CalculationAdvancedAI Opportunity & Model Strategy / Build vs buy vs fine-tune decisions / #5
Your vendor's pricing changes annually. How does that affect a build-versus-buy analysis?
BOUNDthe vendor was the cheaper option for three straight years, and the fourth year is when the math actually flips
Northgate Terminals runs Quaywatch, a feature that predicts how late an incoming vessel will be, built on a vendor's prediction API. Iselin Draper is the AI PM who has to model whether staying on that vendor for five years still beats building the model in-house, given that the vendor raises its price at every annual renewal.
The direct answer
It turns a one-time comparison into a compounding one. A vendor that looks 60 percent cheaper than building in-house on day one can still end up more expensive within four years, because their price grows every year while your in-house cost, once built, mostly doesn't. Model the vendor's cost as a growth rate, not a snapshot, and find the year the two lines actually cross.
Do this, in order
Model the vendor's price as compounding, not flat, using their real renewal history.Why: a single quote hides the growth rate, and the growth rate is what actually decides this over five years.
Find the exact year the two cumulative-cost lines cross, not just which is cheaper today.Why: "cheaper right now" and "cheaper over the life of the decision" are different questions with different answers.
Negotiate a rate cap into the contract before signing, not after the third renewal.Why: the growth rate is the single assumption that moves this answer the most, more than the starting price.
Price the in-house maintenance cost honestly, not just the upfront build.Why: in-house isn't free after year one, it's just flat, and flat is what beats a compounding number eventually.
Set a real trigger for revisiting build, tied to the renewal terms, not a gut feeling.Why: without one, a crossed line stays unnoticed until someone happens to add up five years of invoices.
Sanity-check the five-year totals against something you already know.Why: converting the gap into "more than double the in-house build cost" catches an order-of-magnitude mistake before a renewal meeting does.
How to answer this, stage by stage
Six moves. This is an estimation question, so the structure matters more than any single number, but the structure has to be said out loud.
Stage 1
Ask for the one number you can't invent
Say it like this
"Before I model this, one question: what's the vendor's actual historical renewal increase, not their initial quote? I'll assume 12 percent a year, based on their last three renewals, and flag that the whole answer moves a lot if that number is wrong."
Why this works
Names the load-bearing number in ten seconds, then answers your own question and keeps moving.
Stage 2
State the equation before any numbers
Say it like this
"So the shape is: vendor's year-N cost equals year-one price, times one plus the growth rate, to the power of N minus one. In-house's year-N cost equals the upfront build, plus flat annual maintenance times N. Compare the running totals, not the single-year prices."
Why this works
The interviewer can follow every number that comes next, and can push back on a specific term instead of the whole answer.
Stage 3
Put a stake in each assumption
Say it like this
"Northgate runs about two million predictions a year. At the vendor's signing price of about four cents each, that's 85,000 dollars in year one. In-house, I'd estimate three engineers for about four months to build and validate a proprietary model, call it 220,000 dollars upfront, plus 45,000 a year to run and maintain it."
Why this works
Every number traces to something checkable, volume times price, and engineer-months times a loaded rate, not a guess dressed up as precision.
Stage 4
Run the arithmetic and find the crossover
Say it like this
"At 12 percent a year, the vendor's cost climbs from 85,000 in year one to about 134,000 by year five. Running totals: the vendor is cheaper through year three, 287,000 cumulative against in-house's 355,000. By year four that flips, vendor at about 406,000 against in-house's 400,000. By year five the gap is 95,000 dollars, and it's still growing."
Why this works
Naming the exact year of the crossover, not just "eventually," is what separates a real model from a hand-wave about vendor risk.
Stage 5
Sanity check against something you already know
Say it like this
"Quick check: by year five, the vendor is charging about 6.7 cents per prediction against an in-house cost of about 2.3 cents. That's roughly three times more per prediction, for the same job, which is a believable gap for a compounding subscription against a flat-maintenance system, not an order-of-magnitude error."
Why this works
Converting the total back into a per-unit cost is what catches a mistake before the interviewer has to.
Stage 6
Name the assumption that would break it
Say it like this
"The number I'd want to pressure-test isn't the starting price, it's the growth rate. At a capped 5 percent instead of 12, the crossover doesn't happen until right at the edge of year five instead of partway through year four. A rate cap in the contract is worth more to this analysis than any discount on the sticker price."
