How would you price an agent that completes a task rather than answers a question?
Vesslun sells Ansera, a phone agent that calls a dental practice's patients and books their own next visit, to clinics like Corvine Dental Group. Vesslun's PM, Ellery Falchuk, priced it the fast way at launch: bill on whatever Ansera's own transcript said happened. That held up cleanly for two months, then drifted quietly, because a call ending in "you're booked" and a patient actually landing on the clinic's calendar are not the same event, and only one of them should ever show up on an invoice.
The direct answer
Price Ansera per completed, kept appointment, about five dollars, not per call, per minute, or per seat. Define "completed" as a booking the clinic's own calendar and a held confirmation text can verify, never whatever the agent's own transcript claims. Build that price from real delivery cost, about fifty nine cents per booking, plus margin, bounded above by what a human scheduler and the appointment itself are actually worth to the clinic.
Do this, in order
Price per completed, kept appointment, around five dollars, not per call or per seat.Why: it ties Vesslun's revenue to the one outcome Corvine actually wants, a filled chair, not to how many times the phone rang.
Define "completed" as a calendar-verified booking, never the agent's own transcript claim.Why: an ASR mishear or an over-eager model can report a win that never made it onto a real slot.
Build the price from real delivery cost, about fifty nine cents a booking, not a guess.Why: dialing, connecting, and human escalation cost different amounts, and skipping the arithmetic hides where the margin actually sits.
Cap the price against what the appointment and the human alternative are worth, not just against cost.Why: five dollars sits comfortably under the roughly two hundred ten dollars the visit is worth to the clinic, and close to what a human scheduler already costs.
Watch the escalation rate, not the connect rate, as the number that actually moves the price.Why: more dials hitting voicemail barely touches the dollar cost, more calls needing a person does.
Never cut the redial cadence or the slower confirm-back step to save money.Why: fewer attempts lowers delivery cost but also lowers how many recalls get filled, which is the whole reason a clinic buys Ansera.
How to answer this, stage by stage
Nobody is grading whether you land on exactly five dollars. They're grading whether you know that a task-completing agent needs a receipt from somewhere other than itself before anyone gets billed for what it says it did.
1
Scope it to one product and one buyer, name who owns the price
Say it like this
"Let's ground this in one real case. Ansera is the phone agent Vesslun built to call a dental practice's patients and book their next visit, start to finish, nobody on the practice's side has to pick up a phone. Ellery Falchuk owns the price. First thing worth saying: this isn't a chatbot answering a question, it's an agent that has to actually finish something, so pricing it like a per-message or per-minute product was never going to be honest about what it costs to deliver."
Why this works
Naming what kind of product this is before quoting a number stops the answer from borrowing a pricing model built for a different job.
2
State the equation before naming a single figure
Say it like this
"Ansera's real cost has three parts, added together. The dials that don't connect, mostly voicemail. The dials that do connect and hold a real conversation. And the calls that need a person at Vesslun to step in and finish it by hand."
Why this works
A price with no visible arithmetic under it is a guess wearing a dollar sign. Saying the equation first makes it impossible to hide behind a round number.
3
Reframe what the price actually needs to protect
Say it like this
"The real question isn't what to charge per call. It's what counts as done. A call can end with the agent saying 'you're booked' and nothing landing on the clinic's calendar. If the price is built on the agent's own claim, Vesslun is billing for a task that might not have actually happened."
Why this works
This is the moment that separates a real pricing decision from someone quoting a number off the agent's own success dashboard.
4
Give the one decision
Say it like this
"Here's what I'd actually do: price Ansera at about five dollars per completed, kept appointment, defined as a real write to the clinic's calendar plus a confirmation text the patient didn't cancel. Not per call, not per minute, not a flat monthly seat. The clinic pays for a filled chair, because that's the only thing it ever actually wanted."
Why this works
This is the concrete, defensible price, not a wish that billing could stay simple while the agent's own claim carried it.
5
Own the numbers, and show the range, not one figure
Say it like this
"Getting one booking done takes about five and a half dial attempts on average, since only about a third of dials connect to a live person, and just over half of those end in a yes. Dialing alone costs about fifty nine cents per completed booking once you weight the cheap voicemail dials against the pricier connected ones and the occasional human handoff. A defensible price sits somewhere between three dollars, thin margin, and eight dollars, still under four percent of what the visit is worth."
