Skip to main content
AI // Readiness

Before you buy the AI, find the job it should do.

A focused assessment that lists candidate use cases, checks whether your data can support them, estimates value and effort, and recommends a first pilot.

inference.tracelive

{

"summary"
: "Use cases ranked by value and feasibility, a data reality check, and one pilot to start.",
"format"
: "Workshops + data review",
"output"
: "Ranked roadmap",
"next_step"
: "One scoped pilot",
"reviewed_by"
: "a human"

}

How the assessment runs

  1. Step 01

    Workshops

    Interview teams about repetitive, rules-heavy, or document-heavy work.

  2. Step 02

    Data review

    Check whether the data needed exists, is accessible, and is good enough.

  3. Step 03

    Score

    Rate each use case for value, feasibility, and risk.

  4. Step 04

    Roadmap

    Recommend a first pilot with success metrics and a rough budget.

How we score use cases

CriterionQuestion
ValueHow much time, cost, or error does this remove?
VolumeDoes it happen often enough to matter?
DataIs there enough clean, accessible data?
RiskWhat happens when the AI is wrong?
FitCan results flow into existing tools?

Why start with an assessment

Many AI projects begin with a tool and look for a problem. The demo impresses, the pilot starts, and months later nobody can say what it improved. The usual causes are predictable: the use case was too vague, the data wasn't there, or the result had nowhere to go in the existing workflow.

An AI readiness assessment reverses the order. It starts from the work your teams do today, finds the tasks where AI can save time, reduce errors, or unlock something new, and checks each one against the data and systems you actually have. The output is a short list of opportunities ranked by value and feasibility, with an honest note on the ones that aren't worth doing yet.

It is deliberately short and practical. The goal is a decision about a first pilot, not a strategy document that sits on a shelf.

What the assessment covers

Work and pain points

Where time goes, where errors happen, and which tasks are repetitive or document-heavy.

Data reality

What data exists, where it lives, how complete it is, and who can access it.

Systems and integration

Where results would need to land: ERP, CRM, helpdesk, or a new tool.

Risk and compliance

Sensitivity of the data and the consequences of AI mistakes for each use case.

Value and effort

Rough benefit and build effort, so options can be compared fairly.

First pilot

One recommended pilot with scope, success metrics, and an evaluation plan.

Deliverables

DeliverableWhat it contains
Opportunity listEvery candidate use case with a short description and owner
Scoring matrixValue, volume, data, risk, and fit scores for each
Data findingsGaps, quality issues, and access questions to resolve
Pilot briefScope, success metrics, evaluation set, and rough budget
RoadmapSequenced next steps for the following two or three use cases

Good and poor first pilots

Good first pilots

  • High volume, repetitive tasks
  • Clear right and wrong answers
  • Data already accessible
  • Results land in an existing workflow

Poor first pilots

  • Rare, high-stakes decisions
  • Vague goals like 'use AI for insights'
  • Data that must be collected from scratch
  • No owner for the result

To prepare for the workshops

  • A list of teams and a contact in each
  • Rough volumes for document-heavy or repetitive tasks
  • Access to sample data or exports
  • Any existing AI tools or experiments in use

Common pitfalls we help you avoid

The first pitfall is choosing a pilot because it is exciting rather than useful. Flashy use cases often lack data or an owner, while modest ones such as document processing deliver value quickly and build confidence.

The second is skipping the data check. Many good ideas fail because the data is scattered, incomplete, or locked in systems without APIs, and finding that out after the build starts is expensive.

The third is having no success metric. Without a baseline and a target agreed before the pilot, results become a matter of opinion rather than evidence.

What we measure

MeasureWhy it matters
Use cases assessedBreadth of opportunities considered
Data readinessWhether each use case has usable, accessible data
Expected valueTime, cost, or error reduction estimated per use case
Pilot readinessScope, metric, owner, and evaluation set agreed

Who does what

We handle

  • Workshops, analysis, and scoring
  • Data review and findings
  • Pilot brief and roadmap

Your team handles

  • Access to people and sample data
  • Choosing the pilot
  • Budget approval for next steps

A worked example

Imagine a mid-sized manufacturer that wants to “use AI” but doesn't know where to start. Workshops with finance, purchasing, and quality find a dozen ideas. The data review shows that invoice processing and quality inspection have the data and volume to succeed, while a predictive maintenance idea lacks sensor history.

The assessment recommends invoice processing as the first pilot, with a baseline of current processing time and a target agreed with finance. Predictive maintenance moves to the roadmap with a plan to start collecting sensor data.

Before you start

  • A sponsor who can approve a pilot
  • Contacts in each team involved
  • Sample data or exports you can share

What success looks like

Success is a decision, not a document: a chosen pilot with an owner, a metric, and a budget, plus a short list of what comes next and what isn't worth doing yet.

Structure for the risk conversation

Readiness isn't only about data and use cases; it's about deciding how much risk is acceptable and who owns it. NIST's AI Risk Management Framework, released in January 2023 for voluntary use, and its Generative AI Profile from July 2024 give leadership and compliance teams a shared vocabulary for that conversationSource [1].

Our assessment uses that structure lightly: for each candidate use case, we look at the value, the data available, how errors would be caught, and the controls needed, then rank the options so the first pilot is both useful and safe.

Frequently asked questions

Do we need a data team first?
No. The assessment tells you which use cases work with the data you already have.
What if AI isn't the answer?
Then we'll say so. Sometimes a rule, a form, or an integration solves the problem more cheaply.
Is the roadmap tied to your services?
It's yours to use with any team.
Who should attend workshops?
People who do the work daily, plus someone who owns the budget.
What does a pilot look like?
One workflow, a defined evaluation dataset, and success metrics agreed before building.
How long does the assessment take?
Typically a few weeks, depending on the number of teams involved and how quickly sample data can be shared.
Who from our side needs to be involved?
A sponsor, a contact in each participating team, and someone who can grant access to data and systems.
Do you look at tools we already pay for?
Yes. Sometimes the best first step is using AI features already included in software you own.
Will you recommend vendors?
Only where it helps, and we say how we evaluated them. The assessment is not a sales pitch for any product.
What comes after the assessment?
Usually a scoped pilot, which we can build, or your team or another partner can take forward using the brief.
Can the assessment cover just one department?
Yes. A focused assessment of one area, such as finance or operations, is often the quickest way to find a strong first pilot.
Do you need access to production systems?
Read access to samples or exports is usually enough. We agree what's shared and how it's handled before starting.
Is the roadmap specific to AI only?
It focuses on AI, but notes where integrations, data cleanup, or process changes should come first.
Can you run the assessment remotely?
Yes. Workshops work well over video calls, with on-site sessions where seeing the work in person helps.
Do you cover generative AI and classic machine learning?
Both. Some use cases suit language models, others suit forecasting or classification models, and some need no AI at all.
Do we need a formal AI policy before a pilot?
A short, practical one helps. A framework like NIST's AI RMF gives a useful structure without heavy bureaucracySource [1].
What if our data isn't ready?
The assessment identifies what's missing and whether a smaller pilot or a data clean-up should come first.

Sources

  1. [1]AI Risk Management Framework (opens in a new tab)NIST, released 26 January 2023