Strategic Implementation of Generative AI at Scale

tomtom · Automotive · 2023 · source Agentic: yes · promotional bias: med · source-backed · extracted with prompt 0.5.6 (schema v2)

Generated from extraction.json — do not edit by hand. Every claim cites verbatim source evidence; stated means the quote directly asserts the claim, implied means it supports it indirectly.

Problem

Domain: Enterprise generative-AI adoption for location technology

TomTom needed to adapt its location technology for AI-powered products while using AI internally to improve quality, reduce costs, and scale adoption without a large new hiring or budget wave.

Constraints:

TomTom needed external GenAI products and internal productivity applications.
The strategy had to manage hallucination, confidentiality, governance, and responsible-use risks.
The company wanted broad adoption without placing resident AI specialists in every team.
Sponsored projects needed ownership transfer from the central hub to domain teams after maturity.

Architecture

TomTom used a hub-and-spoke GenAI operating model. A compact Innovation Hub set strategy, built core AI applications, guided infrastructure, and supported teams; local spokes with domain expertise identified opportunities and owned mature products. The program combined Azure OpenAI partnerships, ChatGPT location plugin work, Tommy, documentation chatbots, internal tools such as Chatty and GitHub Copilot, guidelines, training, hackathons, and portfolio reviews.

Diagram confidence: explicit (topology described by the source).

flowchart LR
  innovation_hub["Innovation Hub"] --> spoke_teams["Spoke teams"]
  spoke_teams --> sponsored_project_portfolio["Sponsored project portfolio"]
  innovation_hub --> core_ai_applications["Core AI applications"]
  core_ai_applications --> azure_openai_partnership["Azure OpenAI partnership"]
  innovation_hub --> ai_workplace_tooling["AI workplace tooling"]
  ai_workplace_tooling --> workforce_upskilling_program["Workforce upskilling program"]
  innovation_hub --> responsible_use_guidelines["Responsible use guidelines"]
  sponsored_project_portfolio --> responsible_use_guidelines
  workforce_upskilling_program --> responsible_use_guidelines

Drag to pan · scroll or use +/− to zoom · reconstructed from the source's described component topology

Components

Component	Role	Interacts with
`Innovation Hub`	Compact GenAI specialist team responsible for strategy, core applications, infrastructure guidance, and operational support.	`Spoke teams`, `Core AI applications`, `AI workplace tooling`, `Responsible use guidelines`
`Spoke teams`	Domain teams that identify opportunities, co-build proofs of concept, and own mature solutions.	`Innovation Hub`, `Sponsored project portfolio`
`Sponsored project portfolio`	Joint hub-and-spoke POCs assessed for maturity, preparedness, risk, production effects, and realized impact.	`Spoke teams`, `Responsible use guidelines`
`Core AI applications`	Location plugin for ChatGPT, Tommy in-car assistant, and developer-documentation chatbot.	`Innovation Hub`, `Azure OpenAI partnership`
`Azure OpenAI partnership`	Cloud-model partner used to leverage evolving GenAI capabilities.	`Core AI applications`
`AI workplace tooling`	GitHub Copilot, Chatty, AI code review, Microsoft 365 Copilot beta, and employee tooling support.	`Innovation Hub`, `Workforce upskilling program`
`Workforce upskilling program`	Workshops, roadshows, AMAs, newsletters, hackathons, office hours, tech talks, and communities.	`AI workplace tooling`, `Responsible use guidelines`
`Responsible use guidelines`	Data governance, hallucination awareness, knowledge grounding, training, and evolving policy guidance.	`Sponsored project portfolio`, `Workforce upskilling program`

Data flow

The Innovation Hub identifies strategy and core capabilities, partners with cloud providers, and works with local spoke teams to frame sponsored projects. A small joint team builds a proof of concept over a few weeks, iterates for months, transfers ownership to the spoke team, and tracks production effects, realized impact, risk, and reuse opportunities.

