Roadmap

The Grounded Method: your 10-stage roadmap to trusting AI-assisted analytics

The AI Foundation Track lays the foundation; the AI Implementation Program puts the ten stages into practice.

The ten stages: how to get there

From first experiments to AI on every project. Use it to see where your team is today and what the next step is.

PrepareStages 1–3

1Set the AI stance and measure the baseline

  • Leadership states clearly what is allowed and expected with AI, and actively supports it.
  • Before starting, measure productivity the same way before and after AI, for example with Jira story points, T-shirt sizing or estimated hours, so the gain can be shown fairly.

2Give AI your business context

  • AI should have as much business context as possible.
  • Connect it to as many company resources as possible: meeting notes, email, Confluence, Slack, Microsoft Teams, and more, with the right access controls.

3Govern metrics and documentation

  • One agreed definition per metric.
  • An owner and a review process for each new metric.
  • Documentation prioritized by what's used most.
  • The business context from stage 2 feeds directly into these definitions.
BuildStages 4–6

4Connect AI to your data tools

  • Give AI access to your warehouse, BI tools and ticketing, with a step that finds the right tables before any query is written.

5Make context machine-readable

  • Turn the agreed metric definitions, joins and column descriptions into structured files that AI and your tools can read.

6Build quality checks into the workflow

  • Teach AI the checks your team already trusts: duplicates, join errors, missing values, comparisons against trusted reports.
  • Before anyone outside the team uses it, test AI on 100–200 real business questions and reach a set target, for example more than 99% correct.
ScaleStages 7–9

7Experiment individually, share what works

  • Individuals try AI on their own work, and the useful parts become shared tools for the team.
  • Collaboration matters even more than before: regular check-ins and separate AI team meetings, where everything AI is discussed, help the whole team learn faster.

8Agree on one AI setup and shared best practices

  • Use one shared AI setup, with the same settings for everyone. AI can be customized in many ways, so agree on the best practices and apply them across the team.
  • The result: ask the same question, get the same answer, no matter who asks.
  • Automatic updates and standardized processes keep everyone in sync.
  • AI brings data analytics, data science and data engineering together under one umbrella. With shared tools and shared context, more collaboration happens automatically.

9Define the rules for AI’s behavior and output, then keep improving them as a team

  • Write down how your team answers each type of question, and make AI follow the same steps and checks:
    • “Why did X move?” starts by reconciling the number with a trusted report.
    • “Should we do Y?” needs a robustness check.
    • “Did Z work?” needs a control group.
  • Dictate how each dashboard and analysis looks and reads: the wording, colors and fonts.
  • Keep these instructions in shared version control, such as Git. Whenever an analyst finds an error, or a better way for AI to work, they update the shared version, so the whole team gets the improvement.
  • That way AI can’t jump from a correlation to a recommendation, and each miss is fixed once for everyone.
SustainStage 10

10Maintain and improve

  • Keep documentation in step with data changes, and turn every lesson into an update of the standard.
  • AI increases productivity, but it needs maintenance.
    • AI is an incredible tool, but it is generic and has to be customized heavily to your business context.
    • That's why data analysts stay essential: they guard the data context and the quality checks, and keep adding context so the AI stays up to date.

The order matters. If these stages are skipped, rushed or done out of order, AI gives confident, plausible-looking answers that are wrong. Each stage gives the next one what it needs.

If AI still is not giving the right answers

The other likely cause is the AI tool itself, which may be behind the latest developments. This happens with off-the-shelf tools. In order to test this: run the same questions and the same context through a different LLM or different AI tool. If they pass, the AI tool is the problem.

Quick self-check

Where is your team? A 10-question self-check

Answer yes or no. The first "no" is usually your next step.

  1. Is it clear what's allowed with AI, and do you measure productivity the same way before and after AI?
  2. Is your business context (tables, metrics, pitfalls, decisions) available where AI can use it?
  3. Does each metric have one agreed definition and an owner?
  4. Can AI query your data and BI tools directly?
  5. Are those definitions and your table documentation in a form AI can read?
  6. Is every AI result checked, and did AI pass a test on real questions before wider use?
  7. Does your team share AI tools, rather than each person working alone?
  8. Does everyone on the team use the same AI setup?
  9. Do different question types have their own required checks?
  10. Is someone responsible for keeping AI's context up to date?

Want to go through it together? Book a free call.