Why this works
Naming which assumption moves the answer most is what separates an estimator from someone who just multiplied two numbers together.
If you remember one thing
A vendor quote is a snapshot. A build-versus-buy analysis needs a growth rate. The interviewer isn't checking your arithmetic, they're checking whether you modeled the vendor's price as something that compounds.
Let's learn
Here is the whole model in one block, so the shape is visible before arguing about any single number in it.
Five steps, and the third one, comparing running totals instead of today's prices, is the one most estimates skip.
# inputs
predictions_per_year = 2,000,000
vendor_price_year1 = $0.0425 / prediction # $85,000 in year one
vendor_annual_increase = 12%# last three renewals, averaged
inhouse_upfront = $220,000# 3 engineers, ~4 months, loaded
inhouse_annual_maint = $45,000# compute + fractional engineer time
# vendor cost, year by year
Y1 $85,000 Y2 $95,200 Y3 $106,624 Y4 $119,419 Y5 $133,750
Five-year total cost, vendor versus in-house, broken into its parts
Vendor, compounding at 12% a yearIn-house, upfront buildIn-house, 5 years of flat maintenance
Same five-year window. The vendor's total ends up about 95,000 dollars higher, almost entirely because of the growth rate, not the starting price.
Before anyone modeled the growth rate, the vendor looked like the obviously cheaper option. At signing, 85,000 dollars a year was less than a third of what building in-house would cost just to get started, and nobody ran the comparison any further than that first year.
The vendor wasn't the expensive option. It was the option that got expensive, one renewal at a time, while nobody was adding up the total.
The renewal itself is the actual decision point. Most teams only revisit build versus buy when the vendor forces the question by breaking something.
Cumulative cost over five years, vendor versus in-house, crossover marked
Vendor stays cheaper for the first three years. The lines cross about eleven months into year four, and the gap keeps widening afterward.
Knowledge spark: why does a flat-maintenance line always win eventually?
Any cost that compounds grows faster every single year, even at a constant rate, because each year's increase is applied to an already-larger number. A flat cost never does that. Given enough years, a compounding line crosses above a flat one, no matter how much smaller it started. The only real question is which year it happens in.
The annual percentage is only one of five ways a vendor relationship compounds against you. A real analysis checks all five, not just the one on the invoice.
At its worst, this cost shows up as a finance review that only adds up five years of invoices after the fact, discovering the vendor quietly became the more expensive choice two renewals ago, with no rate cap in the contract to have stopped it.
Where the estimate breaks if you treat the vendor's price as one number
If a renewal ever comes in far above the modeled rate, the runway to migrate has to start the same week, not after the next invoice arrives.
The actual trap isn't a bad growth-rate guess, it's skipping the growth rate entirely and comparing only today's vendor quote to today's in-house estimate. That comparison genuinely favors the vendor in year one, 85,000 against 265,000, and it's still true in year three. The mistake is stopping the comparison there.
Iselin's team had budgeted Quaywatch's vendor line as a fixed number for three straight annual planning cycles, because the first renewal came in close to flat and nobody flagged it as a pattern worth modeling forward. The third renewal came in at 12 percent, matching the two before it, and only then did anyone plot all five years at once.
Not every capability riding on a vendor faces the same exposure. The ones where a company's own data is genuinely the edge are worth building in-house regardless of price, before the math even runs.
BOUND, in one screen
This is an estimation question, so the framework is BOUND. There's no person's habit to track here, the whole answer is arithmetic, and forcing a FLIPS-style story onto it would bolt a parable to the side of a spreadsheet.
B
Break it down.
Vendor year-N cost equals year-one price times one plus the growth rate to the power of N minus one. In-house year-N cost equals the upfront build plus flat annual maintenance times N.
Stating the equation first is what lets the interviewer follow, and challenge, every number that follows.
O
Own numbers.
2 million predictions a year, 4.25 cents each at signing, a 12 percent historical renewal rate. In-house: 220,000 upfront for three engineers over four months, 45,000 a year to maintain.
Each number traces to something a person could check, not a figure pulled from the air.
U
Use a range.
Crossover lands somewhere between year three and year five depending on the real renewal rate, month 47 at 12 percent, closer to month 58 at a capped 5 percent.
A single crossover year implies more confidence in the growth rate than a three-year renewal history actually earns.
N
Nail the sanity check.
By year five, the vendor charges about 6.7 cents per prediction against an in-house cost near 2.3 cents, roughly three times more for the same job.
