01
Why build it?
To deliver HOPE’s phone screening, candidate evaluation, and scheduling to SMBs as one continuous flow—from a vague request to a human interview.
Scroll · swipe · arrow keys
Blucor
SMB alpha · live customer evidence · August 2026
Connect HOPE’s capabilities
into one hiring experience.
An SMB alpha that connects the gaps between HOPE’s existing strengths—AI phone screening and scheduling—across vague requests, customer feedback, criteria revision, and candidate re-selection.
Customer cases anonymizedTwo hires · two stalled engagementsWorking memo for alpha discussion
Answering Dave
HOPE’s strongest capabilities already exist. The bottleneck is thatthe steps before, after, and between them are disconnected.
02
What is success?
An incomplete request reaches a scheduled human interview after a real feedback-and-revision cycle, with no manual Blucor work.
03
Why can’t we do it now?
Email interpretation, criteria updates, re-selection, approval, and state handoff between existing capabilities still depend on people.
04
Minimum experience
One complete loop: vague request → candidates → free-text feedback → approved revision → better-fitting candidates → scheduled interview.
Whose problem are we solving?
The customer does not have a recruiting team waiting to operate software.
The target is an SMB with one HR person—or where the owner, operations leader, or office manager handles hiring alongside their primary job. They may not be skilled at writing JDs or structured interviews; they are hiring while running the company.
“I need someone who can do this job” should be a normal product input.Reality across all four customers
Why this alpha matters
- A model that assigns one person to each customer cannot scale SMB delivery.
- Clarification, follow-up, and scheduling are most needed precisely when the customer is busiest.
- When hiring slows, candidates move to other employers.
- Before the larger October build, we can learn with real customers which UI/UX lets them respond naturally to AI proposals.
Current workflow and stopping points
The capabilities exist. But the critical path connecting them is manual.
Handled manually by a person01 · Vague request
A person collects the location, employment terms, selection process, and other inputs needed to start through follow-up emails or meetings.
Existing HOPE capability02 · Search and AI screening
Job setup, candidate sourcing, résumé evaluation, and AI phone screening can already be executed.
Enterprise · implemented / not used03 · Customer reaction
The feature for receiving customer feedback is implemented, but Enterprise customers are not using it yet.
Handled manually by a person04 · Update criteria
A person separates must-haves from preferences and rewrites questions and evaluation criteria.
Handled manually by a person05 · Restart
A person initiates re-evaluation of existing candidates or a new search.
Handled manually by a person06 · Interpret selection
A person forwards the customer’s “I want to interview” reply into the next process.
Enterprise · implemented / not used07 · Human interview scheduling
The feature is implemented, but Enterprise customers are not using it—possibly because of UI/UX issues.
Handled manually by a person08 · Learn again
A person reconnects the workflow again to feed the human interview outcome into the next search.
The workflow does not fail because AI finds no one. It fails because ownership of the next action disappears somewhere between email, judgment, and the calendar.
The value people were actually providing
Eight human jobs made four critical handoffs in the hiring flow work.
Phase 1 · Request → search launch
Make the vague request actionable
1
Extract missing information
Ask only for what is needed to begin, without demanding a formal new JD.
Phase 2 · Candidate presented → feedback
Turn customer reaction into judgment
2
Interpret free text
Convert a terse rejection reason into a specific mismatch.
3
Decide: change or overfit
Distinguish one person’s impression from a recurring, important signal.
Phase 3 · Criteria revision → next candidate
Apply learning to the next action
4
Apply to criteria
Update the JD, questions, evaluation, and report to the same understanding.
5
Re-source and re-evaluate
Review existing candidates—not only new ones—against the updated criteria.
Phase 4 · Candidate selected → human interview
Keep the selected candidate moving
6
Maintain candidate momentum
Do not make the candidate wait while the customer is busy.
7
Coordinate the human interview
Handle availability, confirmation, reminders, and rescheduling.
8Across every phase · remember state
Across email, ATS, SMS, calls, and calendars, a person remembered who decided what and what needed to happen next.
Evidence from four live customers
In the same hiring loop, two customers reached a hire when people connected the steps; two stalled when people could not keep up.