Why this works
Giving a range, not just a favorite number, is what tells the interviewer you already know where the estimate could move.
6
Sanity check it against something real, then name the trade
Say it like this
"Here's the check that actually mattered: a human scheduler costs about nineteen dollars an hour and gets through roughly five bookings in that hour, so about three dollars eighty a booking by hand. A kept dental visit is worth around two hundred ten dollars to the clinic. Five dollars sits right between those two numbers. I'd rather eat a slightly slower, more careful call, one that repeats the day and time back before hanging up, than chase a cheaper price that risks billing for a booking that isn't real."
Why this works
Naming the trade plainly, a few extra seconds and a touch more scripted for a phantom-booking rate near zero, is what makes this a real decision instead of a wish that speed, cost, and accuracy were all free.
7
Say which assumption swings it most
Say it like this
"The number that actually swings this isn't the connect rate. If connect rate halves, the dial count roughly doubles, but most of those extra dials are the cheap kind, ones that just hit voicemail, so cost barely moves. What really swings it is the escalation rate, how often a call needs a person at Vesslun to finish it. Let that creep from eight percent to twenty percent and delivery cost climbs by close to thirty cents a booking on its own."
Why this works
Naming the assumption that swings the dollar figure, not just the one that swings the raw count, is what a good estimator does that a bad one skips.
8
Close on the one line
Say it like this
"So: price per completed, verified booking, about five dollars, built from real delivery cost plus margin, capped by what the visit and a human scheduler are actually worth. Watch escalation rate, not call volume, and never bill on what the agent says happened instead of what the calendar shows."
Why this works
Ending on the actual number and its basis, in one breath, is what makes this sound rehearsed instead of like a story that trailed off.
Let's learn
Ansera is a phone agent Vesslun built that calls a dental practice's patients and books their next visit for them, start to finish, with nobody on the practice's side picking up a phone.
About six hours of front-desk time a week at Corvine, and roughly 21 booked visits to show for it. The rest of the list waited for next week, or never got called at all.
Before Ansera, Corvine Dental Group's two front-desk staff worked the recall list by hand, about six hours a week between the two of them, and finished around 21 booked visits in a typical week.
Five steps. The last one, the write to the calendar, is the only one that should ever count as a sale.
Knowledge spark: what is ASR?
Automatic speech recognition, the part of the system that turns the patient's spoken words into text the model can read. It is very good, not perfect. A hedge like "maybe Tuesday" can come through sounding closer to a plain "Tuesday" than it actually was.
Now Ansera works through the same list overnight, before the office even opens, and the front desk arrives to a calendar that is already filling in.
The turn: the handful of bookings that didn't stick were never really the problem on their own. Any calling system, human or otherwise, mismatches a few. The real problem was that Vesslun billed Corvine for every call Ansera reported as a win, not for what actually made it onto the clinic's own calendar. Once Corvine's office manager noticed the gap, Ellery went and checked what the numbers actually said.
We were not billing bookings. We were billing what the agent believed about itself.
Bookings Vesslun billed for, versus bookings the clinic's calendar actually showed, week by week
Billed, per the agent's transcriptVerified, per the clinic's calendar
The two lines start together in week one and drift apart by about one booking a week. Nobody notices a one-booking gap. By week eight it is three, and that is the gap Corvine's office manager finally emailed about.
At its worst: one March, Corvine was billed for 34 completed bookings. The clinic's own calendar showed 31. Two of the three phantom bookings came from Ansera mishearing a hesitant "maybe Tuesday" as a firm yes. The third booked a patient into a slot that was already taken. It wasn't much money, about fifteen dollars. What it cost was trust: the office manager's next email asked whether Vesslun was checking its own numbers at all, and for a few days the renewal looked genuinely in doubt over a fifteen-dollar gap that was really a question about honesty.
The decision that mattered
Vesslun's launch price was billed on Ansera's own transcript, whether the conversation ended in a confirmed day and time. It tested clean against a spot check in week one, ten for ten. Nobody set a plan for the day the transcript's version of "done" and the calendar's version quietly stopped agreeing.
What I'd leave alone: the actual call design, the way Ansera offers two open times instead of asking an open question, and the redial cadence across a few days before giving up on a lead. None of that was broken. The billing basis was.