A per-unit comparison catches an order-of-magnitude mistake before a renewal meeting does.
D
Direction.
The annual growth rate moves this answer more than every other assumption combined. A capped 5 percent pushes the crossover to the edge of year five. An uncapped 12 percent pulls it into year four.
Knowing which assumption to pressure-test is what separates an estimator from someone who just multiplied two numbers together.
Try the same model somewhere elseSame five letters, a library metadata tool instead of a shipping-delay model. This time the crossover never comes at all.
Petra Alderman runs AI tooling for the Wrenmoor Library Consortium, which uses a vendor's model to auto-enrich catalog records with subject tags and summaries. Break it down the same way: vendor year-N cost against in-house upfront plus flat maintenance. Own numbers here look different: Wrenmoor processes only 40,000 new records a year, so even at the vendor's aggressive 15 percent annual increase, year-one cost is just 3,200 dollars, growing to about 5,600 by year five. In-house would still cost at least 90,000 dollars upfront for a much smaller consortium's engineering team to build and validate. Use a range: even in a pessimistic scenario with a 25 percent annual hike, five-year vendor cost tops out near 24,000 dollars, nowhere close to the in-house upfront alone. Nail the sanity check: buying is under 3 percent of the in-house cost, every year, for five years straight. Direction: at this volume, no realistic growth rate ever closes the gap, so the honest answer is buy and stop modeling it further, the crossover this method looks for simply never arrives.
Swap the trigger and it still runs.
Speed: an interviewer caps you at sixty seconds. Say "model the vendor's price as compounding, find the year the cumulative lines cross, and negotiate a rate cap before signing," and stop.
Cost: no time to get the vendor's real renewal history before the interview ends. Say so honestly, and use their industry's typical annual increase as a stated placeholder, flagged clearly as a placeholder.
The model got better, for real: say a cheaper open model becomes available partway through year three, dropping the in-house maintenance estimate. Rerun the crossover anyway, a lower in-house cost just pulls the crossover earlier, it doesn't remove the need to check for one.
Where people run it wrong.
They compare today's vendor quote to today's in-house estimate and stop, missing the compounding entirely.
They assume the vendor's next renewal will match its first, without checking whether the first was an introductory rate.
They price the in-house build but forget the flat annual maintenance it still needs, understating the true in-house line.
How to use it live. The moment an interviewer mentions a vendor whose pricing changes over time, ask yourself first: is this a snapshot comparison or a running-total comparison? That question alone buys real thinking time, and it's usually exactly where a shallow answer and a real one split apart.
One short aside about people
Estimation questions don't need a story. This one paragraph is worth adding anyway, because it's the part that makes the model wrong in practice if you skip it.
A rising vendor bill rarely gets challenged one renewal at a time, because a 12 percent increase in isolation looks like a rounding error next to the rest of a budget. The team that keeps approving it isn't being careless, each single renewal really is a small decision. Farrukh Demir, the engineer who first plotted all five years on one chart, put it plainly in the review: "None of us ever said yes to 540,000 dollars. We said yes to 12 percent, five separate times." Worth one sentence in the interview: "I'd want a standing chart that shows the cumulative line, not just this year's invoice, because nobody ever approves the five-year number on purpose, they approve it one small renewal at a time."
Flashcards (tap any card to flip it)
1 · THE FRAMEWORK
Which framework fits modeling how annual vendor price changes affect build versus buy, and why not FLIPS?
Tap to flip
ANSWER
BOUND. There's no person's habit snapping here, just a compounding number against a flat one. Forcing FLIPS onto it would bolt a story onto arithmetic that doesn't need one.
2 · THE EQUATION
State the vendor-versus-in-house comparison in one line.
Tap to flip
ANSWER
Vendor year-N cost is year-one price times one plus the growth rate to the Nth power. In-house year-N cost is the upfront build plus flat annual maintenance times N. Compare the running totals.
3 · THE PEOPLE
Who is this answer about?
Tap to flip
ANSWER
Iselin Draper, the AI PM modeling Quaywatch's five-year cost at Northgate Terminals. Farrukh Demir is the engineer who first plotted all five years on one chart.
4 · THE LOAD-BEARING NUMBER
Which assumption moves this answer the most?
Tap to flip
ANSWER
The vendor's annual growth rate. At 12 percent, the crossover lands around month 47. At a capped 5 percent, it doesn't arrive until near the edge of year five.
