Know what to fix before you build.

3 working days · Fixed price agreed before booking

Work directly with Tal Shany to understand one reporting, data or AI problem before committing to a build. You leave with a decision brief: what to fix, what to test and what it would take to implement.

Start with a problem you recognise

• Reports disagree: trace one agreed metric from a sample source extract to a report and document the mismatches found.
• A manual workflow consumes the week: map the hand-offs, exceptions and data needed to automate it.
• You want an internal AI assistant: assess one use case against data access, freshness, permissions and evaluation needs.

What you receive

1. A current-state map of the agreed workflow and its data sources.
2. A prioritised issue list with evidence, assumptions and unresolved questions.
3. A proposed approach, including access controls, governance and alternatives to AI.
4. A scoped prototype brief with acceptance criteria and an implementation estimate, including dependencies and ongoing cost drivers.

How the three days work

Day 1: agree the question and review the supplied architecture, sample data and workflow. Day 2: inspect the data and test the main assumptions. Day 3: walk through findings and hand over the decision brief.

A small, explicit scope

One workflow, one business owner and an agreed set of sources. The delivery window starts after the scope, access and sample data are ready. Read-only access or approved extracts are preferred. Missing access or unexpected complexity is discussed before changing the scope.

This review does not include a production deployment, a full estate audit, penetration testing or a guarantee that every data issue will be found. Any follow-on build is a separate decision.

Example deliverable: a report reconciliation review

Illustrative example, not a client result. Two monthly revenue reports show different totals. The review traces the agreed metric through date cut-offs, refunds, duplicates and joins. The handover records each mismatch found, the supporting evidence and the proposed correction or further investigation.

What happens afterwards?

You can use the plan internally, stop with the findings, or commission a scoped prototype sprint. AgentData or other tooling is considered only where it suits the problem; buying a platform is not a prerequisite.

Discuss one problem See Tal’s previous data projects