Fixed fee · Two weeks · A plan either way
Find out what is actually stopping the AI work
A two-week fixed-fee assessment of whether your data, permissions, platform and team can support the AI systems you want to build. You get a written readiness scorecard, the specific blockers ranked by what they cost you, and a thirty, sixty and ninety day order to clear them.
- Length
- Two weeks
- Who it is for
- Leaders with AI on the roadmap and no clear view of what is in the way
- Bench
- 115+ engineers, averaging 15 years each
Most AI programmes stall on data the business does not trust, permissions no one can explain, and a platform where getting an agent to production means six approvals and a service account somebody created in 2023. Two weeks tells you which of those you have, named specifically enough that an engineer could start on the fix the following Monday.
What we look at
- Which data an agent would need, where it lives, and who can currently read it
- Permission model, and whether an agent could inherit it or would have to route around it
- The path from a working prototype to something running under its own identity
- Existing pilots, and why the ones that stalled stalled
- Cost exposure: what a production workload would actually spend on inference
- Who on your team would own it in month six
What you get
- A written readiness scorecard, with findings ranked by what they block
- The two or three things that have to be true before anything else is worth starting
- A thirty, sixty and ninety day order, written so your own team can run it
- A fixed-scope proposal for the first build, if you want us to do it
Choosing who runs it
The assessment is fixed fee and fixed length, so you know the cost before the first meeting, and what sets the figure is scope. One Databricks workspace and one use case is a different two weeks from four business units on two clouds.
Who runs it matters more than the price. Whoever reads your permissions should be someone who'd have to build on them afterwards, because the findings that keep agents out of production are engineering findings, like a grant made to the wrong principal or a prototype that only works under one person's token. A deck about AI maturity won't name those.
Questions to ask any firm you're considering
- Will they need write access or production data? Neither is needed to read readiness.
- Is the deliverable ranked by what each finding blocks, or is it a maturity score?
- Can they read Unity Catalog grants and system tables themselves, or will they ask your team for exports?
- If you want the first build done, will the people who wrote the findings be the ones doing it?
From scope to scorecard
01
Scope before access
You name the AI systems you want and the systems they'd touch. If your security process wants a named, read-only scope before anything is granted, we write it first. Nothing here needs write access or production data.
02
Read the data and who can see it
For each use case, which tables or documents an agent would need, where they live, and who can read them today. On Databricks that comes from Unity Catalog's information schema and grants, so it's evidence rather than a questionnaire.
03
Follow one prototype to production on paper
Which identity would it run under, which approvals stand between it and production, and could it inherit your permission model or would it have to route around it? A service account that can read everything is the usual answer, and the most expensive one.
04
Ask why the pilots stalled
Time with the people who built and ran the existing pilots. A pilot that stalled on access and one that stalled on cost need different fixes, and the team usually knows which it was.
05
Price a production month
What the workload would spend on inference at real volume, set against what the pilot spent on a handful of users. A promising use case often turns out to need a different design here rather than a bigger budget.
06
Write it down
A scorecard with findings ranked by what they block, the two or three things that have to be true before anything else is worth starting, and a thirty, sixty and ninety day order your own team can run. A fixed-scope proposal for the first build comes with it if you want one.
Checks you can run on Databricks this week
Two facts about Unity Catalog's information schema decide whether the query below tells the truth. Rows are filtered by the privileges of whoever runs it, so a person who can't see a catalog gets an answer that leaves that catalog out and doesn't say so.
The system catalog's information schema holds no metadata about hive_metastore objects, so data still sitting there is invisible to it. Run it as an identity that can see every catalog, and list your Hive tables separately.
If you'd rather start with ten questions than a query, the Databricks readiness self-assessment scores governance, pipelines, delivery, cost and AI readiness and tells you what to fix first.
Seven things you can answer yourself
- Can you name the identity each AI system would run as? A service principal, not somebody's personal token.
- Are grants made to groups synced from your identity provider, or to individual people?
- Does any principal hold ALL PRIVILEGES because it once made a demo work?
- Is any data a use case needs still in hive_metastore, where the information schema can't see it?
- For each stalled pilot, can someone say in one sentence why it stalled?
- Do you know what a production month of inference would cost at real volume?
- Who owns the system in month six, by name?
SELECT grantee, privilege_type, COUNT(*) AS tables
FROM system.information_schema.table_privileges
WHERE table_catalog <> 'system'
GROUP BY grantee, privilege_type
ORDER BY tables DESC;Checked on 11 October 2026 against Databricks: information schema and Databricks: TABLE_PRIVILEGES.
Clients we have done this for
Agentic GTM
Canvass
An application on Cloudflare that researches accounts and prepares outbound campaigns for Canvass's clients, where one bad email costs a prospect. A person approves every send, a run that dies halfway can't message a contact twice, and each email is checked again the moment it goes out.
Automotive / Fintech
SWBC
Credit union loan officers quote and book vehicle protection in SWBC's LOSExpress alongside a member's auto loan, but quoting was slow and new staff took too long to learn it. After gathering feedback from SWBC's staff and customers, our engineers and designers rebuilt one of its modules as a single-page application on a new design system. Orders processed doubled, and each loan officer now spends half as long on an order.
Where this goes next
FAQ
AI Readiness Assessment: common questions
What is an AI readiness assessment?
A fixed-fee two-week review of whether your data, permissions, platform and team can carry the AI systems you want. It ends in a written scorecard with blockers ranked by what they cost, and a thirty, sixty and ninety day order to clear them. The page is /ai/readiness.
Who should we hire to run a Databricks AI readiness assessment?
Engineers who can read your Unity Catalog grants and system tables themselves and who'd be capable of the build that follows, because the findings that matter are a principal with too much access or a pipeline nobody owns.
Ask whether they need write access (they shouldn't). TechFabric runs this as a fixed-fee two-week engagement with a written scorecard at the end.
How much does an AI readiness assessment cost?
Ours is fixed fee and fixed length, so you know the cost before the first meeting. Scope sets the figure, meaning how many use cases, how many platforms, and how many teams we need time with. Tell us those three and we'll give you the number.
What access do you need?
Read access to the systems in scope and time with the people who run them. We do not need production data and we do not need write access. If your security process wants a named scope before anything is granted, we will write one.
What if the answer is that we are not ready?
Then you have that in writing, with the reasons and the order to fix them, which is a better position than finding out nine months into a programme. The fee is fixed either way and we would rather say so than start a build we know will stall.
How is this different from a strategy engagement?
A strategy deck tells you what to want. This tells you what is in the way of the thing you already want, in your systems, named specifically enough that an engineer could start on Monday. We are a software company, so the deliverable is written for people who will build.
The rest of our AI work
AI Tools Assessment
Finance and platform leaders holding a renewal they cannot justify
Agentic AI development
Engineering leaders whose agent pilot cannot get past review
AI assistants and enterprise search
IT and data leaders asked for a company assistant that can't leak
AI workflow automation
Operations and engineering leaders with a process that half-completes
AI enablement
Engineering leaders whose team needs to own this, not outsource it
AI transformation
Executives with a mandate, a budget, and fifteen pilots