5 · THE CROSSOVER
In which year does in-house become cheaper than the vendor, cumulatively?
Tap to flip
ANSWER
Year four, around month 47. Vendor stays cheaper through year three, at 12 percent annual growth.
6 · THE NUMBER
Fill in the blank: by year five, the vendor's cumulative cost is about ___ dollars, versus about ___ dollars in-house.
Tap to flip
ANSWER
$540,000 versus $445,000. A gap of about $95,000, still widening at the end of year five.
7 · THE SANITY CHECK
How do you check this estimate in one step?
Tap to flip
ANSWER
Convert both totals to cost per prediction. By year five, the vendor charges about 6.7 cents against an in-house cost near 2.3 cents, roughly three times more for the same job, a believable gap, not an order-of-magnitude error.
8 · CROSS PRODUCT TRANSFER
Section 4 runs BOUND again on a different product. Which one, and how does the outcome differ?
Tap to flip
ANSWER
Wrenmoor Library Consortium's catalog-enrichment tool. At low volume, the crossover never arrives even under an aggressive growth rate, so the honest answer stays buy indefinitely.
Check yourself Score: 0 / 0
Short answer
1. Why does comparing only the vendor's signing price to the in-house estimate give a misleading answer here?
Show hint
Look at "where the estimate breaks."
Show answer
Model answer: Because the vendor's price compounds every year while the comparison only captures year one, when the vendor genuinely is cheaper. The two options only converge, and eventually flip, over a multi-year horizon a single quote can't show.
Multiple choice
2. Which single assumption moves this five-year answer the most?
A. The number of engineers assigned to the in-house build.
B. The vendor's annual price growth rate.
C. The exact wording of the vendor's contract.
D. Which programming language the in-house model uses.
Show hint
Look at the Direction step in the BOUND recap.
Show answer
B. At 12 percent, the crossover lands around month 47. At a capped 5 percent, it barely arrives by year five. Nothing else in the model swings the answer this much.
True or false
3. True or false: at Northgate's actual numbers, the vendor is the cheaper cumulative option for the first three years.
True
False
Show hint
Look at the cumulative-cost calc block.
Show answer
True. Vendor cumulative cost stays below in-house through year three, 287,000 against 355,000. The crossover doesn't happen until partway through year four.
Fill in the blank
4. Fill in the blank: the vendor's cumulative cost crosses above the in-house cumulative cost around month ___, during year ___.
Show hint
Look at the line chart's marked crossover point.
Show answer
Month 47, year four. Roughly eleven months into the fourth year, at the modeled 12 percent annual growth rate.
Short answer, apply it yourself
5. Think of a subscription or vendor cost you know that raises its price every year. Model its five-year cumulative cost against a flat alternative in your head. Does it cross?
Show hint
You don't need exact numbers, just a rough starting cost and a rough annual increase.
Show answer
Model answer: A team paying for a cloud-based translation API noticed its price had risen about 8 percent a year for three years running. Modeled forward five more years, the compounding total came out higher than hiring a part-time contractor to run an open translation model instead, a crossover nobody had modeled until it was pointed out.
Short answer, work the number
6. If Northgate's volume doubled to 4 million predictions a year, with all other numbers unchanged, would the crossover happen earlier or later than month 47?
Show hint
Think about which side of the comparison scales with volume, and which side doesn't.
Show answer
Model answer: Earlier. The vendor's cost scales directly with volume, doubling every year's payment, while the in-house upfront build cost is largely fixed regardless of volume, so the crossover would arrive sooner, not later.
Before you close the answer
Why this works
Tests whether you model a vendor's price as a compounding number over the real life of the decision, instead of comparing a single quote to a single build estimate and stopping there.
Follow-up traps
"Couldn't you just renegotiate every year instead of building in-house?" Response: that's a real option, and it's exactly why negotiating a rate cap upfront is ranked above building in the priority list, the trigger should be tested against renegotiation before migration, not skip straight to a rebuild.
"What if the vendor's price actually goes down as their product matures?" Response: model that too, it's the same equation with a negative growth rate, and in that case the crossover may never arrive, which is itself a real, checkable answer, not an assumption to dismiss.
If pressed
The model didn't account for switching cost at the crossover point, migrating off the vendor after four years of integration typically costs an extra 15 to 20 percent on top of the in-house build estimate, which would push the true crossover a few months later than the raw cumulative-cost lines suggest.
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.
