QuickBooks
Core Accounting Automation
- Core
- Automation
- Expansion
Expanding QuickBooks auto-match beyond the bank feed — to checks, Payments Canada, and Bill Pay
Role
Timeline
Stage
Collaborators
Highlights
- QuickBooks Checks — led product discovery on end-to-end automation of 3P-created check transactions and presented MVP findings in senior leadership review within four weeks; launched with a unique deposit ID that cut manual matching time by 80%, and later lifted match accuracy with a "Recommend the right match" A/B test.
- QuickBooks Payments in Canada — led the launch of auto-match in a joint PM + engineering-manager role across six service layers, making the capability global-by-design and unlocking a $X.XM annual revenue opportunity in the Canadian market.
- QuickBooks Bill Pay — launched a "Better Together" integration to 100% of U.S. customers, eliminating seams between QBO and Bill Pay and bringing auto-match into the Bill Pay surface; projected ~$XXXK in automation revenue over the next two years.
- Ecosystem accounting automation strategy — defined and evangelized the product and tech strategy across Intuit by data-sizing the opportunity and pairing it with customer-delight–centered bets; QB Bill Pay accelerated its integration into the capability and three attach products were prioritized for the following year.
- AI trust and control — co-led cross-functional product discovery using the D4D process (13 moderated interviews with SMBs and accountants on attitudes toward automation, 12 competitive AI product teardowns, 14 low-fidelity design concept rapid experiments via unmoderated interviews, 9 in-product experiment moderated interviews) and defined Intuit's first AI trust and control principles for designing and building AI products.
Context
My first rotation had a clear charter: improve QuickBooks ecosystem attach product activation and revenue by enabling ecosystem accounting automation, tracked against a single KPI — the share of transactions that flowed feed-to-books-to-close without human intervention.
Auto-match was the ML capability that drove that KPI. It recognizes an incoming payment and reconciles it against an open invoice, so a small business owner doesn't have to sit and match line items by hand. It had been running for years on the bank feed, where the metadata is rich and the matches are clean — but every other attach product in the QuickBooks ecosystem was still mostly manual.
The bet was to extend ecosystem accounting automation across three of those attach products. QuickBooks Checks came first, with its own key result attached — end-to-end automation in service of lifting gross new attach rate and retention for QB Checks customers. Then Payments Canada and Bill Pay, as the next two ecosystem extensions. The model didn't need to get smarter; it needed to get wider — and the receiving product teams needed to absorb the capability without losing the trust contract customers already had on the bank feed.
Overview
I led three expansions over the year. Each one was a different shape of integration work.
Checks came first. The team designed a unique deposit ID printed onto the check itself, so when the payment cleared, auto-match had a high-confidence anchor instead of guessing from amount and date. Launched with a cross-functional team of 3, the change cut manual matching time on the surface by 80%.
Payments Canada was the same idea, ported across the border. We extended the unique deposit ID concept to global file formats, partnering with Intuit's money services and data exchange teams to keep the ID legible through the rails Canada actually uses. It opened a $2.1M annual revenue opportunity for the Canadian SMB segment.
Bill Pay flipped the polarity — auto-match for outgoing payments, not incoming. Same capability, different mental model: instead of "who paid me," it was "which bill did I just pay." We launched it out-of-the-box as a default value prop for 60K Bill Pay customers, opening a $760K adoption opportunity.
Approach
The KPI gave me a single through-line. Every roadmap decision was a question of how many transactions on a given rail could clear feed-to-books-to-close without a human in the loop. The expansion playbook came together around that: hold the model and the trust contract constant, flex everything downstream — the integration footprint, the customer surface, the regulatory shape.
The trust contract mattered more than I expected going in. Auto-match looks great when it's right and terrible when it's wrong, and three new surfaces meant three new ways it could read wrong to a customer. Alongside the expansion work, I led the AI trust and control discovery — 48 research touchpoints across moderated interviews, competitive teardowns, concept tests, and in-product experiments — and synthesized the findings into what became Intuit's first company-wide AI trust principles. That gave each expansion the same contract to inherit instead of three bespoke ones.
The unique deposit ID became the structural insight that made the playbook portable. It wasn't a model improvement; it was a few bytes in the right place in the payment rail. Checks printed it. Canada's file formats carried it. Bill Pay generated and matched it on the outbound side. Most of the cross-team work was making sure those bytes survived the rails and landed where the model expected them.
Challenges
Auto-match works as well as the data on the rail. Each expansion was a different version of that problem — the deterministic key the model relied on could be sparse, dropped, or shaped wrong for the capability.
Check numbers were the obvious anchor, but the data didn't always carry them through. Checks created through 3P apps didn't print the number through to the bank feed reliably, and even when they did, the field came through inconsistent across issuing banks' clearing formats. We couldn't treat the check number as a deterministic key on its own. The unique deposit ID printed onto the check became the anchor we controlled end-to-end, and probabilistic matching covered the cases where the rail still dropped data on us.