Case A · Successful hire
Regional recycling operator
Mechanic. A person diagnosed a rejection, updated the criteria, and moved the next candidate to interview the same day—ending in an accepted offer.
Case B · Successful hire
Legacy manufacturing group
Engineering roles. People continuously corrected gaps between the JD and plant reality, then moved a strong candidate quickly into interviews and a hire.
Case C · No hire
Industrial components distributor
Sales role. The customer signed, paid, and gave corrective feedback, but the next candidate slate arrived too late.
Case D · No hire
Multi-city specialty contractor
Installer roles. Criteria improved, but email and manual scheduling cost candidate momentum.
Validated: when people operate this loop well, it leads to interviews and hires.
Still to prove: can we build UI/UX that lets customers confidently review, correct, and approve when AI runs the same loop?The correct Wizard-of-Oz interpretation
Still to prove: can we build UI/UX that lets customers confidently review, correct, and approve when AI runs the same loop?The correct Wizard-of-Oz interpretation
Case A · Successful hire · initial request and process
The mechanic search began broadly; the first rejection made the real maintenance bar explicit.
Customer and hiring setup
Regional recycling operator. One HR person; domain decisions sat with operations and executive leaders.
Role
Industrial & Mobile Equipment Mechanic. Open for roughly three months.
Initial request and missing information
“Maintenance experience” was ambiguous. Location, employment type, start timing, shift, and the post-AI interview process were also incomplete.
Final outcome
Offer accepted
Search began with broad “maintenance experience”
The difference between operating, assisting, light maintenance, and independent repair was not explicit.
Candidate rejected
Experience centered on operating equipment, without independent troubleshooting and repair.
Four discriminating questions
Clarified the missing skill, expected level, definition of maintenance, and three must-confirm items before interview.
Customer approved new criteria
Three to five years of heavy-equipment maintenance, electrical schematics, hydraulics, and independent troubleshooting and repair.
The product must turn free text into a before/after criteria diff the customer can review and approve.
Case A · Progress after revision and outcome
The criteria improved the same day, while an existing strong candidate kept moving into a human interview.
Winning candidate completed AI interview
The candidate with strong maintenance experience was already in the pipeline before the later feedback.
Another candidate rejected
Customer identified insufficient hands-on depth and independence.
Criteria and questions updated
Diagnostic questions for the operations leader were incorporated into the selection criteria.
Human interview scheduled
A person obtained the hiring manager’s availability and contacted the candidate immediately.
Human interview
Confirmed 6+ years in heavy-equipment maintenance, hydraulics, pneumatics, and electrical systems.
Offer accepted
The long-open position was filled.
People collected the rejection reason, updated criteria, obtained Hiring Manager availability, contacted the candidate, sent reminders, and handled exceptions. The connection and speed are what AI must replace.Manual work that produced the outcome
Causality:The rejection feedback did not discover the winning candidate. The hire came from improving the process without letting an existing strong candidate stall.
Case A · Verbatim evidence · Email
A person turned “not right” into concrete criteria the next screen could use.
Blucor operator → customer HRDay after rejection
“What specifically was missing from [Candidate A]’s skillset for this role?”
“How far below your expected maintenance level was he? … D. Able to independently troubleshoot and repair”
“If you had to give us the top three ‘must-confirm’ maintenance skills before sending a candidate onsite, what would they be?”
Customer HR → Blucor operatorOperations and executive leaders’ answers
“no real maintenance background”
“wouldn’t necessarily be able to be put on a job without help to explain how to complete the task”
“1. 3-5 Years Hands on experience … 2. General understanding in Electrical Schematics. 3. General understanding in Hydraulics.”
Customer HR → Blucor operatorAfter the revision
“[Candidate B] arrived for his interview today and management felt pretty strongly about this candidate.”
“The offer has been extended to [Candidate B] and he has accepted.”
UI/UX required by this conversationDesign input
- Ask not only what was missing, but also the expected level and the must-confirm evidence.
- Automatically turn the answers into a before/after criteria diff and proposed questions.
- Once the customer approves the diff, continue re-evaluation, candidate contact, and scheduling within the same state.
Human value:Diagnosed an ambiguous rejection, converted it into actionable hiring criteria, and kept the next candidate moving.
