An AI readiness assessment scores whether your company is legible enough for AI to work: one owner per workflow, one source of truth per data category, and decisions written down where the next person can find them. An AI agent dropped into a company that fails those checks scales the mess.
Zero of the ten points webvise scores before a build ask which model you plan to use.
If you pay for AI tools that nobody has wired into a real workflow, you already know the state this article starts from. Below is the 10-point readiness checklist webvise scores prospects against, adapted from Shann Holmberg, plus the assessment arc that took MP Bau, a Brandenburg construction firm, from discovery call to working portal in about 6 weeks. The failed points are the interesting part, because they become the project scope.
- Readiness is organizational. The 10 points check owners, sources of truth, SOPs, templates, data flows, QA standards, decision logs, and feedback loops. None check models.
- The gaps become the scope. Failed points turn into the first work packages of the build, priced and sequenced before any model choice.
- Validate live before deciding. MP Bau's two flagship workflows ran on real company documents in a half-day workshop before the build decision.
- webvise passes its own checklist. The studio's source of truth is an LLM-maintained wiki with a published 3-month report.
- No is a valid result. webvise sells no AI strategy without a named workflow and builds nothing that decides legal, tax, clinical, or credit questions autonomously.
The score is organizational: owners, sources of truth, written decisions
Most stalled AI projects share one diagnosis. The workflow the agent was supposed to run had no owner, its data lived in three places, and what counts as a good output existed only in one person's head. The model was never the blocker.
A company that passes the ten points below can adopt almost any competent model, because the agent finds named owners, reachable data, and written rules. A company that fails them pays an hourly rate to have its own processes reconstructed from memory before any automation can start.
That reconstruction is the scope webvise defines in an AI consulting engagement: the assessment names the failed points, and those points become the work packages. A model decision on day one is never part of it.
The 10-point checklist, adapted from Shann Holmberg
The list comes from Shann Holmberg's readiness framework. webvise adapted it into a scoring pass: every point is checked against one concrete workflow, one point per passed check, and each failure is noted as a candidate work package.
| # | Point | Pass looks like | Common failure |
|---|---|---|---|
| 1 | One owner per workflow | A named person answers for the outcome | "The team" owns it, so nobody does |
| 2 | One source of truth per data category | Prices, contacts, and project data each live in one system | Three price lists, all slightly different |
| 3 | Structured data | Records have fields a machine can query | Knowledge sits in PDFs and inboxes |
| 4 | Documented SOPs | The steps are written down and current | The process lives in one veteran's head |
| 5 | Templates | Recurring outputs start from a shared template | Every offer starts from a blank page |
| 6 | Tool and subscription tracker | You can list what you pay for and why | Licenses nobody remembers buying |
| 7 | Mapped data flows | You know which system feeds which | Data moves by copy and paste |
| 8 | QA standard per output format | "Good" is defined per deliverable | Quality means nobody complained |
| 9 | Decision log | Decisions carry a date and a reason | The same debate repeats every quarter |
| 10 | Feedback loop | Errors flow back into the SOP | The same mistake ships twice |
Score one point per passed check, for one workflow at a time. The passing points show where an agent can stand today. The failing points are your project scope, and most of them are process work you can start without a vendor.
webvise scores itself first
Point 2 and point 9 are the ones I can show receipts for. webvise's source of truth is an LLM-maintained wiki: clients, decisions, and published content get filed with dates, and agents read the wiki before they write anywhere else.
Three months of running it are documented in the LLM wiki 3-month report. How agents reach that knowledge, and why most companies can skip a vector database at the start, is covered in building a business knowledge base without RAG.
Selling an assessment this way has one reason: symmetry. A checklist the studio applies to itself before applying it to a prospect is harder to argue with than a slide deck.
What a full assessment produced at MP Bau
MP Bau, a construction company from Brandenburg, went through the complete arc. No stage involved a model comparison, and every stage produced something the next one needed.
| Stage | What it produced |
|---|---|
| Discovery call | The candidate workflows, in the owner's words |
| Ranking | 5 use cases ordered by payback and data reachability |
| On-site workshop | Half a day with the people who own the workflows |
| Live validation | Both flagship workflows run on real company documents, in the room |
| Build decision | Scope agreed from the failed checklist points |
| Delivery | Discovery call to working portal in about 6 weeks |
Live validation is the stage most assessments skip. A demo on sample data proves the vendor can demo. The same tool run live on the company's own documents, with the workflow owner in the room, proves the workflow.
This arc is the default entry into webvise's AI consulting: assessment first, build decision second.
Two refusals that are part of the assessment
webvise sells no generic AI strategy without a named workflow attached. A strategy document without a workflow has no owner, which means it fails point 1 of its own checklist on the day it is delivered.
The second refusal covers autonomy. No webvise build makes fully autonomous legal, tax, clinical, or credit decisions: those outputs get a human review gate, or the project changes shape until they do.
'Fix these three points first, then build' is a common assessment outcome, and it is the cheap one. The expensive outcome is a finished agent discovering the missing SOPs one support ticket at a time.
Run the score yourself this week
Pick the one workflow that eats the most hours and walk it through the table above, one point per passed check. Write the score down with a date, because that line is the start of your decision log.
The failing points are your to-do list, and most of them need no vendor. An owner can be named in a meeting. A source of truth is often a decision, and at the harder points, structured data and mapped flows, outside help starts to pay off.
webvise has delivered 25+ projects on this sequence: score first, scope from the gaps, build second. If you want the ten points scored against your own workflows, send the workflow you have in mind through the contact form.
Development practices are aligned with ISO 27001 and ISO 42001 standards.