An AI readiness assessment for an engineering firm has almost nothing to do with AI. It measures three things: whether your data is reachable, whether your tools and governance can absorb a new system, and whether your people have any capacity to change how they work. Twelve questions, scored 0 to 2, one afternoon. The total will not tell you whether to adopt AI. It tells you what will break first when you try, which is the more useful answer.

I run this with firms before anyone opens a vendor demo, because the failure pattern is consistent. Firms do not fail at AI because they picked the wrong tool. They fail because the tool landed on a foundation that could not hold it: files nobody can find, no owner for software decisions, and project managers with no room in their week to learn anything.

Why most AI readiness assessments are theater

Most assessments you will be handed are sales instruments. They ask about your appetite for innovation, your digital maturity, your willingness to embrace transformation. Every question has an obvious correct answer, and the score always lands in the band that recommends the assessor's product.

A real assessment is uncomfortable to fill out. It asks whether you could produce the last ten examples of a given deliverable in under an hour, and most firms discover they cannot. It asks who is allowed to approve a new piece of software, and the room goes quiet. Those silences are the finding.

Score each question 0, 1, or 2. Zero means no. One means partially, or in some offices but not others. Two means yes, and you could demonstrate it this week. Be honest rather than generous; a flattering score costs you a quarter.

Data: can the tools reach your work?

Every useful AI application in an engineering firm depends on reaching your own material, past proposals, standard details, spec sections, report language, closed-out project files. A model with no access to your work can only give you generic output, which is the thing your clients are least willing to pay for.

  1. Are your active project files in one findable place, or scattered across server shares, personal drives, inboxes, and desktops?
  2. Is your reference material digital and searchable: past proposals, standard details, spec libraries, report templates?
  3. Do you know which client and project information is confidential, and is that written down somewhere your staff can act on?
  4. Could you assemble the last ten examples of any given deliverable in under an hour?

Question 3 is the one firms underrate. If nobody has classified what is confidential, every AI decision becomes a judgment call made by whoever is on deadline, and that is how client data ends up pasted into a personal account. Answering it is a prerequisite for the usage policy that governs the tools, not a task that follows it.

Tools: is there a foundation to build on?

This axis is about whether your firm can adopt, control, and eventually retire a piece of software without a fight. It has more to do with your operations than your IT stack.

  1. Do you have an approved way for staff to use AI today, or is usage happening anyway on personal accounts?
  2. Does someone in the firm own software decisions, with a budget and the authority to say yes or no?
  3. Can your IT setup grant and revoke access cleanly when people join, move between offices, or leave?
  4. Have you measured any workflow, even once? Hours per proposal, days per submittal review, anything with a number attached.

Question 5 is nearly always a 0 or a 1, and the honest answer is that shadow usage is already happening. That is not a scandal. It is evidence of demand, and it is more useful to you than any survey. The engineers who found a way to use these tools without permission are the people who should be in your first pilot.

Question 8 deserves particular attention, because without a baseline you cannot prove anything later. A pilot with no before-measurement produces an argument, not a result. If you score zero here, fix it first. It costs a week and it determines whether the whole effort is defensible to your partner group.

People: will anyone actually use it?

Tools do not change firms. Habits do. This is where most readiness scores collapse, and it is the axis vendors never ask about.

  1. Is there at least one respected senior technical person, not just the innovation champion, who wants this to work?
  2. Do your project managers have any slack to learn, or is utilization pressure so high that training reads as punishment?
  3. Has the firm ever changed a workflow successfully before? New project management software, a new QA process, anything. Adoption is a muscle.
  4. Can your people say "the tool got this wrong" without it becoming an argument about whether the initiative is failing?

Question 9 separates pilots that spread from pilots that stall. An enthusiastic champion with no technical standing gets treated as a hobbyist. A skeptical senior engineer who becomes convinced is worth more than any rollout plan, because juniors calibrate their behavior on what senior staff actually do rather than what leadership announces.

Question 12 is the quietest and the most predictive. If reporting a bad AI output feels like criticizing the boss's project, nobody reports anything, and errors travel downstream into sealed work. The firms that get this right treat found errors as the system working. That distinction is the whole subject of how these tools fail on construction documents.

Reading your score

18 to 24. You are ready, and your risk is moving too slowly rather than too fast. Pick the highest-payoff workflow and run a real pilot with a measured baseline this quarter. Firms in this band usually underestimate how far ahead of their competitors they are.

10 to 17. Typical, and workable. Run a narrow pilot in your strongest area while you repair the lowest-scoring items in parallel. Do not wait for a perfect score; firms that wait for perfect never start. Choose the pilot workflow deliberately rather than by enthusiasm, the boring back-office workflows almost always beat design.

Below 10. Do not buy tools yet. Spend a quarter on four things: consolidate files into one findable location, write a one-page usage policy, baseline two workflows with real numbers, and recruit one senior technical sponsor. That quarter will double the return on everything after it. Buying software first, in this band, reliably produces an expensive pilot that quietly dies and makes the next attempt harder.

The question that is not on the list

There is a thirteenth question I ask out loud rather than on the form: what happens to the hours you save?

If the answer is "we bill less," adoption dies, because nobody volunteers to shrink their own utilization. If the answer is "we take on the pursuit we have been declining," or "we stop losing Thursdays to submittal logs," or "our early-career staff spend that time on engineering instead of formatting," people move. The score tells you what will break. This question tells you whether anyone will care enough to fix it.

I would rather work with a firm that scored an honest 11 and has a senior engineer who wants this, than one that talked itself into a 19.

Score the twelve. Then take the two lowest and fix them before you take a single vendor call. One afternoon of honest answers is the cheapest AI investment your firm will make this year, and it is the one most firms skip.

If you want a second read on your score, or you would rather have the assessment run by someone who has done it across firms, our AI readiness and roadmap engagement does exactly this and ends with a phased plan rather than a recommendation to buy something. Start a conversation and we will spend thirty minutes on your two lowest scores.

← Back to insights