Canada didn't just have different file formats — it had a different rail, and the difference didn't surface until launch. We extended the unique deposit ID approach expecting it to port cleanly; post-launch, match rates for certain Canadian banks collapsed to near-zero while US rates stayed steady on the same model. The team's first read was a matching-logic bug. I pulled the raw bank-feed payloads side by side instead of trusting the second-hand reports — US payloads carried a remittance reference field tied to the internal invoice key (the deterministic key the model relied on); Canadian payloads had that field null or filled with generic strings like "EFT Transfer." Canadian banking rails treat remittance metadata as optional and frequently strip it in transit — not a code problem, a rail problem. Since we couldn't change the rail, the fix had to move above the model: probabilistic matching (date ±1 day, exact amount, customer name) paired with an explicit Confirm Match step in the UI rather than auto-accepting, to keep false positives from eroding the trust contract. Match rates went from near-zero to ~80%.
Bill Pay was the inverse problem — the data was there, but in the wrong shape. Auto-match had been built around a numeric ID format, what the bank feed had carried since day one. Bill Pay's transaction IDs were alphanumeric. The work was less about the model and more about coordinating across Bill Pay's transaction system, the auto-match service, and the payments layer to agree on a single unique ID format all three could read and produce — matching up transactions across services before the model ever saw them.
What I learned
To scale a platform, you have to sell the integration.
I didn't own the receiving surfaces (Checks, Canada, Bill Pay)—I owned the capability. Getting different product teams to prioritize our integration taught me that platform PMing is an exercise in internal enterprise sales. You have to prove that integrating your capability solves their OKRs (like Bill Pay adoption or Checks retention) faster than building their own roadmap.
Trust is built entirely in the edge cases.
When auto-match works perfectly, users barely notice it. When it makes a mistake, it shatters their confidence in the entire product. Designing the "Confirm Match" step for the Canadian market taught me that you don't build trust by promising a flawless AI; you build trust by giving users graceful, low-effort ways to catch and correct the AI's inevitable mistakes.
Shipping globally requires thinking architecturally.
Playing a hybrid PM/Engineering Manager role for the Canada launch forced me to look beneath the UI. When match rates dropped, the answer wasn't in customer interviews; it was in the raw bank-feed payloads. True product velocity happens when a PM can fluidly cross the boundary between user experience and system architecture to identify exactly where the data is breaking.
Feedback
It was great to see Kaci leading the effort with discovery research for checks automation. Proactively led the whole project, moderated interviews and the synthesis session to find great insights to move the team forward in a short period of time! Kaci's also great at going broad, finding great resources/materials and sharing it to the team to build onto the knowledge together. I'm super grateful to work with Kaci who brings so much positive energy and curiosity into the room.
Product designerQuickBooks partnerKudos to you for doing an amazing job with the customer interviews. It was commendable to see how quickly you learned QB and spearheaded the interview. You displayed great ownership while working on Ecosystem Accounting Automation work. You did a great job with designing the experiment plan for QB Checks and also collaborated with analytics teams to zero in on the tracking requirements. You do a great job in making the life of product dev easy by doing a great job with laying out clear requirements for tasks. Thanks a lot for offering help on multiple occasions while I was working on the QB Checks V1 work.
EngineerQuickBooks partnerKaci's active participation and valuable insights in our metrics discussions for automatch checks have been truly commendable. She consistently brings a fresh perspective and thoughtful analysis to the table, contributing to meaningful discussions that help our team make data-driven decisions. Kaci actively engages in metrics discussions, asking probing questions and providing insightful suggestions that challenge the status quo and lead to innovative solutions. I am impressed with how proactive and resourceful Kaci is when it comes to data analysis. As a PM, she has taken the initiative to dig deep into the data tables and consistently produces accurate and insightful ad hoc queries. During the power analysis exercise, she actively participated in the code logic discussions with the new data analyst on the team. Her attention to detail and analytical acumen have resulted in improved data-driven decision-making on the auto-matching checks project and have immensely helped the new data analyst's onboarding process.
Data analystQuickBooks partnerKaci has been a key asset to the team, particularly in the Automatch Checks initiative. Her exceptional ability to stick to deadlines and keep the entire team on track has been instrumental in ensuring the successful completion of the project. Kaci consistently demonstrates a high level of commitment to meeting project deadlines and has a strong sense of accountability towards project milestones. Her effective project management skills and keen attention to detail have helped to streamline the workflow and ensure that all team members are aligned and working towards project goals.
Data analystQuickBooks partnerKaci, I am so grateful to be part of the team to see you bring so much innovation and speed to our AI trust research. Your work to set up a note-taking space for in-depth interviews that seamlessly integrates into Mural for a group synthesis not only makes it easier for everyone participating, but streamlines the whole process. Analyzing qualitative data is a long, time-taking process and you've managed to simplify something so manual and arduous. I'm hoping you could evangelize this template to teams beyond ours so the whole company can take advantage of an awesome tool you've created. Thanks again and cheers!
