Two machines are writing your RFP but nobody is choosing your partner

When a large RFP arrives at a technology company today, here is what actually happens. The document: often eighty pages, often visibly drafted with AI, complete with the slightly-too-symmetrical question structure that gives it away is loaded into the vendor's own AI, which drafts the answers. The proposal ships back: three hundred pages of confident prose that no single person on either side has read in full.

Then the buyer's team runs the responses through a scoring matrix increasingly with AI assistance and shortlists the vendors whose machines produced the most compliant text.

Sit with that for a moment. A hotel group is selecting a partner for a five-to-ten-year dependency, touching its most sensitive data and its most important decisions, through a process in which one AI interviews another AI while the humans wait for the verdict. The RFP has become theater performed by machines for an audience that left the building.

I say this with no bitterness. Juyo wins its share of these processes. I say it because the process now systematically selects for the wrong thing, and both sides know it.

What the RFP was built for, and why that world is gone

Be fair to the RFP: it was a rational tool for a different era. When information was scarce, capabilities were opaque, and comparing vendors required extracting structured answers to identical questions, a formal written process created comparability and auditability. Procurement departments exist to prevent favoritism and fraud, and the RFP was their instrument. Legitimate purposes, still legitimate today.

But look at what the written response actually measured, even in its best years: the vendor's ability to describe itself. And look at what it measures now, when generative AI can produce a flawless description of anything: the vendor's prompt engineering. Every "yes, fully supported" costs nothing to write and everything to verify. The feature checklist, the RFP's analytical core has become the least trustworthy artifact in the entire evaluation, precisely because it is now free to fabricate fluently.

Meanwhile, the information scarcity that justified the whole apparatus has inverted. A buyer today can verify a vendor's actual capabilities in days: live products can be demonstrated on real data, customer bases can be interrogated, financials and ownership can be examined, integration claims can be tested against running systems. The facts are cheap now. What remains expensive what has always been expensive is the one thing the written process never touched.

The questions no RFP has ever asked

I have completed hundreds of these documents across a decade. I have answered questions about disaster recovery, SOC 2, API rate limits, and the color depth of exported PDFs. Here are questions I have never been asked, in any RFP, by any hotel company, ever:

How long do your employees stay, and why do they leave? You are not buying the demo team; you are buying the people who will answer the phone in year four. Tenure and turnover tell you more about the next five years of service quality than any SLA appendix.

Tell us about your worst incident, and show us the postmortem. Every vendor has broken production, missed a migration, shipped a bad release. The ones worth marrying are the ones who can narrate their failure precisely, show what changed afterward, and name the customers they called first. A vendor with no story here is either lying or too young to trust.

Give us three customers who left you, and permission to call them. Curated references are a genre of fiction. Churned customers are the documentary. How a company loses a client, gracefully, with clean data handover, or clinging and invoicing is how it will one day lose you.

Who owns you, what did you promise them, and are you profitable? As I have argued elsewhere, the cap table predicts vendor behavior better than the pitch deck. A partner whose survival depends on your renewal treats you differently from a partner whose survival depends on the next funding round.

What happens on day one after termination? Data portability, export formats, transition support. the divorce clause, negotiated while everyone is still in love. Vendors who welcome this question are safe. Vendors who deflect it have told you their retention strategy.

What do you refuse to build, and why? Every serious product company says no constantly. The pattern of refusals reveals the philosophy: is the roadmap governed by conviction about the industry, or by whichever prospect shouted loudest last quarter? You are subscribing to a decade of someone else's judgment. Examine the judgment.

Notice what these questions have in common: none can be answered convincingly by a language model, because each one is verified through behavior, artifacts, and third parties. They are AI-proof because they are about character. And character, in a long dependency, is the load-bearing variable.

Culture fit: the variable everyone feels and nobody scores

Ask any executive who has lived through a failed technology partnership what actually went wrong, and you will almost never hear "the features were missing." You will hear: they didn't understand our business. Their support treated us like a ticket number. Their pace and ours never matched. Every conversation was a negotiation. In other words: the cultures were incompatible and everyone sensed it in the first meeting, and nobody was allowed to score it, because "we felt uneasy" doesn't survive a procurement audit.