The lesson: a task-completing agent has to be priced off proof the task actually happened, not off what the agent says happened inside its own conversation. Those two things drift apart quietly, a booking or two a month, long before anyone notices, because the dashboard showing "calls completed" looks exactly as healthy either way.
Now here is the same thing as a story
Read the long version below when you want to feel why a pricing model that tested clean at launch could still drift wrong five months in, one honest booking at a time.
Ellery Falchuk had spent two years pricing Vesslun's other products, all usage-based, translating "the model got cheaper this quarter" into a lower per-minute rate customers could feel good about. Ellery was good at it: clean invoices, few disputes, a reputation on the team for numbers nobody had to argue with.
When Ansera needed a launch price, Ellery did what always worked before: picked the cleanest signal available and billed off it. The cleanest signal was the agent's own transcript, whether the conversation ended with a confirmed day and time. No integration needed with each clinic's own scheduling software, which varied practice to practice and would have taken months to build against. Bill on the transcript, ship in six weeks instead of six.
The first two months at Corvine went well. Ellery pulled a spot check of ten invoiced bookings against the calendar in week one, all ten matched. By week three, the dashboard's "calls completed" number sat healthy and climbing, and Ellery stopped pulling the calendar export, trusting the dashboard the way anyone trusts a number that has never once been wrong.
No single bad booking. A trickle of small, easy-to-dismiss questions, until a second clinic flagging the exact same gap made Ellery actually look.
The habit thinned in three small beats nobody would have flagged on their own. First, the weekly spot check became a monthly one. Then the monthly check became "I'll do it if the dispute rate ticks up." Then, without anyone deciding it, there was no check at all, just an invoice going out on whatever the transcripts reported.
The trigger was one line in an email from Corvine's office manager: "Quick question, why does my calendar show 31 for March and the invoice says 34?" Small. Not angry. Genuinely just a question.
Ellery spent that evening doing what should have stayed a standing habit: pulling every March transcript at Corvine and matching it against the calendar export, booking by booking. Three didn't match. Two were ASR mishearing a hedge as a yes. One booked a slot that was already filled. Then Ellery ran the same check company-wide, across every clinic on Ansera. The gap wasn't a Corvine problem. It was running close to eight or nine percent everywhere, quietly, since launch.
We didn't lose three bookings. We lost the one thing an invoice is supposed to prove: that it happened.
The old decision went back to that first pricing meeting, five months earlier. Billing off the transcript instead of each clinic's own calendar was the obviously right call to ship fast without months of per-practice integration work. Nobody in that meeting decided Ansera should keep billing off its own claim forever. It just kept working, right up until enough bookings had run through it for the gap to become visible.
The old design wasn't careless. It was the honestly fastest way to ship, and it stayed right up until the volume running through it made the gap big enough to see.
Run that same March again, with completion defined by a calendar write and a held confirmation text instead of the transcript. The invoice says 31. The calendar says 31. Corvine's office manager never sends that email, because there is nothing to notice. Run it across the whole clinic book and the quarter's disputed-invoice count drops from six clinics flagging a mismatch to zero.
What Ellery would tell that first meeting: billing off the transcript wasn't wrong to ship with. It just needed a second decision sitting right next to it, a plan for the day billed and verified stopped being the same number, and nobody wrote that plan down.
BOUND, and the price it forced Vesslun to actually defend
Not a story dressed as a framework. This is a pricing problem with a real gap hiding inside it, and BOUND is what turns "charge for a completed booking" into an actual, defensible number.
BBreak it down. What's the actual equation?
The price per completed booking equals the delivery cost of one booking, plus margin, bounded above by what the booking is worth to the clinic. Delivery cost itself is three terms added together: the dials that don't connect, the dials that do, and the calls that need a person at Vesslun to finish by hand.
Say the equation before naming a figure, or the price quietly becomes whatever number sounds fair.
Three small costs add up to about fifty nine cents. The price is almost entirely margin, and that is exactly the point, the margin is what buys the verification.
OOwn the numbers. Where did each one come from?
A dial that hits voicemail or goes unanswered costs about 2 cents, blended telephony and a short compute pass for voicemail detection. A dial that connects to a live person costs about 19 cents, the full ASR, model, and speech cost of an actual back and forth. About 32 percent of dials connect, and of those, about 55 percent end in a booked or rebooked visit, so one dial has roughly a 17.6 percent chance of ending in a completed booking, meaning it takes about 5.5 dials on average to land one. About 8 percent of connected calls get handed to a person at Vesslun, at about $1.25 each, blended scheduler time.