Product designerQuickBooks partnerThank you for your hard work on the Automation research study! You are working on things that haven't been done yet and traveling in uncharted territory. This research has the potential to impact the way the small-business segment as a whole looks at positioning automation to our customers. Not to mention the fact that you conducted 10+ interviews in one week. Hats off to you and thank you for being customer obsessed.
Design managerQuickBooks stakeholderYou've been a champion of quality for our customers and been a critical part of multiple accounting automation services modernization test jams. Moving from an older system to a new one is always challenging. Thank you for making this test jam with 100s of use cases successful and helping inspire confidence in the quality we hold ourselves up to. Thank you for crossing all the t's and dotting all the i's!
Engineering managerQuickBooks stakeholderKaci — congratulations on the launch of QB Check auto-match V1! It's so awesome to see your hard work and determination in launching this capability and experience across two different back ends come to life to delight customers and accountants. This is only the beginning! Can't wait to see the data and to start auto-adding to the books to completely automate the E2E experience.
DirectorQuickBooks stakeholderThank you Kaci for your relentless efforts in getting all teams together and driving everyone toward a potential solution to enable auto-matching for Canadian customers. It was not an easy task and your courage to handle it was inspiring! Thank you! Cheers!
EngineerQuickBooks partnerWhat you shared today was no small feat and the work will be durable and foundational to getting customers Books Complete. The Product Experience is critical to any of our AI work being valid and useful to customers, and what you shared will give us the learnings to deliver that AI-enabled experience. Let's hold ourselves and others to these insights and standards and work to get them to customers quickly. Well done!
Group PMQuickBooks stakeholderHi Kaci, I would like to thank you for bringing so many cross-functional teams together for ecosystem accounting automation. There are many moving parts to the project and many teams involved. Thank you for excellent collaboration, documentation, being on top of all the questions, and patiently answering questions.
EngineerQuickBooks partnerCongrats on an amazing XR! I know you worked super hard to get QB Checks launched, push for the best analytics possible to make data-backed decisions, and tell a super compelling story. You are an incredible PM and I feel so lucky to work with you. People called out how much they loved your provocations about increasing automation over time, the data-backed decision-making, and how you laddered your work up to our overall Intuit strategy. Looking forward to seeing everything else you achieve in the team and in your future career!
ManagerQuickBooks stakeholder
Media
Role
Timeline
Stage
Collaborators
Highlights
- Launched QBO's first AI-driven advice as a plugin in the Intuit Assist conversational interface; concept validated with 12 SMBs in customer interviews, opening the door for personalized, actionable advice across the ~1M SMBs who lack ongoing access to that kind of business expertise today.
- Identified three 0→1 opportunities for embedded LLM automation inside Invoicing and Bill Pay — auditing existing flows, running a MaxDiff survey to rank candidate use cases by SMB perceived value, and grounding the results in customer interviews.
- Led strategy and launch of Intuit Assist feedback collection mechanisms for the private beta as it ramped to 10K users — a UserVoice integration plus a Qualtrics session-level in-product survey for conversations, coordinated across dependent platform teams to land MVP v1. Moved the team from no read on customer satisfaction to a 31% CSAT baseline, a data point that fed the team's roadmap and shaped a broader pivot toward incremental improvements on the Intuit Assist experience.
Context
Intuit Assist was Intuit's response to ChatGPT — a company-wide pivot that spun up a new GenAI platform and a new product surface aimed at being the trusted business expert for small and mid-sized businesses. The vision was a single conversational interface that could give SMBs critical insights, help, and automation of complex tasks: proactive, end-to-end, and rooted in their own data.
I joined during private beta as the team was ramping up to QBO admin users. The space was ambiguous and the team was operating against an outsized how-might-we: how might we help SMBs understand the so-what of QuickBooks and give them advice on how to run and grow their business more effectively within the conversational interface?
When I came in, the experience worked but was narrow. The agent could answer questions about financial data — profit and loss, sales (invoices), spend (bills) — and surface QuickBooks help content like how to send an invoice. What it couldn't do was the part SMBs actually need: telling them what the numbers meant, why it mattered, how to feel about it, and what action to take.
That gap — between data retrieval and business advice — was the brief I came in to close, partnering with experience design, engineering, and data science.
Overview
When I joined, my focus was the advice arc — moving the conversational experience from data retrieval to the so-what. Partnering with experience design, engineering, and data science, I helped launch QBO's first AI-driven advice as a plugin inside the Intuit Assist conversational interface. We validated the concept with 12 SMBs in customer interviews, opening the door for personalized, actionable advice for the ~1M SMBs who lack ongoing access to that kind of business expertise today.
Then strategic direction shifted away from advice, and I picked up two threads.
The first was 0→1 opportunity-finding for embedded LLM automation inside Invoicing and Bill Pay. I audited existing flows, ran a MaxDiff survey to rank candidate use cases by SMB perceived value, and interviewed small business owners to ground the results in real workflow pain — landing on three opportunities worth pursuing.
