As restaurant brands adopt more technology, the challenge is turning fragmented data into useful guidance without creating more work for franchisees. Michael Chen, co-founder and president of Pokeworks, brings the combined perspective of a restaurant operator and software engineer to that challenge. As Pokeworks expanded from a fresh, customizable poke concept into a growing franchise system, the company developed Pokeworks OS to unify information from point-of-sale systems, delivery marketplaces, guest reviews and labor tools.
Pokeworks is now building an AI-powered franchise-support capability trained with direct input from its operators and business coaches. The system is designed to identify patterns across stores and channels, accelerate performance reporting and flag time-sensitive operational issues. In this Spotlight Interview, Chen discusses how Pokeworks is applying AI while keeping human judgment at the center of decision-making, the safeguards it uses to promote reliable insights and why the value of restaurant technology should ultimately be measured by its impact on the guest experience and transaction growth.
To begin, could you share a little about your background and the path that led you to co-found Pokeworks?
I actually started out in engineering, as a software developer, and that technical foundation still shapes how I approach problems. I was always drawn to building businesses, though, and had co-founded and run a few before Pokeworks. But the path to Pokeworks itself was really about a shared passion. A group of us, honestly a bunch of friends and family, had fallen in love with poke on our annual trips to Hawaii, and we kept wishing we could find that same fresh, delicious experience back home on the mainland. In 2015 that turned into a business, and we co-founded Pokeworks to bring the poke we loved to a national audience, with a twist of new flavors adapted to local tastes. For me it has been the intersection of two threads: an operator’s instinct for building and scaling a brand, paired with an engineer’s belief that the right systems and data make a business run better.
What was the original vision for Pokeworks, and how has that vision evolved as the brand and franchise system have grown?
The original vision was simple but ambitious: take the delicious poke we had fallen for on our trips to Hawaii, still a regional specialty at the time, and bring it to the mainland as a fresh, customizable experience that guests could make their own, with new flavors adapted to local tastes. Honestly, we did not expect it to grow the way it did. Early on, a Facebook post about what we were doing went viral, and almost overnight we had people contacting us from all over the country wanting to open their own Pokeworks. That was the turning point. We realized we had to figure out how to scale and spread Pokeworks the right way, and that is what pushed the vision from “build a great concept” to “build a great system that helps every operator succeed.” A franchise brand lives or dies on how consistently and how well its operators run their stores, so more recently the vision has expanded to include the technology and data layer, Pokeworks OS, that gives our franchisees the kind of operational visibility a large corporate chain would have, without asking them to become data analysts.
Pokeworks has described its proprietary technology platform as Pokeworks OS. What operational problems was the platform initially designed to solve?
Operating data was fragmented across point-of-sale, delivery marketplaces, review platforms, and labor systems. Operators were spending their time assembling reports rather than acting on them, and some problems were only visible once the data was aggregated across stores or channels.
What prompted the decision to build an AI-powered franchise support tool rather than rely solely on traditional operating manuals, training programs and field support?
Manuals, training, and field support are essential, but they are static, and they do not surface cross-store, cross-channel patterns in real time. My own background as well as a few of our co-founders are in high tech and R&D, including years working on machine learning and agentic systems, so we had a strong sense of what these systems could actually do for operators. We needed something that could aggregate the data and help operators act on it, not just document how things should be done.
You have emphasized that the system is being trained by Pokeworks operators. How are franchisees, general managers and frontline team members contributing their knowledge and experience?
Coaches and operators correct the system directly, in conversation, and those corrections are retained as durable guidance applied to later work. For example, an operator recently corrected a detail about a dine-in order type we had gotten wrong. Both the corrected finding and the reason we missed it are now permanent and stored in memory.
What types of information and operating knowledge are being used to train the system, and how do you determine which guidance is authoritative?
Our audited data warehouse is the only source for numbers. We also hold a standing rule: when an operator contradicts a figure from their own domain, we assume we are the ones who are wrong and re-audit the method, rather than defend the query.
What are some of the first questions or tasks the AI tool will be able to help franchisees address?
Weekly performance reporting, channel and catering analysis, guest review triage, and delivery-platform error diagnostics.
Where do you see agentic AI adding the most value: answering questions, diagnosing problems, recommending actions or eventually carrying out certain workflows?
Today the clearest value is in diagnosing patterns that only appear when data is aggregated across stores and channels, and in escalating time-sensitive issues, such as a same-day food-safety flag, ahead of the normal reporting cycle.
How do you ensure that the technology supports human judgment rather than encouraging operators to rely blindly on an AI-generated answer?
Every figure carries an inline source tag and a filter legend showing exactly what was included and excluded. The system reports what the data says; what a store does about it stays with the operator.
What safeguards are being put in place to address inaccurate, outdated or conflicting guidance?
Unvalidated sources carry an explicit “unverified” label and can never override an audited figure. Data feeds are checked for freshness before anything is reported, because a stalled feed can look exactly like a real decline. And, by policy, the system will not state forward-looking per-unit revenue to a franchisee.
How do you balance brandwide consistency with the autonomy individual franchisees need to respond to local market conditions?
Metric definitions and reporting standards are brand-level and fixed. What the data says about a given store, and what that store chooses to do about it, stay local.
What does a successful rollout look like from the franchisee’s perspective, and how are you approaching training and change management?
The reports are routed through our field business coaches, so the tool augments the existing operator relationship rather than replacing it. From the franchisee’s perspective, success right now looks like broad adoption: consistent coverage, a reliable cadence, and faster response to flagged issues.
What metrics will you use to determine whether the tool is delivering value, for example, faster problem resolution, improved compliance, labor savings?
Ultimately, our business is still an in-store experience, so every insight and action the system produces has to translate into better guest experience, and better guest experience is what drives sales. That is why our two main measures of success are guest sentiment and transaction growth, not technology metrics for their own sake. On that front, we have now seen two years of positive comparable sales, outpacing our industry. On the tooling side, this is a real investment for us: we have our own in-house development team, and one of our co-founders and our Head of Finance, Wen Wei, who also happens to have a software-development background, has taken on the role of Head of AI, actively building with a small team to make sure we stay on top of a fast-growing and rapidly changing AI environment.