This is also where two rejected alternatives sit: per-minute pricing, dropped because it rewards long, inefficient calls and charges the clinic more for hold music and wrong numbers, and a flat monthly seat price, dropped because it decouples what Corvine pays from how many recalls actually get filled.
Knowledge spark: what does escalation mean here?
Some calls run into something Ansera shouldn't guess at, an unusual insurance question, a patient asking for a specific hygienist. Those get handed to a real person at Vesslun to finish. That handoff, not the phone call itself, is what "escalation" means in this pricing.
UUse a range, not one number.
Weighting 3.74 no-connect dials and 1.76 connect dials per completed booking, dialing alone runs about 41 cents, plus about 18 cents of escalation, for a delivery cost near 59 cents. A defensible price sits between three dollars, thin margin, aggressive land-and-expand pricing, and eight dollars, still under four percent of what the visit is worth to the clinic.
The whole argument for a five dollar landing spot, not a round number pulled from nowhere, lives inside this range.
The build-up: delivery cost, then price, in dollars per completed booking
The whole delivery cost is a sliver next to the price. That gap is not padding, it is what pays for the verification step, the calendar write and the held confirmation text, that the old, transcript-based price never budgeted for at all.
NNail the sanity check. Does the number survive being compared to something real?
A human scheduler costs about $19 an hour and gets through roughly 5 bookings in that hour, about $3.80 a booking by hand. A kept dental visit is worth around $210 to the clinic in chair time. Five dollars sits just above the human benchmark, since Ansera runs after hours and weekends a person won't, and at about 2.4 percent of the appointment's value, nowhere close to what the visit is actually worth.
The hardest step, and the one most answers skip. A price that sounds fair in isolation can still be sitting on top of nothing real underneath it.
DDirection. Which assumption would move the price most?
Not connect rate, even though it swings the raw dial count the most, from 5.5 to about 13 if it halves. Those extra dials are the cheap kind, voicemail, so cost barely moves. The assumption that actually decides the dollar figure is the escalation rate: let it climb from 8 percent to 20 percent, more complicated insurance questions reaching the queue, and delivery cost rises from about 59 cents to about 85 cents, the single biggest lever on the number, worth about 26 cents on its own.
Naming the assumption that swings the dollar figure, not just the one that swings the biggest raw count, is what a good estimator does that a bad one skips.
What actually swings delivery cost per booking, in cents
Biggest dollar swingClose secondBiggest raw dial-count swing, smallest dollar swing
Connect rate swings the number of dials the most, but the extra dials are cheap. Escalation rate is a smaller-sounding assumption that swings actual dollars the most, because it is priced in a person's wage, not a phone bill.
Three things worth naming directly, since this is where the real judgment sits. The AI-specific failure worth watching is a hallucinated completion: Ansera's model, trained to sound helpful, can confirm a hesitant "maybe Tuesday" as a firm booking when the ASR mishears it, and nothing catches that unless completion is checked against a real system of record. The guardrail is a hard rule, not a tuning knob: a booking only becomes billable once it is written to the clinic's calendar and a confirmation text has sat unanswered-negative for two hours, never based on what the transcript says the agent believes happened. And the trade being accepted plainly: Ansera's calls run a few seconds longer and a touch more scripted, repeating the day and time back before hanging up, in exchange for a phantom-booking rate that stays under 1 percent instead of climbing toward 9.
And if you want to be sure it really works, try it somewhere else
Same five letters, a tow truck instead of a dental chair, and this time the lever that swings the price isn't a mishearing. It's whether the truck actually finished the job.
Routewell is the dispatch agent Talonreach Roadside uses to route its tow trucks, about 900 roadside calls a week. Sedric Marlodge, Talonreach's finance lead, owns Routewell's price.
The decision Sedric would take back
Routewell's first price billed per truck dispatched, the moment a driver accepted a job over the radio. A driver accepting a job and a driver actually finishing it are not the same event, and about 6 percent of "dispatched" jobs never closed at all, a driver got pulled onto a bigger job, or the stranded member had already left with a friend.
