An AI readiness assessment is a structured evaluation, typically two to six weeks long, of whether a company's data, systems, people and processes can support AI, and where it would create value first. The output is a ranked list of use cases scored on value and feasibility, an honest read on your data, and a roadmap for the next twelve months. In Canada, a facilitated assessment from an external consultant typically costs $10,000 to $50,000 CAD, based on published market rates.
That five-figure price tag exists to prevent a six-figure mistake. A significant share of AI projects fail not because the technology doesn't work, but because the company's data couldn't feed it, the process around it never changed, or the use case was chosen by enthusiasm rather than economics. An assessment is how you find that out for thousands of dollars instead of hundreds of thousands. This guide covers what a real assessment includes, how to run one, what the deliverable should look like, and how to spot the ones that are sales funnels in disguise.
What it is, and what it is not
A readiness assessment answers three questions. Where could AI plausibly create value in this specific business? Can the company's data, systems and people actually support those use cases today? And in what order should the work happen? It's diagnostic work: interviews, system inventories, data sampling, and use-case scoring, done against your operations rather than against a generic maturity model downloaded from the internet.
It is not a sales audit for a predetermined tool. If the conclusion was written before the first interview, you didn't buy an assessment, you sat through a demo with extra steps. It's also not a strategy deck about what AI means for your industry. Industry context matters, but a deliverable you could swap onto a competitor's letterhead without editing has told you nothing. The test of a real assessment is specificity: it names your systems, quotes your people, and says no to use cases that don't fit.
The five dimensions a serious assessment covers
Vendors slice the pie differently, but a credible assessment covers the same five dimensions one way or another:
- Data. What you have, where it lives, who owns it, how clean it is, and whether it can legally and practically be used for the use cases on the table. This is where most assessments find the bad news, and finding it now is the point.
- Infrastructure and systems. The ERP, CRM, MES or homegrown tools your operations run on, what they can integrate with, and what's stuck behind a vendor's closed API or a server under someone's desk.
- People and skills. Who would build, operate and maintain AI systems, what skills exist in-house, and how the people doing the work today actually feel about the prospect. Resistance discovered in week three of an assessment is cheaper than resistance discovered in month six of an implementation.
- Process. How work flows today, where the bottlenecks and manual handoffs are, and which processes are stable enough to automate. Automating a broken process gets you a faster broken process.
- Governance and risk. Privacy obligations, regulatory constraints, security posture, and who decides what an acceptable AI output looks like. Frameworks like the NIST AI Risk Management Framework give this dimension structure without requiring an enterprise compliance department.
How to run one, step by step
Whether you run it internally or hire help, the sequence is the same. The order matters: companies that start with the technology and work backwards toward a business case tend to produce impressive pilots that nobody asked for.
- Define business goals first. Before anyone says the word model, write down the two or three business outcomes that matter this year: margin, throughput, customer response time, whatever moves your numbers. Every use case gets judged against these, and anything that can't be tied to one gets parked.
- Inventory data and systems. List the systems your operations run on, what data each one holds, how it's accessed, and how reliable it is. Pull real samples. A field that's technically in the schema but empty in practice is a finding worth having in writing.
- Interview the people doing the work. Not just the executives. The dispatcher, the estimator, the customer service lead. They know where the duplicate data entry, the workarounds and the tribal knowledge live, and that's where the best use cases usually hide.
- Score use cases on value versus feasibility. Plot every candidate on two axes: business value if it works, and feasibility given the data, systems and skills you actually have. High value and high feasibility go first. High value and low feasibility go on the roadmap with the gaps named. Everything else waits.
- Deliver a roadmap with quick wins. Sequence the top use cases over twelve months, name the data and skill gaps that need closing, and include one or two quick wins shippable in weeks. Early visible results buy the patience the bigger projects will need.
What a good deliverable looks like
The deliverable is what you're paying for, so be specific about it before signing. A good one fits in a document your leadership team will actually read, and it contains decisions, not just observations:
- A ranked list of use cases, scored on value and feasibility, with the reasoning shown. Including the ones that were considered and rejected, and why.
- A data gap list: which datasets each top use case needs, their current condition, and what closing each gap costs in time and effort.
- Build-versus-buy calls for the top use cases. Most SMB use cases are better served by configuring proven vendor tools than by custom development, and an honest assessment says so even when the assessor sells development.
- A twelve-month roadmap with sequencing, dependencies, rough cost ranges, and the funding programs each phase might qualify for.
- Named quick wins: one or two items deliverable in weeks, with an owner and a success metric attached.
Notice what's not on that list: a maturity score. A spider chart telling you that you're a 2.3 out of 5 on AI maturity is fine decoration, but it doesn't tell anyone what to do on Monday. If the deliverable's centrepiece is a benchmark against an anonymous peer group, you bought a poster.
DIY or facilitated: which one you need
You can run an assessment internally, and for some companies that's the right call. The honest comparison looks like this:
| Factor | Internal self-assessment | External facilitated assessment |
|---|---|---|
| Cost | Mostly staff time, often underestimated | $10,000 – $50,000 CAD at published market rates |
| Time | 6 to 12 weeks part-time, competing with day jobs | 2 to 6 weeks, full attention |
| Objectivity | Hard. Every department lobbies for its own pet project | Easier to say no, easier to deliver bad news upward |
| Blind spots | You don't know what other companies' data problems look like | Pattern recognition from past projects, but needs time to learn your business |
| When it's right | You have in-house data or AI expertise and an executive sponsor with time | First serious AI initiative, no in-house expertise, or internal politics around use-case selection |
A middle path works well for many mid-sized companies: an internal lead runs the inventory and interviews, an external consultant designs the framework, pressure-tests the findings and scores the use cases.
The strongest argument for outside help isn't expertise, it's permission. An external assessor can write that the CRM data is unusable or that the most senior manager's favourite project doesn't clear the bar. Internal teams know these things too. They just can't always say them in a document with their name on it.
Red flags in assessment offers
The assessment market has its share of bait. Three patterns are worth walking away from:
- Free assessments that funnel into one vendor's product. If the assessment costs nothing and the assessor sells a platform, the conclusion was reached before you signed. A real assessment costs money precisely because it's allowed to disappoint you.
- Assessments longer than eight weeks. Past that point you're funding analysis paralysis, or a discovery phase priced as a strategy engagement. Two to six weeks covers a mid-sized company. If a provider quotes four months, ask what they're doing in month three.
- Deliverables that are all slideware. If the sample deliverable has industry statistics and maturity quadrants but no evidence anyone queried a database, you're buying a presentation, not an assessment. Ask to see a redacted data gap list from a past engagement. A provider who never produced one can't show you one.