The second was feedback infrastructure for the Intuit Assist private beta as it ramped to 10K users. I led the strategy and launch of the collection mechanisms — a UserVoice integration plus a Qualtrics session-level in-product survey for conversations — which required aligning across dependent platform teams to land an MVP v1. The team went from no read at all on customer satisfaction to a baseline 31% CSAT — a data point that became one of the inputs to the team's roadmap and shaped a broader pivot toward incremental improvements on the Intuit Assist experience.
Approach
GenAI work didn't reward the same instincts that worked in my first rotation. In Core Accounting Automation, the brief was: build the feature for X offering, ship; build the offering for Y product, ship. The capability was mature, customer mental models were known, and the bottleneck was integration. Intuit Assist was the opposite — the tech was new, the playbook didn't exist, and the bottleneck was figuring out what the tech was actually world-class at solving.
That changed how I worked. The operating principle became learn first, launch later — and the team practice that supported it was simple: share what we hoped to learn at the start of every week, and share what we actually learned at the end. Whatever any one person picked up — a customer interview signal, a platform constraint, a model quality limit — went back to the rest of the team so the next phase didn't have to re-discover it.
The other shift was leaning earlier and harder on leadership. Platform dependencies on a brand-new GenAI stack didn't surface neatly — a build I'd scoped with three teams would suddenly involve teams six and seven we hadn't known about, each pushing their own version of a shared component. What took days or weeks for the team to untangle could be resolved in an hour-long meeting by leadership. Making timely escalation a default response to dependency knots — instead of treating it as a last resort — was a productivity multiplier on its own.
Challenges
The hardest parts of this rotation weren't technical — they were structural. The work was scoped under a strategic direction that kept moving, on top of a platform stack the company was building in parallel, in a space where nobody had a playbook yet.
Strategic direction moved underneath the work. The advice arc validated in 12-SMB interviews and pointed at a ~1M-SMB opportunity, but the path to production-quality advice was longer than the team's window. Direction shifted on, and the work I picked up after — the 0→1 opportunity finding and the feedback infrastructure — had to fit a roadmap that was still in motion. Re-anchoring the team's through-line through that shift was harder than re-scoping the work itself.
Surfacing the right 0→1 opportunities meant ranking against perceived SMB value without overweighting what the tech happened to be good at. The MaxDiff results said one thing; the interviews said something more nuanced; not every high-ranked use case had a clean LLM-shaped problem in it. The cut down from the candidate set to three pursuable opportunities was as much a discipline of saying no as a discipline of choosing.
Landing the feedback infrastructure MVP v1 required aligning two dependent platform teams on what their components needed to expose for our use case to work. Both teams were busy with their own roadmaps; we were a downstream dependency, not a primary customer. The 31% CSAT baseline that came out the other side was as much a cross-org alignment win as a measurement one.
What I learned
Learn first, launch later.
In Core Accounting Automation, the brief was to ship. In Intuit Assist, shipping took a backseat to learning what the tech was actually world-class at solving. The team ran a simple weekly ritual — what we hoped to learn at the start of the week, what we actually learned at the end — so each phase of the build inherited everything the team had picked up the week before. The point wasn't slower; it was that fewer build cycles got wasted on the wrong questions.
The new norm became having no norms at all.
I assumed the team would have the answers neatly laid out. It took me a beat to internalize that generative AI is so fundamentally new that nobody fully has the playbook. The upside of the blank canvas is the freedom to build new norms and shape the team culture from the ground up — closer in feel to a startup than to an org with established rhythms.
Timely escalation is a habit, not a last resort.
Platform dependencies on a new GenAI stack don't surface neatly. A build scoped with three teams would suddenly involve teams six and seven we hadn't known about, each pulling a shared component in a different direction. What took the team days or weeks to untangle could be resolved in an hour by leadership. Treating escalation as a default response to dependency knots — not a sign of failure — was one of the biggest mindset shifts the rotation taught me.
Customers come first, even in the hype cycle.
It's easy to get swept up by the excitement of a new technology. The job didn't change — advocate for customers, understand their needs, solve their problems. What changed was the toolset. The tech is a means to an end; what matters to the customer is what it can do for them.
Feedback
Kaci approaches every project fearlessly and autonomously and is always rolling up her sleeves and coming up with new ways to approach problems and voice her perspective, especially in ambiguous spaces, leading to an ability to effectively partner with her teammates to make an impact for our customers. This has shown up in numerous examples including a new in-product survey, shipping the third ever plugin within Intuit Assist, and coming up with new ways for the team to improve the overall Intuit Assist experience.
ManagerIntuit Assist teamKaci takes a customer-centric approach to her projects and consistently talks with and learns from customers directly to ground her thoughts and her team in customer needs, which helps ensure the team has shared context on who customers are and what they need.
