AI Roadmap
Practical business AI, not generative snippets. The AI story for Marketing Central has two chapters: what AI is doing right now to deliver a better estimate, and what AI can do for Liberty's operations in the years after launch — on Liberty's terms, on Liberty's timeline.
The AI conversation in most proposals is about AI features — chatbots, recommendation engines, automated summaries. Those are real and we'll discuss them. But the more immediate AI story is about how AI tooling is being used in the development process itself — and what that means for the estimate you're looking at.
Phase 1: AI-Assisted Development (At Launch)
The 1,309-hour net estimate already reflects a 30% efficiency gain from AI-assisted development. This is not a marketing claim — it is a concrete reduction from a gross estimate of approximately 1,676 hours, verified against the detailed bottom-up estimate by task.
Code Generation & Review
AI tooling accelerates the initial implementation of CRUD operations, form handling, data validation, and API integration scaffolding — the repetitive structural work that doesn't require senior architectural decision-making but does require significant implementation time. Senior developers review, refine, and own the output — AI produces the draft, humans own the result.
Test Writing
Automated test generation significantly reduces the time required to achieve meaningful test coverage. Unit tests, integration tests, and API contract tests that would take hours to write manually are scaffolded in minutes, then reviewed and refined. Higher test coverage at lower time cost is a direct quality benefit to Liberty.
Documentation
API documentation, integration specifications, and inline code documentation are generated alongside implementation — not as a separate documentation phase that often gets cut or deferred. Liberty inherits a codebase that is documented as it is built.
Integration Scaffolding
Third-party API integrations (carrier rate APIs, Entra, GP) involve significant boilerplate — authentication flows, error handling, retry logic, response parsing. AI tooling generates this scaffolding reliably, reducing the time each integration takes without reducing the quality of the result.
Phase 2: AI Features for Operations (Post-Launch)
These are not features in the launch scope. They are a roadmap of capabilities that become available once the platform is live and Liberty is ready to invest in them. They are presented here because they represent meaningful business value — not because we're padding the proposal with speculative features.
Semantic Search
Replace keyword-only catalog search with semantic understanding. A franchisee searching "summer outdoor materials" surfaces the right products even if those exact words don't appear in the product names. Built on vector embeddings of product descriptions and campaign metadata.
Smart Asset Tagging
Automatically tag new products added to the catalog with appropriate campaign, product type, and attribute tags based on the product's design and metadata. Reduces the manual tagging burden on Liberty's marketing team when adding new products.
Asset Recommendations
Suggest marketing assets to franchisees based on their ordering history, their office location, the time of year, and what similar offices in their region are ordering. Surfaces relevant products that a franchisee might not have found through direct search.
Campaign Recommendations
Analyze historical order data and fund utilization patterns to recommend which campaigns a specific office should prioritize — based on what's worked for similar offices in similar markets at similar points in the season.
Reporting Insights
Surface anomalies and trends in the reporting dashboards automatically: "Office count using Zee Funds is down 15% YOY in the Southeast region" or "Three vendors have average fulfillment times 40% longer than the peer average." Insights that would require manual analysis to surface become automatic.
Receipt OCR for Claims
Automatically parse uploaded receipt images to extract vendor name, date, and total amount — pre-populating the claims form for franchisees. Reduces manual data entry and improves claims data quality. Flags discrepancies between OCR-extracted data and franchisee-entered data for reviewer attention.
Approval Assistance
Analyze historical claims approval patterns to flag claims that are likely to be approved (high-confidence, complete documentation, eligible expense category) for expedited review vs. claims that match denial patterns and should receive closer attention. Helps reviewers prioritize their queue without automating the decision.
Platform Chatbot
A conversational interface that can answer franchisee questions about their account, fund balance, order status, and how to complete common tasks — reducing support volume for Liberty's team. Grounded in Marketing Central's actual data, not a generic LLM. Surfaces the answer from the system; doesn't hallucinate fund balances.
Liberty Controls the AI Roadmap
The most important thing about the Phase 2 list above is not any individual item — it's that Liberty controls which ones get built, in what order, and when. Because the platform code lives in Liberty's environment and is owned by Liberty, there is no vendor to petition, no feature request queue to join, and no licensing negotiation required to add AI capabilities.
Add AI When the Technology Is Ready
AI capabilities are evolving rapidly. What requires a significant engineering investment today may be a commodity service in 18 months. Because Liberty owns the platform, AI features can be added when the technology is mature enough to deploy confidently — not on a vendor's product cycle, not as a upsell, and not locked behind a new pricing tier.
Build What Actually Matters
Liberty's operations team will know — after a season on the new platform — which workflows are still friction-heavy, which data would be most valuable if surfaced automatically, and where AI would actually change behavior vs. where it would be a shiny feature that nobody uses. Building AI on top of an owned platform means building what matters, not what's impressive in a demo.
Every SaaS vendor in the AI space is currently promising AI features on a roadmap they control. "We'll add AI-powered tagging in Q3" — but you don't control Q3. You don't control the pricing when it arrives. You don't control the model it uses or whether it suits your use case. When Liberty owns the platform, "we'll add receipt OCR for claims" is a decision Liberty makes and executes on Liberty's schedule, using the technology that's best at that point in time.
The RFP's AI Questions, Answered Directly
The RFP poses specific questions about how AI is used and governed. Rather than answer them in the abstract, here is a plain response to each.
| RFP Question | Answer |
|---|---|
| Are the AI models proprietary, open-source, or third-party? | Third-party foundation models, accessed through Liberty's own Azure tenant (for example, Azure OpenAI Service). There is no proprietary Bonfire model and no lock-in. Prompts and data stay within Liberty's Azure boundary and are not used to train external models. |
| How does the system handle uncertainty or misunderstood requests? | AI features are retrieval-grounded against Marketing Central's real data. When confidence is low or the answer is not in the data, the assistant says so and points the user to the right place rather than fabricating — it never invents fund balances or order facts. And a human always makes the final decision on claims and approvals: AI assists, it does not decide. |
| Does the platform support role-based access to AI capabilities? | Yes. AI features respect the same six-role RBAC and office/territory scoping as the rest of the platform. A user only ever sees AI output drawn from data they are already entitled to — AI does not create a side channel around the access model. |
| Which components are Liberty-managed vs. vendor-managed? | Liberty owns all code, data, and the Azure environment. During the engagement Bonfire builds and operates the platform; after launch, operational ownership follows the support agreement. AI features are optional and Liberty-controlled — Liberty decides if and when each one is switched on. |