Source: actual email. Only names, company, and contact details are anonymized. English spelling, grammar, and word order are preserved verbatim.
Case B · Successful hire · initial request and process
Candidate interviews exposed the gap between the engineering role on paper and the work the plant actually needed.
Customer and hiring setup
Legacy manufacturing group. HR was the point of contact, but the true criteria were distributed across multiple Hiring Managers.
Role
Mechanical, MES, and controls engineers.
Initial request and missing information
The documents conflicted with reality on experience level, plant-floor ratio, legacy equipment, CAD, hours, after-hours work, and visa terms.
Final outcome
Offer accepted
Three years of experience appeared sufficient
In practice, the expected bar was roughly six-plus years.
Advanced robotics
The real work centered on diagnosing and improving legacy equipment.
SCADA and MES examples were vague
The résumé was strong, but concrete examples of ownership were missing.
CAD and plant-floor experience were missing
Candidates were too desk-oriented or centered on a different technical domain.
Communication risk
A one-way monologue left the collaborative style unclear.
Do not add a question after every case
Wait until the same gap recurs across candidates, then update only the smallest useful criterion.
Case B · Progress after revision and outcome
People propagated learning across systems and moved a strong candidate to hire in roughly one month.
Correct job documents
Aligned experience level, plant-floor ratio, legacy equipment, and working conditions with reality.
Edit AI questions
Asked briefly for the needed evidence without making the interview endlessly longer.
Re-evaluate and regenerate
Reviewed prior candidates and regenerated reports with the updated reasoning.
Connect ATS and feedback
Reflected candidate state in the system the customer actually used.
AI interview → video → onsite
10+ years of equipment design, troubleshooting, CAD, and hands-on plant experience.
Offer accepted
The handoff from evaluation through human interviews kept moving.
Learning must be selective, cumulative, and operational. Customers need UI/UX that makes small AI-proposed changes easy to understand and approve.Case B product requirement
Causality:Initial role calibration preceded discovery of the winning candidate. Some later feedback arrived after the candidate was already in the pipeline, so we do not claim every feedback item caused discovery.
Case B · Verbatim evidence · Circleback + Email
People learned the needed specificity without adding questions after a single concern.
Customer HR · Feedback meetingAfter candidate interviews
“Her resume looked really good. … But once we started digging into her exact experience, she wasn't really able to give us detailed examples.”
“What production problems did you fix? Give us an example. … Okay, well, give us an example of a problem.”
“Again, they're looking for specificity. They want the candidate to provide a specific example.”
Customer HR · Feedback meetingDecision to add questions
“If I see a consistent theme emerging where we miss … then I'll add a question. Otherwise … does it make sense to change this question or do we keep it the same?”
Blucor operator → customer HRApproval request after update
“Based on the feedback from the hiring manager, I updated the interview questions and evaluation criteria for the two positions.”
“We also ran simulation tests … Could you please review them?”
UI/UX required by this conversationDesign input
- Group free text into recurring signals such as “insufficient examples,” “plant-floor experience,” and “technical center.”
- Show evidence for whether a comment is a one-off impression or a gap shared across candidates.
- Show the proposed question and criteria diff plus the expected re-evaluation impact, then make approval one step.
Human value:Accumulated Hiring Manager feedback and judged the tradeoff between interview length and screening precision.
Source: actual Circleback transcript and email. Only person, company, and candidate names are anonymized. English is preserved verbatim.
Case C · No hire · from initial request to stall
The customer signed, paid, and provided corrective feedback—but we did not produce the next fitting candidates.
Customer and hiring setup
Industrial components distributor. The HR leader was the point of contact and repeatedly followed up for candidates.
Role
Outside Sales Representative.
Initial request and missing information
Required B2B sales plus field experience with parts, bearings, conveyors, gears, motors, and related products.
Final outcome
No hire
Candidate recommended
Strong sales, CRM, and account-management background, but a gap in direct conveyor experience.
Customer gave a specific correction
Replied that more field-based industrial product experience was needed.
Customer chased the next candidates
No visible result from the revision.
Human apology and promise
Promised two to three candidates the following week.