3 supporting quotes

“We deploy a hub-and-spoke model to structure and scale our innovation efforts.”article, stated

“Both would allocate a small amount of resources to jointly develop a proof-of-concept (POC) solution - typically a task for 1-2 people over a few weeks.”article, stated

“Once the product matures, the hub team provides consultative support as necessary, while the spoke team takes over full product ownership and maintenance.”article, stated

Why this architecture

TomTom chose hub-and-spoke because it needed centralized GenAI expertise and risk management while letting domain teams discover and own use cases. Buying or partnering for non-core foundation capabilities kept the company focused on location-tech applications rather than building base models.

3 supporting quotes

“Distribute our resources in small, coordinated ventures for broader coverage and risk mitigation through a hub-and-spoke model.”article, stated

“This framework promotes GenAI’s broad impact across different teams while enabling the central team to develop expertise, share knowledge, spot common trends, and identify opportunities for reuse”article, stated

“We prioritize purchasing over building non-core applications such as AI tools in Office, Enterprise Search, Salesforce, Workday, etc.”article, stated

Models used: Azure OpenAI models · ChatGPT · Microsoft 365 Copilot · GitHub Copilot

Scale

TomTom delivered multiple GenAI products in less than a year of investment.
“In less than a year’s investment in GenAI innovation, TomTom has introduced to the world a ChatGPT location plugin, an AI assistant for in-car digital cockpits (Tommy)”article, stated
GenAI development reduced timelines from quarters to weeks in TomTom's framing.
“This enables teams to reduce development time from quarters as with traditional AI, to mere weeks with GenAI”article, stated
Sponsored POCs typically use one to two people over a few weeks.
“typically a task for 1-2 people over a few weeks.”article, stated
TomTom hosted two GenAI hackathons in 2023.
“In 2023, we hosted two GenAI hackathons, nurturing innovative ideas and swift implementation”article, stated

Alternatives considered (roads not taken)

Build foundational models in-house

Rejected because: TomTom chose to focus on applications and partner with providers because foundation-model building carried upfront cost and ROI concerns.

“we explore cost-effective applications and avoid building foundational models due to upfront cost and ROI concerns.”article, stated

Put resident AI specialists in every team

Rejected because: The hub-and-spoke model let teams access central GenAI expertise and cloud capabilities without needing specialists in every domain team.

“eliminates the need for resident AI specialists in every team.”article, stated

Fund every GenAI idea immediately

Rejected because: The hub prioritizes core-business ideas, defers ideas waiting on future tech, and discourages unrelated ideas in favor of purchasing.

“We hold off on ideas that might benefit from future tech advancements and discourage ideas unrelated to TomTom’s core business, favoring purchasing instead.”article, stated

Patterns

`human-in-the-loop-gate` (organizational governance)

The hub-and-spoke operating model keeps humans in the loop through sponsored-project framing, portfolio supervision, maturity assessment, and ownership transfer.

“the hub and spoke teams work together to frame the problem, outline anticipated outcomes and impact, and design the solution.”article, stated

“The hub team supervises the sponsored projects portfolio, assessing project and tech maturity, team preparedness, and risk.”article, stated

`rag-retrieval` (grounding)

TomTom uses grounding techniques in product and process development to reduce hallucination and broader-context failures.

“we not only incorporate knowledge grounding techniques in our product and process development, but also conduct training sessions”article, stated

`safety-policy-enforcement` (responsible AI)

TomTom uses data governance, responsible-use guidelines, internal ChatGPT deployment, and training to manage hallucination and confidentiality risks.

“We prioritize data governance and offer guidelines for responsible use of GenAI. GenAI poses certain challenges like hallucination, where the output seems plausible but is factually incorrect, or lack of awareness of the broader context.”article, stated

“We’ve deployed an in-house version of ChatGPT, both as a measure to safeguard sensitive data and to standardize GenAI best practices.”article, stated

`eval-driven-development` (portfolio management)

TomTom tracks sponsored projects, production effects, realized impact, progress, success metrics, audits, and strategy reviews.

