Case Study: How a Multi-Brand Food Franchisor Built a Real-Time Competitive Intelligence System Across 100+ Brands

The brief version: a major multi-brand food service franchisor operating seven nationally recognized concepts across thousands of US franchise locations needed competitive intelligence that could keep pace with a market moving faster than any analyst team could track. What Velocity AI built was not a dashboard. It was an intelligence infrastructure — ten categories of continuously refreshed competitive data, an AI-generated readout on every section, and a conversational agent that lets anyone on the team ask questions and get sourced answers in seconds.

This is the build, in detail.

The Problem at Scale

Running competitive intelligence across a multi-brand portfolio is structurally harder than running it for a single brand. Each concept has its own competitive set. Auntie Anne's does not compete against the same brands as the ice cream concepts. The competitive pressure on a fast-casual sandwich brand comes from different directions than pressure on a pretzel-focused QSR. One size of intelligence does not fit all seven brands.

Before this deployment, the intelligence process had three structural failures:

Cadence mismatch: Analysis happened on a cycle measured in weeks. In a market where top QSR chains are launching LTOs monthly and running value promotions on a near-weekly basis, two-week-old intelligence is historical data. A competitor can announce a value bundle, build trial, and start cannibalizing traffic before the brand team even sees the signal.

Fragmented coverage: Pricing data lived in one place. FDD filings lived somewhere else. Marketing and social signals were monitored informally if at all. No single system gave anyone a unified view. Brand teams made decisions with partial maps.

No self-serve access: When a brand manager wanted to know what competitors had done with value positioning in the last quarter, they could not simply ask. They submitted a request, waited for a report, and received information filtered through another person's framing. The intelligence was analyst-gated in both directions.

The Architecture: Ten Intelligence Categories, One Unified View

Velocity AI architected the system around ten discrete intelligence categories — each with its own data pipeline, its own structured storage schema, and its own AI-generated readout. Every category is accessible through a shared dashboard interface with a brand selector at the top, allowing users to toggle between the seven concepts in the portfolio and see the competitive intelligence relevant to each one's specific competitive set.

The AI Layer: Intelligence Readout on Every Page

The data pipeline is one half of the system. The AI layer is the other half — and it is what separates a useful intelligence tool from a data warehouse.

Every category page includes an AI Intelligence Readout. This is not a static summary. It is an AI-generated analysis — grounded in published competitor events and stored for reuse by timeframe — that synthesizes the most important signals from that category into plain-language insight. When a brand manager opens the Promotions & Value page, they do not just see a table of deals. They see a paragraph that tells them what the data means: which competitors are accelerating, which promotional subcategories are heating up, and what it implies for their own positioning.

The readout updates daily. It is scoped to the brand currently selected in the dashboard. A brand manager looking at Carvel's competitive intelligence sees a readout written specifically for Carvel's competitive set — not the same readout shown to the team running a different concept.

The GoTo Agent: Conversational Intelligence Access

At the end of the navigation sits the GoTo Agent — the conversational AI interface that operates as the dashboard's intelligence assistant.

The GoTo Agent is designed for one job: letting anyone on the team ask questions about the competitive dataset in plain language and get sourced, synthesized answers without navigating through ten category pages.

The agent's suggested entry points capture the range of use cases:

• "Summarize the latest activity" — cross-category synthesis of what has happened in the last period
• "What should I pay attention to?" — AI-prioritized signal ranking across the competitive set
• "Compare recent competitor moves" — side-by-side analysis of what specific competitors have done
• "How can I use this dashboard?" — onboarding assistance for new users

The underlying mechanism is a retrieval-augmented generation system that queries the structured intelligence database, retrieves relevant records based on the question, and synthesizes a sourced answer. The agent has access to all ten categories simultaneously, which means it can answer cross-category questions that would otherwise require a user to manually synthesize across pages.

Per-Brand Competitive Sets

One of the critical architectural decisions in this build was treating each brand's competitive set as a distinct configuration — not as a subset of a global competitor list.

The intelligence brief for Auntie Anne's tracks 14 direct competitors. The intelligence brief for a separate ice cream concept tracks 19. These are not the same 14 or 19. The competitive dynamics, the relevant data sources, the signal categories that matter most — all of these differ by brand.

The system supports this through brand-level configuration at the pipeline layer. Each brand has its own defined competitive set, its own data ingestion scope, and its own AI readout context. The brand selector in the dashboard header is not just a filter — it switches the entire intelligence context. The same Promotions & Value page looks fundamentally different depending on which brand is selected, because it is drawing from a different competitive universe.

This architecture also supports role-based access. Enterprise leadership sees a cross-portfolio view. Brand managers see their brand's intelligence layer. Both views draw from the same underlying data infrastructure — the access layer controls scope, not data freshness.

The Results in Full

What Made This Build Work

Three decisions shaped the outcome:

Intelligence design before data engineering: The ten category structure was defined before a single scraper was written. The question "what decisions does this intelligence need to support, and at what level of granularity?" came first. That sequence matters because it determines what data gets collected, how it gets structured, and what the AI readout needs to be able to say about it. Teams that start with infrastructure and figure out the questions later build systems that nobody uses.

AI synthesis as a first-class feature, not an add-on: The Intelligence Readout was part of the architecture from the beginning — not added after the data pipeline was complete. This meant the underlying data had to be structured in a way that supported AI synthesis, not just visualization. Every extracted record needed attribution (source, date, competitor, category) so the readout could generate grounded claims rather than hallucinated summaries.

A conversational agent that answers cross-category questions: The GoTo Agent was not treated as a chatbot bolted onto a dashboard. It was designed as the intelligence interface for users who do not want to navigate ten pages to get a synthesized picture. The agent needed access to all ten categories simultaneously and the ability to reason across them — which required the data architecture to be unified at the query layer even though the pipelines were separate.

What This Means for Multi-Brand Food Franchisors

The QSR and food service industry is in an intelligence arms race. LTO launches have surged. Value wars are running on near-weekly cycles. Consumer expectations for novelty are compressing the window between "trend emerging" and "table stakes." A brand team that learns about a competitor's promotional move two weeks after launch has already missed the window to respond during peak consumer attention.

The organizations winning this race are not the ones with the largest analyst teams. They are the ones with the shortest path from competitive signal to brand decision — and the infrastructure to compress that path continuously, across every brand in their portfolio, without adding headcount.

Velocity AI has built production competitive intelligence systems for food service, financial services, and multi-brand franchise operators. If your brand or strategy team is still relying on periodic reports to understand what the market is doing, we can show you what a ten-category, continuously refreshed intelligence layer looks like for your specific competitive landscape.