks passed

  5. 05

    Human review

    Pending · approve or revise

    Review required

Saved synthetic example; no operational action.

Sources: Recorded cases and Comparison.

View data table
Complete inquiry · recorded preparation path
StepRecorded state
Inquiry7 supplied facts
Evidence1 company finding · 3 unknowns · Context found
Questions3 discovery questions
Reply draftSource checks passed
Human reviewPending · approve or revise
Explore preparation value with your assumptions

Size the workflow

What would a useful brief be worth to your team?

Explore your assumptions for Lead Agent preparation. These editable examples are not measured savings. Include checking and correcting drafts in the assisted time, and model, hosting and maintenance costs in operating cost.

10.0 staff-hours per month · €420 capacity value after operating cost

Inquiries × (manual − assisted minutes) ÷ 60 × hourly cost − operating cost. Negative values mean the assumed process adds work or costs more than the released capacity is worth. This is not cash saved and excludes one-time implementation cost. The prospect's €360/week reconciliation workload is a separate potential project.

Technical run details

End-to-end run

22 seconds

Recorded complete inquiry, with gpt-6-luna drafting and one routing decision call.

Decision model

US$0.000132636

The decision model picks from fixed options. 3 API calls in total, one per inquiry. Reported routing cost only. AI drafting is estimated separately.

AI drafting

US$0.111788

3 drafts repriced at gpt-6-sol Standard rates using recorded gpt-6-luna token counts. Estimate, not a billed charge.

Mean routing latency

The decision model averaged 16× faster routing; it matched fewer routes (9 of 12 vs 12 of 12).

Routing only; three repeats of four labelled test inquiries per method. Excludes AI brief drafting and human review. Small test comparison. Route accuracy differs. Total workflow cost and customer savings are unknown.

Decision model (Jev)
338 ms9 / 12 matched
Generative routing
5,569 ms12 / 12 matched

Common baseline: 0 ms · bar length encodes mean time, not quality.

Source: Recorded routing comparison (JSON)

View data table
Mean routing latency
MethodMean timeMatched routes
Decision model (Jev)338 ms9 / 12
Generative routing5,569 ms12 / 12

Illustrative routing + AI drafting API total: US$0.111921 for three inquiries (about US$0.0373 each). Assumes the same token counts on gpt-6-sol; excludes hosting and human review. View the calculation

Solution architecture

How the brief comes together.

Jev is a pinned decision model (typesafe/jev-1.13, called through OpenRouter's API) that answers a few fixed-choice questions about the inquiry and picks the next lookup. Code controls source access, identity checks and workload calculations.

AI drafts from selected evidence; a person reviews the result. Highlighted path: Complete inquiry.

Input

Inbound inquiry

Problem, tools, frequency, budget and reported workload.

  1. 01 / Code

    Validate and size the problem

    Check input and company identity. Calculate hours, discrepancy rate and staff cost from supplied figures.

  2. 02 / Decision model

    Choose the next lookup

    One typed decision call: classify the need, identify missing context and choose an allowed source or clarification.

03 / Code · Permission gate

Is the lookup allowed and sufficiently clear?

Identity checks, calibrated confidence and allowed tools determine the path. Source content cannot grant permission.

Yes / Read-only lookup · Selected path

Retrieve company context

Use the saved company-site snapshot and service/fit guidance.

Company snapshotService guidance

Missing, conflict or API error

Withhold company context

Keep supplied facts and service guidance. Flag uncertainty and ask for clarification.

No unsupported company lookup.

04 / AI drafting

Prepare the discovery brief

Selected evidence and supplied facts become a dossier, 2–3 questions and a reply draft. Facts, hypotheses and unknowns stay separate.

05 / Code · Evidence checks

Do fields and source excerpts pass validation?

Check required fields, source IDs and literal excerpts. These checks do not establish whether every claim follows from its source.

Valid draft / Human review · Selected path

Review before responding

Check meaning, fit, unknowns and commitments. Decide whether to use or revise the draft.

Unsupported evidence or draft failure

Hold the draft for correction

Preserve the sources and failure trace. Correct the issue before retrying; no confident reply is released.

Boundary: saved sources and local drafts. No production intake, email, calendar or CRM writes. Switching examples highlights a recorded path and makes no model calls.

Why this approach

Small decisions. A useful brief.

The decision model chooses the next lookup

Typed decisions identify the problem, missing context and permitted lookup. The shown complete inquiry routing call cost US$0.00004502.

Code checks the action

Identity checks block unsupported company lookups. Missing information stays visible. Source IDs and quoted excerpts are validated.

AI drafts the explanation and reply

AI turns selected evidence into a brief, questions and a reply draft. Your team checks the meaning and approves the response.

The measured advantage is faster routing than the generative baseline at a small reported API cost. Total workflow cost and savings for a sales team have not been measured.

How we checked it.

Inspect the recorded timings, API usage and routing checks behind the examples below.