The fix wasn't a faster dispatch. It was moving the billable event later: a job counts once GPS confirms the truck reached the location and the driver taps "job complete" in the app, not the moment a driver says yes over the radio. A dispatch attempt costs about 60 cents, radio or SMS plus a GPS check, and it takes about 1.4 attempts on average to land a truck that accepts and follows through, since some decline or are too far. Add roughly $2.00 for the geofence-and-signoff verification and delivery cost per closed job runs about $2.85. A human dispatcher costs about $22 an hour and closes about 6 jobs in that hour, near $3.67 a job, and Talonreach values a closed job at around $45 once membership retention and avoided liability are counted. Routewell prices between $9 and $18 per closed job, comfortably under that $45 ceiling and above the human benchmark, since Routewell runs at 2am on a holiday weekend a phone-only dispatcher can't always staff for.
A question answered is its own proof, the text is right there. A task completed needs a receipt from somewhere else, a calendar, a geofence, a signed-off job, before it is safe to bill.
Same method, different lever: for Ansera, the risk was a mishearing that made a call sound done when it wasn't. For Routewell, there's no conversation to mishear, the risk is a driver getting pulled away after accepting. Same fix in spirit, move the billable event to the latest, most verifiable point, a real-world signal outside the agent's own report, but a completely different mechanism to build it: a geofence and a driver's tap, not a transcript check.
Swap the trigger and it still runs.
Speed: an interviewer caps you at ninety seconds. Skip straight to it: name the completion event first, a calendar write, a closed job, never the agent's own claim, then price around that.
Cost: if Vesslun's telephony or compute costs doubled, the five dollar price wouldn't need to move right away, delivery cost is a small slice of price, so margin absorbs the hit before the sticker price has to change.
The model got better, for real: say Ansera's booking-given-connect rate rose from 55 percent to 70 percent. Delivery cost would fall toward 45 cents, but the price shouldn't fall in lockstep, since the appointment's value to the clinic never changed at all.
Where people run it wrong.
They price on the agent's own claim of success instead of a verified event, because it's the easiest thing to compute at launch.
They copy per-minute or per-seat pricing from a different kind of AI product, because it's familiar, not because it fits an agent that finishes things.
They chase the cheapest possible delivery cost by cutting redial attempts or the confirm-back step, and quietly grow the phantom-completion rate right alongside the savings.
How to use it live. Ask "what does done look like as a receipt in someone else's system" before naming a single dollar figure. If the honest answer only exists inside the model's own output, the price isn't safe to build yet.
Flashcards (tap any card to flip it)
1 · THE FRAMEWORK
What framework is this, and what's its one job?
Tap to flip
ANSWER
BOUND: show the arithmetic, own the assumptions. Built for estimation and pricing questions like this one, not a story about a habit.
2 · THE PERSON
Who is this answer about?
Tap to flip
ANSWER
Ellery Falchuk, the Vesslun PM who priced Ansera at launch, with two years of clean, dispute-free usage-based pricing behind them before Ansera existed.
3 · THE BLIND SPOT
What did Vesslun bill on that the clinic's own calendar didn't always match?
Tap to flip
ANSWER
Ansera's own transcript, whether the conversation ended in a confirmed time. About 8 to 9 percent of billed bookings never actually landed on a real calendar slot.
4 · THE EQUATION
What three costs make up one completed booking at Ansera?
Tap to flip
ANSWER
The dials that hit voicemail, the dials that connect to a live person, and the calls that need a person at Vesslun to finish by hand.
5 · THE OLD DECISION
What decision would Ellery take back?
Tap to flip
ANSWER
Billing off Ansera's own transcript instead of each clinic's calendar, chosen at launch to skip months of per-clinic scheduling-software integration.
6 · THE NUMBER
Fill in the blank: Ansera's delivery cost runs about ___ cents per completed booking, and the recommended price is about $___.
Tap to flip
ANSWER
About 59 cents delivery cost, and about $5.00 price, above the roughly $3.80 a human scheduler costs per booking and well under the roughly $210 a kept visit is worth to the clinic.
7 · THE REPLAY
Same March, new pricing basis, what changes?
Tap to flip
ANSWER
The invoice matches the calendar exactly, 31 for 31, and the quarter's disputed-invoice count drops from six clinics flagging a mismatch to zero.