ManagerIntuit Assist teamKaci was fantastic at helping wrangle the myriad deadlines and requirements to get the advice plugin into production. She worked well with the advice team and the AI PM to ensure the right technical requirements were met and that the approval docs were written and reviewed by the necessary parties. She has also been incredibly available, always helpful in pointing to the right resources, and personally helping gather analytics numbers (dashboards, etc.) for last-minute communications with leaders.
Colleagueadvice plugin partnerIf I had to describe Kaci in one word, it would be "fearless" — and I mean that in the best way possible. One of her greatest strengths, which constantly inspires me, is her fearlessness in voicing her opinion and asking questions. I have witnessed this many times, a consistent trait that I believe is critical for a product manager. It allows her to gain an in-depth understanding of the project she is working on.
ColleagueIntuit Assist teamKaci is a superstar in her strong ability to take initiative and be customer-centric. One example of this is when she led the effort to have multiple customer interviews to better understand the advice users want to receive, and worked with designers to ensure the output of the advice plugin is well phrased for consumption. Kaci's love for customers is remarkable. A couple of months back, we had our hands full with work. But Kaci, being Kaci, still managed to conduct a weekly customer study and even summarized and shared her findings with the rest of the team.
ColleagueIntuit Assist teamI appreciate Kaci's promptness when it comes to responding to my messages. I was working with her to design how Advice would appear in the chat panel, but I realized I forgot to ask for the actual content that should be displayed. So I reached out to her and got a response in minutes. This isn't just a one-time thing; she's consistently quick to reply. I know it may seem small, but it often helps me work smoothly by unblocking obstacles.
DesignerConversational UX partnerDuring the feedback integration project, Kaci had to collaborate with multiple teams, each with their own unique challenges to overcome. This was possibly her first project after joining Intuit Assist, but she proactively coordinated with different teams and successfully united all the threads to integrate the feedback widget.
ColleagueFeedback infrastructure partnerThank you for leading a session to help all the Intuit Assist PMs learn how to use the analytics authoring tool and run their own SQL queries via Databricks. You taught all of us something new that will enable the collective team to make data-driven decisions faster and ultimately design better experiences for our users. Kudos to you for learning this skill set and teaching all of us.
ManagerIntuit Assist PMCongratulations to the Intuit Assist for QuickBooks team and all of our supporting partners for launching v.1 of the new Beta release! We appreciate the hard work that this team has undertaken in the past two months. It's a huge first step towards an inspiring new direction. Take a bow and congratulate yourselves (and each other) for a job well done!
VPIntuit Assist Leadership
Media
Role
Timeline
Stage
Collaborators
Highlights
- Increased QuickBooks subscriptions by 11% — A/B tested a personalized feature ranker recommendation model on the landing page against a rules-based approach
- Drove $XM in ecosystem revenue — led a team of 4 data scientists and ML engineers to scale the QuickBooks recommendation system from a single campaign touchpoint into a multi-surface system, in partnership with product, sales, GTM, and marketing
Context
I came into this rotation chasing a gap I'd noticed in my own work. My prior rotation had been my first close look at building with production GenAI, and I realized how much product leverage lived in places I couldn't speak to with fluency: model behavior, eval discipline, signal taxonomy, and the infra stack. I could brief teams on product outcomes, but I couldn't reason about technical tradeoffs at the layer where they actually mattered.
So, I chose a rotation that would put me next to the people who could. The Personalization & Recommendations team was ML-dense by construction—data scientists and ML engineers running the production recommendation systems QuickBooks customers actually saw. I came for three things: empathy for what it takes to build and operate these systems, proximity to technical expertise, and a working understanding of how production ML decisions actually get made.
Overview
I led a team of ML engineers and data scientists shipping a feature ranker for the QuickBooks home page and a bundle recommendation model driving cross-sell and upsell across the ecosystem.
The harder problem was organizational — ML teams and product teams measure different things, ship on different cadences, and disagree about what counts as a launch. The operating rhythm we built is what I'm most proud of, not the models themselves.
Approach
I led the team of ML engineers and data scientists responsible for defining the QRS roadmap, which subsequently expanded into shipping the feature ranker and the bundle recommendation models.
QuickBooks Recommender System (Ecosystem Roadmap)
I started by helping define the broader roadmap for QRS, an engine designed to drive upsell and cross-sell across the entire ecosystem. This required aligning with teams for every vertical product (Payroll, Payments, Accounting, etc.).
This is where the operating problem shifted from algorithms to organizational incentives. Vertical product segments are inherently focused on their own local goals, even if those goals aren't the highest priority for the company globally. I quickly realized that because Marketing operates horizontally across all these segments, they held a much more objective pulse on the global health of the business. Partnering with Marketing allowed me to prioritize the QRS ML roadmap based on actual enterprise leverage rather than getting trapped in vertical silos.
Feature Ranker (Subscription Conversion)
With the ecosystem strategy taking shape, we took on the feature ranker. This was an ideal bet because it had a clean experimental setup: an ML ranker vs. the existing rules-based baseline on the homepage.
