AI Readiness Assessment

What an AI readiness assessment answers

Most teams that ask where to start with AI already have a list of ideas: a chatbot on the website, smarter search in the app, automatic summaries of service visits, an assistant that drafts replies. The hard question is not which idea sounds best but which one your data, systems and people can support right now. Tepia’s AI readiness assessment exists to answer that question with evidence instead of opinions.

The assessment answers four things. Which problems in your business are a good fit for AI at all, and which are better solved with ordinary software or a cleaner process. Whether the data those problems depend on exists, is reachable through an API or export, and is accurate enough to trust. What could go wrong for customers, staff or regulators if the AI gets an answer wrong. And what the first build should be, sized so you learn something useful within one release cycle.

It is deliberately not a strategy deck. You will not get a maturity score benchmarked against an industry average. You get a short list of use cases, the evidence behind each one, and a recommendation you can act on next quarter. If you want the broader picture first, our AI strategy and implementation consulting page covers how this fits into a longer roadmap.

The five areas Tepia scores

Tepia scores each area from one to five using evidence gathered in interviews and from samples of your real records. A low score is not a failure. It tells you what to fix before you build, or which use case to avoid for now.

The data row usually decides the outcome. Gartner’s research on AI ready data matches what Tepia sees in practice: projects stall when the records a model needs are scattered, stale or locked in a system nobody can query. When that is the finding, the first project is often a cleanup or integration effort, covered on our data preparation for AI projects page.

How Tepia ranks candidate use cases

Tepia lists every candidate use case raised in interviews, usually ten to twenty, and places each on two axes. Value asks what changes in staff time, revenue or customer experience if it works. Feasibility asks whether the data exists, whether current models handle this kind of task reliably, and how hard the integration is. The top right of that grid becomes your shortlist.

A few patterns show up in almost every assessment:

• Strong first candidates: summarizing long records staff already read, classifying incoming requests, drafting replies a person approves, and searching internal documents in plain language.
• Good but data dependent: recommendations, forecasting and anything that needs a history of outcomes the business may never have captured.
• Usually not first: fully autonomous decisions about money, health or eligibility, and features where a confident wrong answer reaches a customer with no review.

Each shortlisted use case gets a one page brief: the user, the trigger, the data source, the model approach, how success will be measured and the main risk. Those briefs are written so that any competent team could scope the build, including your own. For a deeper look at turning a brief into a plan, read how to scope an AI project before you build.

How Tepia approaches AI readiness assessments

The assessment is the Discovery phase of Tepia’s six phase process, pointed at AI. A US based project manager runs it, and the later phases apply once you choose a use case to build.

1. Discovery: stakeholder and frontline interviews, a review of your systems and their APIs, and sampling of real data. Deliverables are an Investigation Summary covering the five scored areas, an Interview Summary, and User Stories for each shortlisted use case.
2. Design: for the chosen use case, wireframes showing where AI output appears, how users correct it and what happens when the model is unsure.
3. Development and Testing: a proof of concept or Alpha against an evaluation set built from your records, with functional and user acceptance test plans.
4. Training: sessions that teach reviewers how to judge and correct AI output, using scenarios from your own workflow.
5. Launch: a limited rollout to one team, with data connections moved from samples to production systems.
6. Support: monitoring of accuracy and feedback, bug tracking, and feature requests as you add the next use case.

Discovery typically lasts 1 to 3 months and runs shorter when you arrive with one use case already defined. See our process for how each phase works.

Signs you are ready, and signs to fix first

You are probably ready to build if you can name the person who will own the AI feature, point to the system where its data lives, and describe how someone checks the work today. You are also in good shape if a wrong answer would be caught by a person before it reaches a customer. Tepia looks for those three signals in the first week.

Fix these first if they apply:

• The knowledge the AI would need sits in personal inboxes, unstructured shared drives, or a retiring employee’s memory.
• Your core system has no API and no reliable export, so any AI feature would be working from stale copies.
• Nobody has decided what the AI is allowed to see, especially customer records, health data or payment details.
• Leadership wants AI on the roadmap but cannot name a measurable outcome it should change.

None of these are reasons to give up. They are reasons to sequence the work, and Tepia will tell you plainly when the right first step is integration or cleanup rather than a model. When a use case passes, the next step is usually an AI proof of concept on your own records. To start an assessment, talk to Tepia.

Frequently asked questions