8 · CROSS-PRODUCT TRANSFER
Section 4 answers this same question for a different product. Which product, and what's the different lever?
Tap to flip
ANSWER
Routewell, the tow-dispatch agent Talonreach Roadside uses. The lever there isn't a mishearing, it's whether the truck actually finished the job, verified by a geofence and a driver's signoff, not by the dispatch being accepted.
Check yourself Score: 0 / 0
True or false
1. True or false: the safest place to bill Ansera from is its own transcript, since it already ran the whole conversation and knows what happened.
True
False
Show hint
Check the key point block in Let's learn, and the ASR knowledge spark.
Show answer
False. An ASR mishear or an over-eager model can report a booking that never made it onto the clinic's real calendar. The safe billing event is a verified write to the calendar plus a held confirmation text, not the transcript's own claim.
Fill in the blank
2. Ansera's delivery cost per completed booking is about 8 cents for dials that never connect, about 33 cents for dials that do connect, and about ___ cents for calls that need a person at Vesslun to finish, for a total of about 59 cents.
Show hint
Look at the O step in the framework recap, and the labeled-parts diagram.
Show answer
18. That is the escalation cost, the piece of the price that turned out to matter most once the escalation rate itself started to move.
Multiple choice
3. Per the sensitivity chart in the framework recap, which assumption swings Ansera's dollar cost per booking the most?
A. Connect rate dropping from 32 percent to 14 percent
B. Escalation rate climbing from 8 percent to 20 percent
C. The average call getting a few seconds longer
D. The clinic's appointment value changing
Show hint
Look at the D step and the horizontal bar chart right after it.
Show answer
B. Escalation rate swings the dollar cost the most, about 26 cents, more than connect rate's 18 cents, because it is priced in a person's wage, not a phone bill.
Short answer, name the rejected alternatives
4. What two pricing models did Vesslun consider and reject before landing on price-per-completed-booking?
Show hint
Look at the O step's arrow line in the framework recap.
Show answer
Model answer: Per-minute pricing, rejected because it rewards long, inefficient calls and charges the clinic more for hold music and wrong numbers. And a flat monthly seat price, rejected because it decouples what the clinic pays from how many recalls actually get filled.
Short answer, apply it yourself
5. Think of an AI tool you use or have heard of that finishes a real task instead of just answering something. Name what "done" would look like as a receipt in some other system, not just as something the tool itself claims.
Show hint
Think about a system outside the tool's own output that could confirm the task actually happened.
Show answer
Model answer: A grocery app's AI reordering agent. "Done" isn't the agent saying "order placed," it's a real confirmation number from the store and a charge that actually posts to the card, both outside the agent's own claim.
Multiple choice
6. If Ansera's booking-given-connect rate rose from 55 percent to 70 percent because a better model converted more of the people who picked up, what is the honest move on price?
A. Raise the price, since the agent got better.
B. Lower the delivery cost estimate and let margin absorb it, since the clinic's appointment value did not change.
C. Lower the price to match the lower delivery cost, dollar for dollar.
D. Keep everything the same, since conversion rate does not affect price.
Show hint
Look at "the model got better" line in Section 4's swap-the-trigger bullets.
Show answer
B. The value of a kept appointment to the clinic never moved, only Vesslun's delivery cost did. The price is bounded by value delivered, not by however cheap the model got to run.
Before you close the answer
Why this works
Tests whether you will price a completing agent off proof it finished, not off what the model says about itself, and whether you can build a defensible number instead of picking a round one.
Follow-up traps
"Why not just price per minute, it's simpler for everyone to understand?" Response: per-minute pricing charges the clinic more for a long, inefficient call and rewards Vesslun for wasted time instead of a booked chair, the two incentives point in opposite directions.
"Isn't five dollars kind of arbitrary?" Response: no, it's bounded on both sides, above the roughly $3.80 a human scheduler already costs per booking, and well under the roughly $210 the appointment is worth to the clinic, with a real 59 cents of delivery cost underneath it.
If pressed
The confirmation text carries a two-hour grace window before a booking counts as billable, long enough for a patient to text back "actually can't make it" and short enough that Vesslun isn't waiting days to invoice a real, kept appointment.
From U2xAI Academy
From answering questions to owning outcomes.
A live workshop where you ship a working AI agent, defend a launch decision, and walk away with a portfolio recruiters can't wave off, not just more questions to study.