Additional manual clarification
Multiple follow-ups sought more detailed criteria.
No next fitting candidates delivered
Momentum declined despite demand and feedback.
The problem was not a lack of customer engagement. A reply to the bot email did not trigger immediate criteria revision and a new search.
Case C · The experience that was needed
One customer reply should have produced clarification proposals, followed by immediate re-evaluation and re-sourcing after approval.
01
Is field industrial sales a must-have?
Separate a hard requirement from a strong preference.
02
Which adjacent products count?
Define acceptable substitutes across parts, bearings, motors, drives, and related products.
03
Which distribution or technical sales experience transfers?
Judge when similar customers, products, and field conditions can substitute for direct experience.
04
Which matters more: new business or account growth?
Identify the actual sales motion hidden inside generic “sales experience.”
AI presents clarifications conversationally → the customer corrects or approves → prior candidates are re-evaluated immediately → updated candidates appear within 48 hours. This UI/UX would remove the wait for a person to find and interpret the email.The revision loop that was needed
Case C · Verbatim evidence · Email
One customer email already contained the information needed to update the criteria.
Customer HR → HOPE report emailAfter candidate presentation
“Doesn’t seem like a good fit.”
“We’re looking for industrial sales type like parts, bearings, conveyor belts, gears, motor boxes, etc.”
“Education and customer accounts seem fine but we need someone that’s more in the field.”
Structure already present in the replyWhat AI could extract
- Keep: education and customer-account management.
- Strengthen: direct experience with industrial product categories.
- Strengthen: field sales activity rather than desk-centered work.
- Clarify: are the listed products mandatory, or do adjacent products count?
Blucor operator → customer HRManual reply seven days later
“We have tightened the screening criteria to prioritize candidates with more direct field and industrial sales experience…”
“We will do our best to have 2-3 more candidates ready … next week.”
A person wrote the promise, but the product did not own the search, visibility, or deadline management that followed.
UI/UX required by this conversationDesign input
- Recognize an email reply immediately as feedback intent.
- Confirm the proposed change in a short conversation and launch a new search with one-click approval.
- Automatically report the 48-hour deadline, candidates under review, and the reason when no one matches.
Failure point:Not the value hypothesis—the ownership and speed from reply to execution.
Source: actual email. Only person, company, candidate, and contact details are anonymized. English is preserved verbatim.
Case D · No hire · initial request and criteria revision
At a company with no dedicated recruiter, a criterion biased toward direct experience was successfully improved.
Customer and hiring setup
Multi-city specialty contractor. No dedicated recruiter; executives, operations, and office staff shared the work.
Role
Thirteen equipment installer openings.
Initial request and missing information
Direct specialty-equipment experience was required. Start date, benefits, worker classification, notice, and training were also incomplete.
Final outcome
Paused · no hire
Direct experience required
AI rejected even strong candidates with adjacent experience.
Welding · repair · tools · construction
The candidate had transferable hands-on capability.
Adjacent trades as the must-have
Direct experience became preferred; specific vendor experience became a plus.
Customer approved
The criteria-revision conversation itself succeeded.
The alpha should reproduce this experience: overlooked transferable skill → AI change proposal → customer correction and approval.
Case D · The system had to behave like a recruiter within 24 hours
Even when the customer replied “I want to interview,” it did not automatically become an interview action.
24h
Required response speed
Customer and candidate replies had to happen within one day or momentum and availability disappeared.
2 people
Lost to other offers
Two qualified candidates accepted other jobs while the process was moving.
1 person
Left after a few hours
One person quit only a few hours after starting, making a live pipeline essential.
Flexible
Pause / resume
The customer needed to pause and resume hiring volume without rebuilding the process.
“Perfect to interview” / “worth a call”
Hiring intent arrived in the bot email.
A person forwarded it to another operator
Someone had to notice, understand, and route it.
Asked again for the candidate’s phone number
The customer followed up because the information and next action had not arrived.
Original availability expired
By the time the process restarted with an apology, the original interview windows were no longer usable.
Case D · What HOPE already had—and what people still handled manually
AI phone screening existed. But people still handled the workflow from customer selection to the human interview manually.