I quickly learned that most of the product leverage lived in the signal taxonomy I defined upstream, not in the algorithm. I started with 1P clickstream data—which was easy to acquire and already flowing through Intuit's data platform—and manually triaged which events were high-signal versus noise. Once that baseline was stable, I layered in 3P data (like ZoomInfo for industry and team size) to expand our signal coverage.
Payroll & Upgrade Bundles
Later in the rotation, we expanded our scope again, building targeting models specifically for QuickBooks Payroll and upgrade bundles to drive deeper ecosystem penetration.
Challenges
The bundle recommendation model (launched for Payroll and upgrades) came in at $XXXK against an expected ~$XM. It was a real miss, and the launch I learned the most from. We had chosen a highly sophisticated ML approach for what was a first-of-its-kind targeting model at Intuit. That complexity cost us twice: it pushed the launch timeline back, and it made the resulting customer segments harder for our marketing stakeholders to actually interpret and act on. A simpler propensity model would have shipped faster, been more interpretable, and given us a baseline to iterate from. Because sizing carries such wide error bars for any novel model, the right move is always to ship the simplest version first, benchmark reality, and then add sophistication.
Across all these launches, the harder overarching problem was getting ML and Product onto a shared definition of "shipped." ML evaluates offline metrics on long cadences; Product evaluates downstream business metrics on weekly campaign cycles; Marketing and Sales carry entirely different success criteria. Aligning everyone—without forcing the ML team into a product cadence that didn't fit, or vice versa—was a massive operational challenge, but building that shared rhythm is what made the QRS multi-surface expansion possible at all.
What I learned
Signals beat algorithms once the model works.
Most of the lift on the home page ranker came from upstream—a clean 1P clickstream cut, followed by ZoomInfo for industry and team size. The algorithm matters far less than which inputs make it in. Model quality on a recommendation system is ultimately bottlenecked by the signals you let it see.
For a first-of-its-kind model, ship the simplest version first.
The complexity of our payroll and upgrade bundle recommender cost us time and usability for our stakeholders. Because sizing carries wide error bars on a novel model, the right move is to ship the simplest possible version to benchmark reality, then add algorithmic sophistication only once you know which dimensions drive impact.
Marketing is the horizontal truth-teller in a vertically siloed org.
While building the roadmap for QRS, I had to navigate priorities across Payroll, Payments, and Accounting. Vertical product teams are incentivized to hit local metrics, which can make them myopic. Because Marketing operates as a horizontal layer, they have the clearest pulse on what actually moves the needle for the overall business. To prioritize an ecosystem-wide ML roadmap correctly, you have to align with the horizontal truth-tellers.
The model is only one of six levers (and rarely the highest-leverage one).
Problem framing, data/signals, targeting, UX/delivery, infra, and the model itself all move the exact same business metric. Spending all your time tweaking the algorithm only pays off if the other five levers are already tuned. Most of the impact I drove in this rotation came from optimizing the levers outside the algorithm.
Feedback
It would be an understatement to say that Kaci came in to rescue our joint efforts and collaboration with the product team on forming an AI-native future for the product. This was the first time the team had tried to establish a solid and deep collaboration with the product team, and with Kaci joining the team, she helped completely transform the way we worked together, pushing the collaboration to another level. Kaci hit the ground running immediately after joining the team, quickly immersing herself in the complicated landscape of the team's work. She wasted no time in forming a deep understanding of the customer problem and developing a solid grasp of the technical stack. This enabled her to start making a meaningful impact, even before anyone expected her to. Kaci has shown exceptional competence as a product manager, introducing the team to a complete new suite of best practices in terms of project/dependency/stakeholder management. She helped formalize the process of cross-functional collaboration, which was literally non-existent before. The team's "PM stack" received a complete transformation after her arrival. Thanks to Kaci's highly meticulous and methodical approach, we have formed a close-knit relationship with the product team, embarking on the journey of forming a shared AI-native product roadmap. Kaci embodies a customer-first mindset, always starting by closely inspecting assumptions and anchoring on customer problems. I was amazed by her data-driven approach, to the extent that if she couldn't find data to support a story, she would go the extra mile to uncover it herself. I'm particularly grateful for how Kaci has consistently translated vague requests from the product team into executable plans and a clear future roadmap, helping to orient and ground the entire team. I also appreciate how she has encouraged the AI team to take more ownership, raise our voices in contributing to the shared roadmap, and take the driver seat to push the boundary of building an AI-native future, even during highly turbulent and uncertain times when we observed priorities could shift overnight and plans had to adapt quickly.
