Internal AI Assessment: What to Map Before You Call a Vendor

Before you call a vendor, the most expensive thing you can do is skip your homework. An internal AI assessment, run in-house over 1-2 weeks, puts you in a completely different position: you know where AI makes sense, where it doesn’t, and you have the numbers to judge quotes instead of being at their mercy.

This guide is the operative method to do it yourself, at zero cost, before spending a euro. It doesn’t replace a partner’s paid assessment (that comes later), but it decides whether that’s worth doing, and on what.

Why run the internal assessment first

Most European SMEs call a vendor asking for “AI consulting” in the abstract. It’s the most expensive mistake: with no defined use case and no baseline, you have no yardstick for anything. You end up comparing a €10k quote against a €50k quote “for the same scope on paper” with no idea which is right.

An internal assessment solves three problems:

1. It gives you a concrete use case instead of “we want to do AI”. A serious vendor literally can’t quote without this.
2. It gives you a measurable baseline (hours, costs, volumes), the number against which you’ll judge any promised result.
3. It exposes the processes that aren’t ready before you pay for them. A process running on data scattered across spreadsheets and knowledge in people’s heads costs far more to automate, and sometimes isn’t worth it.

This is the same work a good partner would do in their paid assessment. Doing it first, in a lightweight form, saves you time and makes you a much harder client to fool. On how to actually pick the partner afterward, the guide on AI consulting covers criteria and red flags.

What you need (and what you don’t)

You need: 1-2 weeks of diffuse time (not full-time), involvement from department heads, and the willingness to time real processes instead of trusting memory-based estimates.

You don’t need: a tech lead, a budget, or technical AI knowledge. This is an operations and management exercise, not an engineering one. The question isn’t “which model do we use”, it’s “where are we burning hours on repetitive, rule-based work”.

When NOT to do it yourself: if you’re a company with very complex, multi-department processes and you want to start at scale immediately, skip to the paid version. The internal assessment is for those who want clarity before committing, not for those who’ve already decided and have budget for a detailed plan.

The 5 steps of the internal assessment

Step 1: Map the candidate processes

List your company’s repetitive processes, department by department. Not creative projects, not strategic decisions: the processes that repeat with predictable rules. Typical SME examples:

• CV screening and first candidate contact (recruitment)
• Data extraction from invoices and bank reconciliation (finance)
• Inbound lead qualification and first-level replies (sales)
• Ticket triage and recurring answers (customer support)

For each one, note: who does it, how many times a month, how long per run.

Step 2: Time the baseline (real numbers, not estimates)

This is where the serious separate from the wishful. For the 3-4 most promising processes, time 3-5 real runs. Don’t ask “how long does it take you?” (the answer is always off by ±50%). Observe, or have times logged for a week.

Then compute: runs/month × average time = hours/month. That’s your baseline. It’s the number a vendor will have to commit to improving, and the number against which you’ll measure ROI.

Step 3: Check the data under each process

A process is cheap to automate only if the data is accessible. For each candidate, answer:

• Where does the data live? ERP (TeamSystem, Zucchetti, Odoo), email, PDFs, scattered spreadsheets?
• In what format? Structured (database, API) or unstructured (non-standard PDFs, photos, free text)?
• Is it accessible? Is there an API, an export, or does it depend on knowledge in one person’s head?

Rule of thumb: structured and accessible data = cheap automation. Scattered data, non-standard PDFs, or tacit knowledge = high cost and more risk. This doesn’t kill the project, but it changes the quote, and it’s better to know it beforehand.

Step 4: Separate the rule-based part from the judgment part

Inside every process there’s a repetitive, rule-based part (an AI candidate) and a part that needs human judgment (stays with people). A good AI agent handles the first and escalates the second.

Example: in CV screening, AI can read 100k+ candidates and filter on objective criteria (as in the APraise case, where the agent handles 100k+ candidates, equivalent to roughly 4 extra recruiters), but the final call on a borderline candidate stays with the recruiter. Estimating this percentage tells you how much of your wasted time is genuinely recoverable.

Step 5: Prioritize with the impact/feasibility matrix

Plot the processes on two axes: recoverable hours (impact) and data accessibility + % rule-based (feasibility). Start with a single process in the high-high quadrant.

Don’t start with more than one process. The guide on AI consulting costs explains why narrow scope is almost always the right call for a first agent.

The final document: 1 page

At the end of the internal assessment you should have one page containing:

1. The chosen priority process (one).
2. The timed baseline (hours/month, runs, average time).
3. Where the data lives and in what format.
4. The estimated % of rule-based work vs human judgment.
5. The target you’d want to reach (e.g. “recover 40 of the 60 hours/month”).

With this page in hand, the conversation with a vendor changes nature. You no longer ask “what can you do for us”: you present a defined problem with numbers, and you see who replies with a serious plan versus who replies with marketing slides.

If you want an even faster read on your company’s AI maturity before starting, our 3-minute check-up is a good starting point. And if you then want to hand the deep assessment to whoever builds, our AI agents start from a paid assessment (refunded if you proceed) with a baseline timed by the team.

Important note

This is general operative guidance, not an analysis of your specific case. Every company has constraints that don’t show on paper: real data quality, governance, cash flow, internal resistance to change. A lightweight internal assessment gets you far, but it doesn’t replace the opinion of someone who’s seen dozens of similar cases.

If you want an honest read on your case, with no pitch and no surprise quote, talk to us for 20 minutes. If, after seeing your numbers, we think the right answer is “not yet” or “an off-the-shelf tool is enough”, we’ll say so. We’d rather tell you the truth than sell you a sprint you don’t need.