Service Data and AI
Data readiness and enterprise AI pilots
We look at one of your business processes, check what data is available and define how to measure whether artificial intelligence actually improves the work. You can sign up for the assessment alone and decide afterward whether to prepare the data, run a pilot or stop there.
Summary
What tends to go wrong
The pressure to use AI usually arrives before anyone knows where to start. Without a well-defined use case, the right data and an agreed definition of success, a pilot turns into a demo that goes over well in the meeting and never reaches daily work, or into a recurring cost nobody can justify.
Who it is for
Executives sponsoring an AI initiative, IT teams who will support whatever gets built and process owners who want to know whether AI helps with a specific task: answering questions about business metrics, finding an internal procedure or checking documents.
Deliverables
What gets delivered
- An opportunity map with the value, feasibility and risk of each use case
- A source assessment covering quality, permissions, owners and the current baseline
- A pilot plan with an initial architecture, an estimate with stated assumptions, dependencies and success criteria
- When contracted, an evaluated prototype, a report listing known errors and a recommendation on next steps
Steps
How the engagement runs
Initial conversation
You describe the task, who does it, how often, what makes it hard and what happens when it goes wrong. Before any technical work, we clarify the terms of the assessment. The conversation produces a written problem statement and, where it makes sense, a scope proposal.
Data and AI assessment
We interview the process owner and IT, review sources, quality, permissions, owners, costs and the manual baseline, and compare AI against simple data or BI improvements. You receive the opportunity matrix and, for the priority use case, an initial architecture, an estimate with stated assumptions, dependencies and success criteria.
Contracted pilot
We define sources, profiles and questions or tasks, keep development examples apart from acceptance examples and log the version, cost and time of every run. You receive a working demonstration in an agreed environment, the evaluation report, known errors and a recommendation to expand, adjust or stop. The results are yours even if you decide to stop.
Rollout and ongoing support
We define integration, support, access, auditing, data refresh, human review and a rollback procedure, with owners and recurring costs agreed. You receive the accepted version, operating documentation, training, quality indicators and the procedure to suspend use if quality drops.
Roles
Who takes care of what
Trustwise
Run the assessment, prepare the agreed data, build and evaluate the pilot, log the version, cost and time of each run and deliver a reasoned recommendation.
Your company
Name a sponsor and a business owner for the use case, grant access to sources within the agreed scope, decide who evaluates and with what tolerances, supply reference examples and make the call at the end of each stage.
Engagement
Where to start
The engagement runs in stages, and each one ends with a decision on your side. The assessment can be contracted on its own and ends with a recommendation: stop, prepare the data or move on to a pilot. The pilot, rollout and ongoing support are separate proposals, with scope, responsibilities and recurring costs defined before work begins.
Scope
What is included
-
Data and AI assessment
Interviews with the process owner and IT, a review of sources, data quality and permissions, and a comparison of AI against simpler data or BI improvements.
-
Data preparation
Organizing, cleaning and controlling access to the sources the use case needs, before any model comes into play.
-
Scoped pilot
One use case, one business owner and defined sources, with acceptance examples kept separate from those used during development.
-
Quality and cost evaluation
A comparison with the current process on quality, time and cost per accepted task, always reported with the sample size.
-
Rollout and ongoing support
Integration, support, auditing, human review and a procedure to suspend use if quality drops.
Use cases
Three places to start, and how each one is measured
Every pilot starts with one use case, one business owner and a defined set of sources, and the success criteria are set beforehand, together with the team that will use the result. The scenarios below are starting points for the conversation, not projects already delivered.
- Management metrics: finance asks which stores saw margin decline in the last closed month, and the answer comes from approved measures, with the period, filters and source shown. It starts with one model and a handful of metrics.
- Measuring the metrics case: values reconciled with the official report within each measure's tolerance; tests with ambiguous filters, an incomplete period and two access profiles; time to answer compared with the current process.