Data scientistPersonalization teamI have had the opportunity to witness Kaci's continuous personal growth and increasing contributions to the team since she joined us. One aspect that truly stands out about Kaci is her deep customer obsession. Although she functions as the AI PM and focuses on shaping and building AI products, she genuinely cares about the experiences we create for our customers. I have observed her proactively putting herself in the customers' shoes and collaborating closely with product managers to deliver the best possible experience through AI. For instance, she successfully influenced the team to effectively utilize the feature ranker model for one of our key surfaces while consistently identifying and cultivating AI opportunities to solve critical customer problems. In addition to her product acumen and deep customer empathy, Kaci possesses a genuine interest in technology. She has consistently helped the team discover opportunities for innovation and the development of outstanding AI/ML solutions. For example, she showed strong interest and support for the customer understanding initiative, carving out space for the team to explore areas we hadn't previously touched, such as model fine-tuning and representation learning, and connecting those efforts to actual business impact. Innovation within this team would not have been as prosperous without Kaci as the strong advocate. I am excited to see Kaci's ever-expanding scope and the even greater impact she will make. I believe she is ready for the challenge of leveraging her tech enthusiasm and customer obsession while further developing her architectural thinking. I am confident she will excel if given the opportunity to take ownership of a larger part of the system and build something from the ground up.
Data scientistPersonalization teamKaci, thank you for putting together the listening tour reflections and sharing feedback for the team. It was great for us to hear your goals so we can work together to support you. Also, with the ups and downs of the team and stakeholders, you have really leaned in to review their strategies, push for clarity, and create clear plans to get alignment. Not an easy task, and we're grateful to partner with you on it. Thanks for bringing your perspective and energy. Looking forward to working with you more! And on the recent shareout: thank you for sharing out to cross-functional leadership. It was an engaging discussion and a good way to bring stakeholders into how AI solutions are built.
Engineering managerRecommendation systemsKaci, thank you so much for all your hard work on the signup project. You proactively audited our jobs-to-be-done taxonomies for maximum impact and consistency, and did so in such a thoughtful way! Your input and expertise helped us answer big questions and connect the dots so we can make the most informed design decisions for our future customers. You're very appreciated!
Content designerSignup projectThis is another huge milestone towards driving the ecosystem revenue for QB. Thank you for your dedicated effort to constantly looking for ways to improve and iterating our solutions to deliver the best customer benefit. I would like to also highlight the collaboration among teams to utilize the AI capability, new data source, and more to expand the recommendation coverage to our customers. Great work and congrats on the launch of the recommendation model!
Engineering managerRecommendation modelCongratulations to the team on the successful launch of the signup work! This achievement is a testament to our collaboration, customer obsession, and commitment to excellence. From the initial concept to the final execution, this launch was a true team effort. Each individual brought their unique talents and perspectives to the table, and together we created a solution that truly meets the needs of our customers. I appreciate your hard work and dedication. It is through our collective efforts that we have demonstrated the true meaning of "Stronger Together, Customer Obsessed." Look forward to seeing the impact to the QBO ecosystem and continuing our collaboration.
Engineering managerSignup projectThank you for going above and beyond during our innovation sprint and collaborating on a generative rec-system proof of concept! This team accomplished an impressive feat exploring multiple data sources, novel feature engineering techniques, and building multiple variations of generative and traditional solutions for cold start recommendations. Thank you for pushing the envelope with bold innovative ideas and out-of-the-box thinking, and for showcasing true care and passion in your contributions. It would not have been possible without you!
BuilderRec-system PoCWe want to personally thank you for your outstanding contributions and dedication that led to the successful major launch of our recommendation system enhancements! This incredible achievement includes increasing the number of recommendation slots in our platform, which will help increase impressions for the recommendation system, and the simultaneous rollout of a major release featuring expanded recommended actions.
VPMarketingHuge thanks for your outstanding work on the bundle recommendation! Your meticulous effort in understanding the end-to-end integration of the recommendation engine into our GTM campaign launch was absolutely key. You expertly navigated every complexity, perfectly aligning all teams to launch a new monetization stream. Your detailed approach and seamless execution are truly appreciated!
Principal PMBundles GTMThank you for handling all the pressure from external partners and managing to keep us on the right track. Congrats on the payroll bundle rollout!
Data scientistPayroll bundleWe wanted to thank you for the amazing job you have done to launch the targeting model MVP. As the business expands its bundles monetization strategy, the learnings from this launch will be critical to driving the direction we take to deliver the right offers to the right customers. Your hard work and thought partnership have helped us get a working model out the door so we can learn and iterate quickly. We are so proud to be a part of this team and work alongside such talented and hardworking individuals. Thank you for everything you do!
LeadershipBundles
Media
Role
Timeline
Collaborators
Highlights
- Owned end-to-end recruiting strategy for RPM full-time and intern pipelines — generated 2K+ applications via targeted sourcing and campus partnerships, and closed 10 of 10 extended offers by designing a high-touch candidate experience
Overview
Alongside the product work, I led the recruiting strategy for the QuickBooks Rotational PM program — both full-time and intern pipelines — designing the sourcing approach, campus partnerships, and the candidate experience.
Feedback
Team, thank you for helping plan and organize the RPM Onboarding for the 2023 March RPM cohort! You've helped assess which sessions were priority, organized social events to help build community, provided resources and welcome materials, and coached them through their RPM Project. The new RPMs had a lot of FUN and feel really grateful for the experience. Thank you for helping create an awesome first impression of RPM and for supporting your community!