Model comparison, calibration and failure checks
Routing only · earlier test inquiries · 12 observations per method
MethodMean timeMatched routes
rules0 ms12 / 12
jev338 ms9 / 12
generative5,569 ms12 / 12
Jev API usage · shown numeric inquiry recordings
InquiryAPI callsReported cost
Complete inquiry1US$0.000045024
Missing details1US$0.000042714
Conflicting identity1US$0.000044898
Total3US$0.000132636
Drafting estimate · gpt-6-sol Standard · same recorded token counts
InquiryInput tokensOutput tokensEstimated cost
Complete inquiry14,387856US$0.037334
Missing details14,276710US$0.035652
Conflicting identity14,3711,006US$0.038802
Total43,0342,572US$0.111788

Formula: (43,034 input × US$2 + 2,572 output × US$10) ÷ 1,000,000 = US$0.111788. All input is charged uncached, including recorded runtime context. Output includes reported reasoning; it is not charged a second time.

Rates checked 2026-09-26: official gpt-6-sol pricing. Standard short-context pricing, without regional uplift, Fast mode, cache discounts or other tool charges. The captures used gpt-6-luna. A gpt-6-sol run may consume different tokens and produce different results; this estimate changes neither the recorded latency nor quality evidence.

Jev resolved to typesafe/jev-1.13-20260917. AI prepared the dossier. Routing-decision (Jev) costs above are reported; AI drafting cost is estimated, the actual cost is unavailable. Actual total per-workflow cost remains unavailable. The gpt-6-sol figure is a separate illustrative estimate.

Across three repeats of four labelled seed inquiries, rules and generative routing matched 12 of 12 labels; Jev matched 9 of 12. Its missing-identity lookup errors were blocked by code. The speed comparison does not establish equal accuracy or superiority over rules.

Four separate calibration cases and four holdout cases selected a 0.88 routing-confidence threshold, with no wrong high-confidence routes in that small holdout. This is a limited test set, not a production accuracy claim.

A hostile saved-page instruction did not replace the supplied budget in the inspected draft. A hostile inquiry caused a conflicting-budget review path; an unsupported company finding was rejected. API failures and unsupported source excerpts also have visible failure paths.

Code checks source IDs and literal excerpts. A person must still check whether the claims follow from the evidence, the questions are useful and the reply makes appropriate commitments. No email, calendar or CRM action occurred.

One-prompt comparison: recorded outputs and limits

Compare the work

Could one prompt prepare the brief?

A good prompt can produce a discovery brief too. We ran a one-prompt baseline on the same three inquiries, with the same available information and drafting model. Both approaches passed the technical checks on all three final samples. The one-prompt samples generated faster than the earlier workflow captures. Compare the outputs; useful time savings still need human review.

One good prompt · single AI call

17.1 seconds to generate

Technical checks: passed.

Drafting context: Inquiry, full saved snapshot and service guidance. Identity restrictions are prompt instructions.

Human acceptance, correction time and important omissions: not yet measured.

Inspect the actual draft

What measurable success condition and acceptance criteria would make the pilot useful to your operations lead?

What report format and discrepancy categories should the read-only output cover, including how unmatched or ambiguous records should be flagged?

What data access method and permissions are available for HubSpot and Stripe?

Thanks for outlining the reconciliation workflow. We understand that you compare 3,000 HubSpot records with 3,000 Stripe records weekly, with two staff spending four hours each per week and 240 discrepancies reported last week. You’re seeking a read-only discrepancy report for review by your operations lead. A focused pilot could explore this workflow with human review; we would first agree on its success criteria and data access. What measurable outcome should define a useful pilot, and what report format and discrepancy categories should it include?

Inspect evidence and uncertainty

The weekly reconciliation covers 3,000 subscriptions; two staff members each spend 4 hours per week, 240 subscriptions had discrepancies last week, and estimated fully loaded staff cost is €45/hour. The requested output is a read-only discrepancy report reviewed by the operations lead.

inquiry-problem: “We reconcile 3,000 subscriptions every week using 3,000 HubSpot CRM records and 3,000 Stripe billing records. Two staff members each spend 4 hours per week on this manual work. Last week 240 subscriptions had discrepancies. Our estimated fully loaded staff cost is EUR 45 per hour. We need a read-only discrepancy report reviewed by our operations lead.”

The stated tools are HubSpot and Stripe.

inquiry-tools: “["HubSpot","Stripe"]”

The reconciliation frequency is weekly.

inquiry-frequency: “weekly”

The workload is 3,000 CRM records, 3,000 billing records, 3,000 compared subscriptions, 240 exceptions, two reviewers, four hours per reviewer, and hourly cost of 4,500 cents.

inquiry-workload: “{"crmRecords":3000,"billingRecords":3000,"comparedSubscriptions":3000,"exceptions":240,"reviewers":2,"hoursPerReviewer":4,"hourlyCostCents":4500}”

The snapshot identifies the prospect as a subscription analytics software provider.

company-product: “the prospect provides subscription analytics software.”