Exists in HOPEAI phone screening
Screening call and report for the candidate.
Handled manually by a personInterpret customer selection
A person read and forwarded “I want to interview” from the bot email.
Handled manually by a personSchedule the human interview
A person collected customer and candidate availability and confirmed the interview time.
Handled manually by a personSMS and reminders
A person sent confirmations, SMS notifications, and reminders to the candidate.
Handled manually by a personLaunch and monitor calls
A person launched call transfer and checked Call History.
Handled manually by a personRecover exceptions
Tracked delays, no-shows, stopped replies, and rescheduling.
Handled manually by a personWrite outcome back to state
Synchronized progress across customer, candidate, and internal systems.
Handled manually by a personContinue to the next candidate
Restarted the same loop for each role and city.
The problem was not an inability to run AI phone screens. People manually connected customer selection, candidate replies, scheduling, SMS, calls, and outcome updates. The customer later returned with another role, showing demand remained.Case D conclusion
Case D · Verbatim evidence · Circleback + Email
Before “who should we hire,” the problem was “I do not have time to take the next hiring action.”
Owner · Feedback meetingNo dedicated recruiter
“I'm overwhelmed with what I need. I know who I need.”
“You want to know the truth? I don't even have the time to call those people.”
Owner → HOPE report emailCandidate selection
“This person is perfect to interview.”
“short list him to call!”
Blucor operator → customerManual scheduling clarification
“Will you be the one conducting the phone interview next, or will it be [Interviewer B]?”
“It would be helpful if you could let us know the preferred date/time for the call and the phone number to use.”
Customer: “[Interviewer A] will conduct the first interview on Wednesday. Then if we like them we will set up a follow up interview … on Thursday.”
UI/UX required by this conversationDesign input
- Treat “perfect to interview” as intent to start the interview process—not as a comment.
- Store the interviewer, contact details, and availability once instead of asking again by email every time.
- Send proposed times to the candidate and continue automatically through confirmation, reminders, rescheduling, and outcome updates.
Human value:Expanded a short customer signal into multiple operational actions: confirm the interviewer, contact the candidate, and finalize the schedule.
Source: actual Circleback transcript and email. Only person, company, candidate, and contact details are anonymized. English is preserved verbatim.
Cross-case comparison
The roles differed, but the structure separating success from stall was the same.
| Dimension | A · Recycling | B · Manufacturing | C · Components | D · Contractor |
|---|---|---|---|---|
| Outcome | Hire | Hire | No hire | No hire |
| Initial ambiguity | Operation vs independent maintenance | Desk vs floor, experience bar, legacy equipment | Generic sales vs field industrial products | Direct experience vs adjacent trades |
| Useful feedback | Operations leader defined the true maintenance bar | Multiple interviews revealed the need for plant-floor specificity | Customer named the required product categories | Customer approved a transferable-skill model |
| Human action | Diagnosed, revised, contacted, scheduled | Calibrated, edited questions, re-evaluated, bridged ATS | Interpreted late; next candidates did not arrive | Criteria improved; communication and interview operations stayed manual |
| Critical difference | The next action always had an owner | Learning propagated through the workflow | The reply did not trigger immediate execution | Interview intent did not automatically become an interview |
Success pattern · conditions for turning HOPE into one experience
Feedback becomes a decision, the decision becomes a system change, and the change becomescandidate action—fast.
State
Know what is true now
Maintain the current job version, evidence behind each criterion, candidate stage, unanswered questions, availability, and the promised next action.
Judgment
Know what should change
Separate must-haves from preferences, anecdotes from consistent signals, and automation from exceptions requiring approval.
Agency
Make the next move
Ask, update, re-evaluate, search, contact, schedule, remind, reschedule, and report without waiting.
HOPE already has execution capabilities such as AI phone screening and scheduling. The alpha adds the connection layer that preserves context between them, proposes the right changes, and lets customers respond without confusion.Connection layer = State + Judgment + Agency + usable UI/UX
Responsibility boundary between people and AI
“Zero manual Blucor work” and “customer control” can coexist.