SVPSmall Business GroupThank you to each of you for planning, budgeting, and executing an awesome RPM social or community service event or celebration! This program is so special because YOU put in the time to create amazing memories together and celebrate each other's milestones. The RPM Community dearly appreciates your time and ideas and commitment to helping us bond and support each other. THANK YOU!
SVPSmall Business GroupThank you for planning and executing an incredible onboarding experience for the new RPM cohort. Everything from messaging months before they start, to planning the sequence of training, to putting together an awesome hands-on project, to finding and purchasing and setting up amazing swag, to planning community events and socials, and so much more! This is the first impression the new RPM cohort has of Intuit — and what a way to start! Thank you for your time, your thoughtfulness, and your continued participation in and contribution to growing PM talent!
Program LeadRPMThank you for planning and leading a training session last week for our new cohort of RPMs. We very much appreciate your ongoing investment in growing PM talent, and working with the RPM program to ensure we are building strong PM competencies, skills, and getting them ready for their upcoming roles. Thank you!
Program LeadRPMThank you Kaci for knocking our RPM Offer Day out-of-the-park: from providing well-thought-out sell-day programming for our prospective RPMs and PM interns, to gathering a superstar team of ex-RPMs/managers and current RPM panelists to speak with them! Your energy and passion for delivering the best candidate experience are infectious, and it shows with most of our candidates having already accepted their offers! These incoming classes of PM interns and RPMs are the future of the program — so THANK YOU for giving them a glimpse of what it is like to be a part of Intuit's RPM community!
PeerRPMThank you SO MUCH for being a lead for this year's recruitment cycle. We could not have done this without you! This work is integral to building a strong PM community — by hiring top talent for Intern and RPM roles, we ensure the success of this program far into the future. Go team!
Program LeadRPMKaci, thank you for leading the RPM birthday celebrations for our community and proactively organizing and recognizing those small but big milestones in a fun way. We appreciate your leadership and joy for the special moments!
Program LeadRPMKaci, thank you for leading and organizing the PM Intern community service event on Friday at Alemany Farm. We appreciate your initiative in setting this event up, and encouraging the PM Interns to not only give back to our community and deeply understand an important Intuit value, but also build bonds with each other.
Program LeadRPMThank you for crafting and leading a training session for our newest cohort of Rotational Product Managers this week. Your investment in coaching and training incoming talent, and building the fundamental PM craft skills and professional skills for our PM community, is incredibly appreciated and valued!
Program LeadRPMTeam, big thank you for your hard work and dedication that you put into reviewing countless resumes for the RPM and RPM Internship positions. Your diligence and attention to detail were truly invaluable in helping us select the best candidates to move forward to the interview stage.
Program LeadRPMKaci and Tim, huge thank you for thinking of and coding an RPM resume review scoring system to weigh and account for various degrees of strictness across resume reviewers. This is an impactful new addition to our process to ensure we are fair and that we lower the bias between reviewers across hundreds of resumes for the RPM program. Incredible innovation, execution, and initiative!
Program LeadRPMKaci, you were nominated by an RPM for going above and beyond in spontaneously hopping on a call to help add emails for 100s of candidates to the resume review rubric sheets and ensure the sheets were ready in time for the resume review. Thank you for stepping up!
Program LeadRPMThank you for taking the time to interview candidates for the RPM and RPM Intern roles over the last 2 weeks. Your participation in the hiring process was critical to finding the right additions to our program, and we appreciate your thoughtful contributions and your time investment in hiring amazing PM talent!
Program LeadRPMTim and Kaci, huge THANK YOU to you both for your exceptional leadership during the RPM Recruiting season. Your meticulous organization of the committees was instrumental in our success, and your willingness to step in and provide assistance at every stage of the recruiting funnel was greatly appreciated (including generating the z-score, assessing the interview case studies, reviewing the resume rubric, and stepping in as last-minute interviewers). Thank you also for leading the retro session and capturing the feedback to improve future seasons.
Program LeadRPMKaci, thank you for leading the RPM Training on how to write an effective year-end assessment, how to ask for feedback, and how to share feedback. You covered not just best practices, but also shared rationale and personal examples. Really appreciate your investment in coaching RPMs!
Program LeadRPMSarah, Kaci, and Anjali — THANK YOU for your time and ideas in helping rethink the RPM case approach for our next recruiting cycle, so we can clearly assess for technical and AI proficiency, critical thinking, and adaptability. We greatly appreciate you thinking about the tools, the approach, the case itself, and the questions to ask — this will make our process so much more insightful and smooth!
Program LeadRPMKaci and Alyssa, thank you for stepping up to lead the Intuit RPM Info Session at the virtual Grace Hopper event! You did an amazing job in preparing the content and sharing your advice with insight and energy. Thank you for representing Intuit and our RPM program so effectively!
Program LeadRPM