This is a solvable problem, and solving it does not mean surrendering rigor to vibes. Culture fit can be evaluated with the same discipline we apply to security questionnaires:

Watch them work, on your data, with your people. Replace the middle of the funnel with a paid working sprint: one or two weeks, real property data, the vendor's actual delivery team (contractually named the people in the sprint are the people on the account), a defined business question to answer together. What a written proposal hides, a sprint exposes within days: how they handle ambiguity, how they say "we don't know," whether your team enjoys thinking with them, whether their product survives contact with your messy reality. Pay them for it — paying keeps the best vendors in your process and filters the ones whose economics only work with free pre-sales theater.

Score the questions they ask you. The quality of a partner is visible in their curiosity. A vendor who spends the first meeting understanding your ownership structure, your operating model, and your definition of success is showing you year three of the relationship. A vendor who spends it presenting is showing you the last relationship they had.

Do the reverse site visit. Spend half a day inside the vendor's company. Sit in on a real support standup. Listen to how they talk about customers when the customers aren't buying. Look at who is in the room and how disagreement travels. Hospitality executives are professionally trained observers of service culture  you can read a hotel's soul from its staff canteen. Use exactly that skill on your future partner; it is the most underused due-diligence instrument in our industry.

Check the speed and honesty of small things. How fast did they answer the odd question outside the process? What happened when you asked something their product genuinely doesn't do? The vendor who says "we don't do that, and here's who does it well" has just passed the most important test in the entire evaluation — you have found an organization that tells the truth under commercial pressure. That trait does not coexist with a fabricated RFP answer.

The redesigned process

Put together, the better process looks like this — and it is faster, cheaper, and harder to game than what it replaces:

Stage one, machine to machine. Stop pretending the factual layer deserves months. Let vendors publish verified, structured capability profiles: certifications, live integrations, architecture, compliance evidence, standardized and machine-readable. Let the buyer's AI screen against requirements in days. This is the one part of the process AI genuinely should own so give it that part completely, and reclaim the human months it currently wastes.

Stage two, humans and evidence. The AI-proof questions above, answered with artifacts and third parties: postmortems, churned references, ownership disclosure, exit terms, refusal philosophy. A half day per vendor, and more predictive than three hundred pages.

Stage three, work together before deciding. The paid sprint on real data with the named delivery team. Selection by observed collaboration. The final scoring weights inverted from today's practice: perhaps a fifth on capabilities — mostly verified by machine — and the great majority on evidence of character and demonstrated fit. Weighted that way because that is how the risk is actually distributed: technology gaps get fixed in quarters; culture gaps compound for a decade.

For the buyers reading this: your procurement function will object that soft criteria can't be audited. They can. Behavior, artifacts, and reference calls generate exactly the paper trail auditors need, and a sprint scorecard is more defensible than a subjective read of vendor prose. What genuinely can't be audited anymore is the authorship of a written answer.

And for my fellow vendors: we bear half the responsibility for this theater, because we keep performing in it. The next time an eighty-page machine-generated questionnaire arrives, consider the braver reply: a respectful note explaining what you will and won't do, the offer of a working sprint, and your churned-customer references attached, unasked. Some buyers will disqualify you. The ones who don't are the ones worth ten years of your best work — and you will have learned more about their culture from one reply than from anything in their document.

The deeper point

Hospitality, of all industries, should have seen this first. We are the people who know that a hotel cannot be judged from its brochure — that truth lives in the arrival experience, the recovery from a mistake, the way a doorman treats a guest who looks like she can't afford the room. We built an entire industry on the understanding that character reveals itself in behavior, under pressure, over time.

Then we turned around and selected our decade-long technology partners by reading their brochures.

The machines have now made those brochures perfect, and in doing so they have done us an accidental favor: the written answer is worthless as a differentiator, so we are finally forced to evaluate what always mattered. Choose your partners the way you would want a guest to choose your hotel by how it feels to be inside, by how mistakes are handled, by whether the people believe what the brand promises.