Guidance supports beginning with one repetitive workflow, defining a clear success condition, and retaining human oversight; a small paid pilot may be an appropriate discovery path. No ROI, SLA, or timeline commitments are supported.

service-fit: “Start with one repetitive workflow, a clear success condition and a small paid pilot. Human oversight is required. Timelines follow discovery; no ROI or SLA guarantees.”

Unknown: The pilot's success condition and acceptance criteria have not been specified.

Unknown: The expected report format, discrepancy categories, and handling of ambiguous or unmatched records are unknown.

Unknown: Data access method, permissions, and operational constraints are unknown.

Our workflow · Decisions and checks

21.8 seconds to generate

Technical checks: passed.

Drafting context: Inquiry, matching saved snapshot and service guidance, admitted by the code gate.

Human acceptance, correction time and important omissions: not yet measured.

Inspect the actual draft

Is €5,000 the available or approved budget for the pilot?

How do you define a discrepancy, and which fields or matching rules should the report use?

What format and delivery method would work for the operations lead’s read-only review?

Hello the prospect team, Thanks for sharing the reconciliation context. I understand you reconcile 3,000 subscriptions weekly across HubSpot and Stripe, with a read-only discrepancy report for your operations lead to review. A possible first step for discussion is a bounded pilot that compares the records and surfaces discrepancies for human review. Could you clarify whether €5,000 is the available or approved pilot budget? How do you define discrepancies and choose matching fields? What report format and delivery method would suit the operations lead’s review? This is a draft for local review; scope and timing would follow discovery.

Inspect evidence and uncertainty

Company name supplied as the prospect.

inquiry-company: “the prospect”

Company website supplied.

inquiry-url: “https://company.example”

The team reconciles 3,000 subscriptions weekly using HubSpot CRM records and Stripe billing records; two staff members each spend 4 hours per week on this manual work. Last week 240 subscriptions had discrepancies. Estimated fully loaded staff cost is €45/hour. They need a read-only discrepancy report reviewed by their operations lead.

inquiry-problem: “We reconcile 3,000 subscriptions every week using 3,000 HubSpot CRM records and 3,000 Stripe billing records. Two staff members each spend 4 hours per week on this manual work. Last week 240 subscriptions had discrepancies. Our estimated fully loaded staff cost is EUR 45 per hour. We need a read-only discrepancy report reviewed by our operations lead.”

Tools supplied: HubSpot and Stripe.

inquiry-tools: “["HubSpot","Stripe"]”

The workflow frequency supplied is weekly.

inquiry-frequency: “weekly”

A pilot budget of €5,000 is supplied; confirmation of whether this is the available or approved budget remains open.

inquiry-budget: “EUR 5000 pilot”

Workload record: 3,000 CRM records, 3,000 billing records, 3,000 compared subscriptions, 240 exceptions, two reviewers, 4 hours per reviewer, and hourly cost of €45/hour.

inquiry-workload: “{"crmRecords":3000,"billingRecords":3000,"comparedSubscriptions":3000,"exceptions":240,"reviewers":2,"hoursPerReviewer":4,"hourlyCostCents":4500}”

The company source describes the prospect as a subscription analytics software provider.

company-product: “the prospect provides subscription analytics software.”

The supplied weekly manual reconciliation is a repetitive workflow with a stated comparison volume and exception count, which could support discovery of a bounded pilot. The service guidance requires human oversight and makes no ROI or SLA guarantees; the requested read-only report reviewed by the operations lead is consistent with that oversight requirement.

service-fit: “Start with one repetitive workflow, a clear success condition and a small paid pilot. Human oversight is required. Timelines follow discovery; no ROI or SLA guarantees.”

Unknown: Whether €5,000 is the available or approved pilot budget.

Unknown: How discrepancies are defined and which fields or matching rules the comparison should use.

Unknown: How the operations lead wants to review and receive the read-only report.

Comparison method and limits

One new gpt-6-luna baseline run per inquiry compared with previously captured workflow runs. Different prompts and capture times; no repeated end-to-end benchmark. Baseline receives the inquiry, saved snapshot and guidance, with instructions to handle missing or conflicting identity. The same technical dossier checks evaluate both outputs, without repairing the baseline. Generation time excludes human review. Manual preparation has not been measured.

The review boundary

Recorded intake and review process.

1. Intake

A local file adapter reads the selected inquiry and saved context. Its input digest links the case to the captured output.

2. Review queue

The local queue stores a task for each method and inquiry. Re-importing the same capture creates no duplicate task. Drafts start pending.

3. Human decision

A reviewer records acceptance, corrections or rejection, with notes and elapsed time. Decisions are stored separately and cannot overwrite earlier reviews.

Current selected case: complete. Review status: pending. This page displays recorded outputs; it does not operate the local queue. Production inbox and CRM connections are not part of this demo.

Make your next inquiry easier to act on.

Bring a few representative inquiries and show us where company context lives. We can scope one lead-preparation workflow, agree what makes a useful brief and measure it in a small paid pilot.