Formats
- Inventory as CSV and Parquet
- Scores as a spreadsheet and PDF
- Plan as a slide deck and a written memo
Find out which of your data can teach a model before you pay to train one.
In three weeks we inventory the systems your business runs on, sample what is actually inside them, and score every AI use case you are considering against the data that would have to carry it. You leave with a ranked list: what to build, what to fix first, and what to drop.
AI projects rarely stall on the model. They stall on the data underneath it.
Gartner predicts that through 2026, organizations will abandon 60% of AI projects that aren't supported by AI-ready data, and its survey found that 63% of organizations either lack the right data management practices for AI or aren't sure whether they have them.
The pattern behind those numbers is familiar. A pilot runs on a few hundred hand-picked examples and looks promising. Then the team goes looking for the other ninety thousand: they sit in four systems, half of them are scanned, a third mention patients or policyholders by name, and nobody can say who owns the rights to the vendor files. The project doesn't fail in a demo. It fails in a spreadsheet three months later.
A readiness assessment moves that spreadsheet to the start. Before anyone fine-tunes, labels, or buys a GPU hour, you know what each use case would need, what you already have, and what the gap would cost to close.
The assessment at a glance
You provide
- Read access to up to 12 source systems: shared drives, document management, ticketing, CRM or ERP exports, call recordings, data warehouse tables
- Three to eight candidate use cases, each described in a paragraph
- Two hours with a process owner for each use case
We deliver
- Data inventory with owner, volume, format, and sensitivity for every source
- Readiness score per use case across nine dimensions
- Rights and privacy screen with the records that need legal review
- Ranked 90-day plan with effort and cost ranges
Quality gates
- Every score backed by a sample of at least 200 records per source
- Sensitivity findings reviewed by a second analyst
- Plan reviewed with your process owners before the readout
Turnaround
Three weeks from kickoff to readout
Team
A lead data engineer, an annotation lead, and a privacy reviewer
Starts at
$6,800
Three weeks, fixed fee: up to 12 data sources inventoried and sampled, a readiness score for each candidate use case, and a 90-day plan.
How the three weeks run
Most of the work happens inside your environment. Raw records never leave it.
Scope and access (days 1–3)
We agree on the use cases, the source systems, and the access route: a read-only service account in your cloud, a virtual desktop, or an on-site workstation in Jackson or at your office.
Inventory (days 3–7)
We profile every source: record counts, file types, date ranges, owners, duplication, and how much of it a machine can actually read. Scanned pages, embedded tables, and audio get their own measurements.
Sample and score (days 6–12)
Analysts read real samples against each use case. Could a person learn the task from these records? Are the outcomes recorded? Are the hard cases in there, or only the easy ones? Each answer becomes a scored dimension with its evidence attached.
Rights and privacy screen (days 9–13)
We flag personal and health information, contractual restrictions, and third-party content, and we mark the records your counsel should look at before anything is used for training.
Plan and readout (days 13–15)
We rank the use cases by value and by the cost of closing their data gaps, then walk your team through the plan. The readout is a working session, not a presentation.
What we score
Nine dimensions, each scored 0 to 4 per use case, with the sample behind every score.
| Dimension | The question we answer | Why it matters |
|---|---|---|
| Volume | Are there enough examples of the task, including the rare ones? | Fine-tuning and evaluation both need coverage of the cases that go wrong, not just the common ones. |
| Outcome labels | Does the record say what happened next, or who was right? | Without outcomes, every example has to be labeled by hand, which changes the budget. |
| Machine readability | How much is scanned, handwritten, or locked in tables and images? | Parsing quality caps every downstream result, including retrieval. |
| Noise and duplication | How many records are copies, templates, or junk? | Duplicates inflate scores and teach a model to repeat boilerplate. |
| Freshness | Do the records reflect how the work is done today? | Old procedures teach old answers with full confidence. |
| Sensitivity | How much personal, health, or financial information is present? | Determines redaction work and where processing may happen. |
| Rights | Do you have the right to use the records for training? | Vendor files, customer uploads, and licensed content often carry limits. |
| Ground truth for testing | Can we build an evaluation set your experts would trust? | If you can't measure it, you can't tell whether training helped. |
| Ownership | Is there a person who can answer questions about this source? | Data with no owner tends to stay unfixed. |
What you receive
- A data inventory covering every source in scope, with owner, volume, formats, date range, and sensitivity
- A readiness scorecard for each use case, with the records behind each score so your team can check our reading
- A rights and privacy register listing the categories and sources that need legal review before training
- A ranked 90-day plan: what to build first, what data work it needs, and a cost range for each step
- Three example records per use case, cleaned and formatted the way a training set would need them
- A two-hour readout with your technical and business owners, and a written memo for those who missed it
An assessment, not a strategy deck
A typical AI strategy engagement
- Interviews leaders about where AI could help
- Describes the data in categories
- Recommends a platform
- Leaves the data work for a later phase
A Benner House readiness assessment
- Reads samples of the actual records in each system
- Scores each use case against the data it would need
- Is vendor-neutral and recommends no platform by default
- Prices the data work so you can fund it with real numbers
One fixed fee
AI Data Readiness Assessment
Starts at
$6,800
Three weeks, fixed fee: up to 12 data sources inventoried and sampled, a readiness score for each candidate use case, and a 90-day plan.
Request a scoped quoteEstates with more than 12 sources, or assessments that need on-site work in more than one location, are quoted after a 30-minute scoping call. If you go on to a build with us within 90 days, half the assessment fee is credited against it.
Questions before an assessment
Do our records leave our environment?
No. We work through a read-only account in your cloud, a virtual desktop you control, or a workstation on your premises. Our notes contain counts, scores, and short redacted excerpts, never bulk copies of your data.
We haven't chosen a model or a vendor yet. Is it too early?
It's the right time. The assessment is model-neutral, and its findings usually narrow the choice: a use case with strong outcome labels points toward fine-tuning a small model, while one built on large, changing document sets points toward retrieval.
What if the answer is that our data isn't ready?
Then you'll know which gaps matter and what they would cost to close, which is cheaper to learn now than after a pilot. Most assessments find at least one use case that can start immediately and one that should wait.
Can you take the self-assessment results as a starting point?
Yes. If you've used our readiness scorecard, send us the result link and we'll start from your answers.
Do you provide legal advice on data rights?
No. We identify where rights or privacy questions exist and document the evidence. Your counsel makes the call, and we're glad to work alongside them.
Keep reading
Bring us one messy data source and one question.
Tell us what the data is and what you want a model to do with it. A data engineer, not a salesperson, will reply within one business day with questions and a proposed first step.