Always autonomous
Low-risk operations
- Understand email and SMS replies
- Ask factual follow-up questions
- Re-evaluate existing candidates
- Collect availability
- Schedule within approved rules
- Reminders, state updates, progress reports
Customer one-click approval
Criteria and company representation
- Change must-haves and preferences
- Add or remove AI interview questions
- Material changes to job language
- Material candidate messaging
- Show before/after diff and rationale
Escalate as exception
High risk, conflict, or repeated failure
- Conflicting Hiring Manager feedback
- Discriminatory hiring criteria
- Material changes such as compensation
- Legal or contractual risk
- Repeated scheduling failure
Alpha success is not “a person handled it invisibly behind the scenes.” In the normal flow, customers can understand and respond to AI clarifications, proposed changes, and next actions without confusion; only exceptions are surfaced explicitly.
Minimum coherent alpha experience
Complete one learnable loop—not a broad recruiting system.
01 · Receive
Customer sends an incomplete JD or short email.
02 · Clarify
AI conversationally asks only for missing information that blocks action.
03 · Approve
Customer approves the job and selection criteria in one click.
04 · Execute
AI posts, sources, screens, and produces reports.
05 · Learn
Customer provides free-text feedback on candidates.
06 · Revise
AI presents a hiring-criteria diff and obtains approval.
07 · Return
Present re-evaluated or newly sourced candidates within 1–2 days.
08 · Schedule
When the customer selects someone, automatically coordinate the human interview.
Product promise: give us an incomplete request and correct us in ordinary language; the hiring process keeps moving.
Alpha scope
The major components already exist. What we need to build is the connection layer.
Already available
- Job setup and posting
- Candidate sourcing
- Résumé screening
- AI phone screening
- Candidate reports
- SMS send and receive
- Some call transfer
- Candidate status management
Missing connective tissue
- Understand inbound customer email
- Ask only the missing questions
- Structure free-text feedback
- Propose and approve a criteria diff
- Re-evaluate prior candidates
- Re-source automatically after a change
- Collect availability from both sides
- Schedule, reschedule, notify, and recover no-shows
- Preserve state across every channel
Primary alpha success criterion
Start from an incomplete request, completeat least one real feedback-and-criteria-revision cycle, and schedule a human interview with a customer-chosen candidate—with zero manual Blucor work on the critical path.
≤1 day
Request → approved search
Launch the search within the customer’s realistic time constraints.
Minutes
Feedback → criteria diff
Propose the change while interview memory is fresh.
≤48h
Approval → updated candidate
Re-evaluated existing candidates or newly sourced candidates.
≤24h
Selection → scheduled interview
AI handles availability, confirmation, and notification.
0
Normal-flow human work
No manual email forwarding, criteria editing, calendar chasing, or reminders.
2+ customers
Repeat with real customers
Complete the full loop at least twice across real roles in different SMB contexts.
How to treat the hiring outcome
An offer matters enormously—but it should not be the alpha’s only pass/fail gate.
Alpha pass condition
Autonomous operation through scheduled human interview
Directly test the value the product controls: interpretation, approved learning, execution, state management, and speed.
＋
North star / stretch goal
At least one offer
Ideally accepted or started. This is strong evidence of quality, but it also depends on compensation, headcount, employer decisions, background checks, and competing offers.
AI runs normal operations
Contact, factual clarification, scheduling, notifications, rescheduling, and state updates.
Customer approves material changes
Review before/after diffs for criteria, questions, and job language.
Only exceptions go to a person
Conflicts, discrimination or legal risk, material term changes, and repeated failure.
An alpha that informs the larger October build
Run the narrow loop end to end with two real customers that have lean HR teams.
1
Choose the right roles
Active, urgent searches with at least one participating decision-maker, a meaningful candidate pool, and willingness to provide free-text feedback.
2
Instrument every handoff
Measure response time, approvals, AI actions, exceptions, human interventions, candidate wait time, and interview completion.
3
Carry evidence into October
Decide which judgments can be automated, which need lightweight approval, and which rare exceptions require an expert.
Do not rebuild HOPE.Connect the strong capabilities that already exist into one hiring loop that does not stop.
The four cases show that value appears when people connect the gaps. The alpha must prove that we can build UI/UX where customers confidently respond to AI-proposed clarifications, criteria changes, and next actions.