- Internal document search: an employee looks up a procedure and gets an answer that cites the current version. Retrieval pulls passages from the documents and the answer is written from them (RAG); only documents the user is allowed to see are included.
- Measuring document search: answers and sources checked by subject-matter experts, with unanswerable questions included in the test. Content from a deleted or superseded document, or from one whose access was revoked, must not appear in the answer, which also has to say when evidence is missing.
- Document triage: in an administrative workflow, the pilot extracts the fields from one document type and routes mismatches to a person for review before the final record is made.
- Measuring triage: accuracy per field, rework, time per document and cost per accepted item, using samples from different layouts and keeping human review for critical fields.
Before the pilot, your company and Trustwise record the current process, the reference questions or tasks, the sample, the tolerances, the budget, who evaluates and the rule for moving forward. The comparison with the current process looks at quality, time and cost per accepted task, including consumption, infrastructure and human review. Any unauthorized access blocks release, and every rate we report comes with the sample size.
Sample answer
A question and the expected answer
An illustration with fictitious data, to show the format a metrics pilot aims for. It is not a client result.
- Question: which stores saw margin decline in the last closed month?
- Answer: the stores with a decline, the change for each one, the period covered and the filters applied
- Source: the margin measure approved by finance and the model it was read from
- Limits: if the month has not closed yet or the user's profile cannot see one of the stores, the answer says so instead of guessing
Choosing the technology
The task comes first, then the technology
We choose the technology after understanding the task, the authorized sources and the cost of running it. SQL Server, Power BI and Microsoft Fabric are considered based on your environment, region, licensing and the skills available. The proposal spells out who is responsible for the data, connectors, hosting, models, consumption and support, and how to swap a source, a model version or a vendor and run the evaluation again.
- Microsoft SQL Server, evaluated against the client's environment and licensing
- Microsoft Power BI, evaluated against the client's environment and licensing
- Microsoft Fabric, evaluated against the client's environment and licensing
FAQ
Questions we often hear
Does our data need to be in order before we start?
No. Finding out what state the data is in is part of the assessment. The conclusion is often that the first step is to clean up a source or fix a metric in BI, not to use AI.
Will you need access to our systems and documents?
During the assessment, access is agreed with IT on a least-privilege basis, through a dedicated channel, and can be revoked at any time. In a pilot, sources and profiles are defined in the proposal. On the website form, describe only the problem: do not send documents, credentials or other people's personal data.
Can we use the Power BI we already have?
It depends on your environment. Copilot in Power BI and Fabric data agents require paid capacity and administrator configuration, which in some regions includes allowing data to be processed outside the capacity's region. The assessment checks what your environment and licenses allow before making any recommendation.
How much does a pilot cost?
It depends on the use case, the sources and the technology chosen, so the website does not list prices. The assessment provides an estimate with explicit assumptions, and pilot acceptance includes the budget and the cost per accepted task, counting consumption, infrastructure and human review.
What if the pilot does not work?
The pilot exists to answer that question with data. If the result falls short of the agreed criteria, the recommendation is to adjust or stop, and you still receive the evaluation report and the errors found.
What happens after the pilot?
Running an AI solution day to day requires support, data refresh, auditing, human review and a procedure to suspend use if quality drops. That is covered in a separate rollout and support proposal, with owners and recurring costs defined, along with how to swap a source, a model or a vendor and run the evaluation again.
What does this page mean by example, pilot and proven case?
An example is an illustration of how a use case would work, written to explain the method. It uses no client data and shows no actual result. A pilot is a contracted test with agreed scope, sample and acceptance criteria. Its result applies to that use case and that sample. A proven case is a result measured in real use, published only with the client's permission and a recorded source.
How does this service build on the others?
An AI pilot depends on data that already exists and is looked after by other services. The metrics model built in a business intelligence project is the foundation for the management-metrics use case. Data infrastructure defines where the platform runs, who can reach it and how much it consumes. Personal data governance sets which sources each purpose may use and how to revoke access.