“We keep track of these sponsored projects and their production effects, as well as their realized impact. This method allows us to scale innovation with a small team of specialists, and upskill the broader teams through collaboration.”article, stated

“We plan to hold biannual audits and strategy reviews to assess GenAI initiatives, using Azure ML’s responsible AI guidelines.”article, stated

Tradeoffs

Use a hub-and-spoke GenAI operating model.

Chose Central specialist hub plus domain spoke teams over Fully centralized or fully decentralized GenAI adoption.

Rationale: The model scales expertise, spreads ownership, manages risk, and lets domain teams take over mature products.

“This framework promotes GenAI’s broad impact across different teams while enabling the central team to develop expertise, share knowledge, spot common trends”article, stated

Partner for foundation-model capability.

Chose Cloud providers like Azure OpenAI over Building foundation models internally.

Rationale: TomTom wanted to focus on application development and avoid upfront foundation-model cost and ROI risk.

“We choose to focus our innovative efforts on applications development, and closely partner with cloud providers like Azure OpenAI”article, stated

Use internal GenAI tooling for sensitive employee work.

Chose An in-house ChatGPT alternative called Chatty over Uncontrolled public tool usage.

Rationale: Chatty helped safeguard sensitive data and standardize best practices while still giving employees access to GenAI.

“‘Chatty’, an open-source, internally hosted version of ChatGPT available at https://github.com/rijnb/chatty-server/ for enhanced data privacy”article, stated

Gotchas (what bit them)

GenAI rollout can stall if every team needs its own specialists.

Root cause: Domain teams need GenAI capability, but specialist expertise is scarce and expensive to distribute everywhere.

Resolution: Use a compact central hub to provide expertise, guidance, and operational support while domain spokes retain ownership.

“This enables teams to reduce development time from quarters as with traditional AI, to mere weeks with GenAI, and eliminates the need for resident AI specialists in every team. This new pattern encourages us to reconsider business opportunities to apply AI across many tail applications.”article, stated

“Our Innovation Hub is a compact group of GenAI specialists who oversee strategic directions and boost GenAI proficiency across the organization.”article, stated

Hallucination and confidentiality risks can undermine adoption.

Root cause: GenAI output can be plausible but incorrect and may expose sensitive data if public tools are used carelessly.

Resolution: Use knowledge grounding, training sessions, responsible-use guidelines, and an in-house ChatGPT deployment.

“GenAI poses certain challenges like hallucination, where the output seems plausible but is factually incorrect, or lack of awareness of the broader context.”article, stated

“We’ve deployed an in-house version of ChatGPT, both as a measure to safeguard sensitive data and to standardize GenAI best practices.”article, stated

GenAI portfolios can drift away from the core business.

Root cause: Many ideas become attractive once model capabilities are broadly available.

Resolution: Prioritize core-business projects, reassess unfunded ideas, and purchase unrelated non-core applications.

“We hold off on ideas that might benefit from future tech advancements and discourage ideas unrelated to TomTom’s core business, favoring purchasing instead.”article, stated

Evaluation

TomTom evaluates its GenAI program through sponsored-project supervision, production-effect tracking, realized-impact tracking, audits, strategy reviews, and responsible-AI guidelines.

Metrics: Production effects and realized impact for sponsored projects. · Biannual audits and strategy reviews. · Success metrics for initiative portfolio monitoring.

“We’ll continue investing in GenAI applications for location technology, optimize our portfolio of initiatives on ROI, monitor our progress, and track success metrics.”article, stated

Outcomes

TomTom launched a ChatGPT location plugin, Tommy, developer documentation chat, and several internal GenAI applications.
The company reports delivering these without substantial extra investment or a large hiring influx.
TomTom launched workplace tooling, hackathons, training, newsletters, and communities to broaden adoption.

The program reports less than one year of GenAI investment, POCs typically built by one to two people over a few weeks, two GenAI hackathons in 2023, and no significant increase in budget or team size.

“We achieved these without substantial additional investments or a large influx of new hires. This experience provided valuable insights and helped crystalize our strategy.”article, stated

“Through the initiatives implemented in 2023, we’ve successfully delivered GenAI products, optimized workflows, and upskilled our workforce”article, stated