Most AI projects fail before a single line of code gets written. They fail at the readiness stage — the period when a company decides they're going to use AI, picks a vendor, and starts a project without having answered the questions that determine whether the project can succeed. By the time we're called in to fix the resulting mess, the answer is usually: you skipped step one, and now we have to redo step one while the project is already burning money.
This post is the 10-question readiness audit we run before any engagement. You can run it yourself in an afternoon. If you can't answer 'yes' to at least seven of the ten, you are not ready to start an AI project — and any vendor who tells you otherwise is selling you something.
Data readiness (questions 1-4)
1. Do you have the data?
Obvious but frequently skipped. 'We want an AI to answer customer support questions' requires a corpus of support docs. 'We want AI to recommend products' requires product metadata, user interaction history, and enough scale to find patterns. If you don't have the data, the first 8 weeks of the project will be data acquisition, and the vendor's AI experts will be doing data work they're overqualified for. Pay a data engineer for this phase, not an AI specialist.
2. Is the data accessible?
Data existing isn't enough. Is it in a format and location where an AI system can reach it? Locked behind a legacy ERP with no API counts as 'not accessible.' Scattered across 40 Google Docs counts as 'not accessible.' Spread across a Salesforce, a Zendesk, and three spreadsheets counts as 'partially accessible, needs integration work before AI work begins.'
3. Is the data clean enough?
AI is not a substitute for bad data. Dirty data produces dirty AI. Ask: are duplicates handled? Is PII redacted where needed? Are fields consistent across sources? If the honest answer is no, the first project deliverable is a data cleaning pipeline — which is fine, but it should be budgeted as its own workstream.
4. Can you explain the data lifecycle?
Where does new data come in? How often? Who owns updates? How is deleted data handled? AI systems amplify data lifecycle problems. If you can't explain the lifecycle cleanly, you'll discover gaps in production six months after launch — when they're expensive to fix.
Problem readiness (questions 5-6)
5. Can you describe the problem in one sentence?
'We want to use AI to improve customer experience' is not a one-sentence problem. 'We want AI to answer 80% of tier-1 support questions without escalation while maintaining 90% CSAT' is. Vagueness at the problem stage multiplies into chaos at the delivery stage. If you can't get to a specific, measurable problem statement, the engagement is not yet ready to start.
6. Do you have a baseline?
How do you currently do this thing? What does it cost? What's the quality? Without a baseline, you can't measure AI's impact. Teams that skip baseline discovery end up in the 'did AI work?' debate six months post-launch with no numbers to settle it. Spend a week establishing baseline metrics before anyone writes code.
Team readiness (questions 7-8)
7. Is there an internal owner?
Someone on your team needs to own this. Not 'be involved.' Own — make decisions, review output, unblock when stuck. Without an internal owner, every decision routes through the vendor, the engagement slows to the vendor's iteration speed, and internal stakeholders drift. The owner doesn't need to be technical, but they need authority and availability.
8. Does the engineering team have capacity to productize?
The vendor will ship the AI. Your team will integrate it, maintain it, monitor it, iterate on it. If your engineering team is at 100% capacity on other priorities, the AI system will ship, perform acceptably for two months, and then slowly decay as integration debt accumulates. Budget 15-30% of an engineer's time for steady-state AI ownership. Our cost modeling post explains this in detail.
Infrastructure readiness (questions 9-10)
9. Can you run (or pay for) the required infrastructure?
LLM API costs, vector DBs, monitoring, evaluation infrastructure, possibly GPU. Our TCO post walks through the categories. For production-grade deployments, budget $5-50K/month in ongoing infrastructure beyond the headline LLM API cost. If your budget is smaller than that, scope smaller.
10. Do you have a compliance path?
Data privacy law, industry regulation (HIPAA, SOC 2, PCI, sector-specific), internal data governance. AI systems touch data in new ways. If your compliance team hasn't weighed in, they will weigh in later — often after launch, often expensively. Get legal and security involved before the first sprint. Our security page covers how we approach this for regulated clients.
The scoring
- 10/10: You are ready. Start. Pick a vendor using the vendor selection framework.
- 8-9/10: You are mostly ready. Resolve the open items in the first two weeks of the engagement.
- 7/10: Borderline. Scope a smaller pilot to resolve the unresolved items before committing to a full build.
- < 7/10: You are not ready. Spend 4-8 weeks closing gaps before starting any AI project. Starting now will hurt.
The honest version
Most companies who answer honestly land at 6 or 7 on first read. That's okay — it's the right time to find out, before money is committed. The readiness gaps are almost always fixable in 4-8 weeks by a focused effort: a data engineer to clean and pipe data, a product manager to define the problem crisply, a senior engineer to validate capacity. After that prep, the subsequent AI engagement runs 2-3x faster than it would have otherwise and ships with 80% fewer surprises.
We offer a paid pre-engagement audit service that runs this process with deliverables — gap analysis, prioritized fixes, sample data review, baseline metrics, and a readiness scorecard — in two to four weeks. Most clients who do it don't engage us immediately after (they address gaps first), but the ones who do see engagements run measurably smoother than our benchmark.
The audit format
We've published a free version of the readiness audit as an interactive tool that walks through the same 10 questions with scoring. It's the exact framework above, just packaged as a questionnaire. Takes 15-20 minutes. Delivers a report you can take to your team and your CFO.
An AI project that starts prepared beats one that starts fast. Speed comes from readiness, not from skipping readiness.
Closing
The most valuable pre-engagement conversation we have with any client is the readiness conversation. It surfaces problems that would otherwise be discovered in week 6 — at which point they cost 10x more to fix. Run the audit on yourself before talking to any vendor. If you score low, fix those things first. If you score high, you'll get far more value from any vendor you pick, including us. If you want help running through the audit, book a 30-minute call — no pitch, just a walkthrough.