A 60-person analytics company rebuilds its product around a conversational AI agent and announces a sunset date for the classic dashboard. Across 52 simulated voices, 25 percent leaned favorable and 55 percent leaned opposed, reception risk came back high, and the bench returned its rarest verdict: Hold off, rework before exposing. Not because the AI move is wrong, but because the plan asks the most loyal customers to give up workflows they spent years building, with nothing firm about what carries over.
The pressure is real. Two AI-native competitors launched agent-first products this year and one of them just closed a large funding round. The board pushed for a decisive AI move at the last meeting. Growth sits around 4 percent a quarter with 19 months of runway: enough to act, not enough to act twice. Engineering estimates the rebuild at nine months, and the estimate has already slipped once. So the plan is bold on purpose: the agent becomes the default interface for every new account, 70 percent of engineering moves to the agent roadmap, and the classic dashboard gets a sunset date. The bench does not vote on the strategy. It reads the announcement the way a power user reads it, the way a support rep reads it, the way a board member reads it, and it ranks what breaks first.
Every voice in the panel, ranked from the most favorable to the most opposed. Read the two ends. The top belongs to the board's AI hawk, at plus 90, and to the customer-success voices selling to new accounts, who finally have a story against the legacy players. The bottom belongs to the power users, at minus 90: people who spent years refining custom reports their teams run every week, and who hear one thing in the announcement, a sunset date on their work. Customer support sits at minus 72 on average, bracing for the confusion wave. The two ends are not arguing about the same thing: one talks relevance, the other talks respect for what they built. That is why the bench returns Hold off rather than Adjust: rework the migration story before any real exposure.
Each dot is one voice of the panel, from pushback to support. A lukewarm average can hide a panel cut in two. Here the cloud shows it.
Long-time customers, especially power users, perceive the rebuild as a personal loss of workflows they’ve refined over years, framing it as a betrayal of trust. Meanwhile, AI-native adopters and new users see the agent as a competitive advantage, creating a fracture where the company’s identity is pulled between legacy loyalty and innovation signaling.
The engineering team’s skepticism isn’t just about timelines, it’s a crisis of morale, with senior engineers feeling their institutional knowledge is being discarded and junior engineers fearing a half-baked product. The 70% resource shift risks turning technical debt into a cultural debt, where the team’s faith in leadership erodes before the rebuild even ships.
The decision strains the company’s claimed mission of empowering e-commerce brands with reliable analytics, as power users and customer success managers argue the agent-first pivot prioritizes investor signaling over the workflows that built the business. The rebuild is read as a departure from serving users to chasing AI hype, undermining the identity of the company as a trusted partner.
Launch a ‘legacy workflow preservation’ beta with the top a large share of accounts, offering white-glove migration support and a grandfathering clause for custom reports. This signals respect for their investment while testing the agent’s ability to replicate nuanced workflows, reducing the perception of amputation.
Reframe the rebuild as a phased, modular transition where the classic dashboard’s core features (e.g., custom reports) are preserved as optional modules within the agent. This addresses the principal architect’s concern about scalability and the senior engineer’s fear of technical debt, while giving the team a tangible win to rally around.
For low-engagement or price-sensitive customers, avoid framing the agent as a replacement, instead, position it as an optional ‘AI assistant’ that augments their existing dashboards. Highlight grandfathering and emphasize that the classic interface remains available during the transition, reducing silent drift from users who don’t need AI features.
This is an illustrative simulation: a panel of plausible voices generated by an AI from sourced sociological profiles. It is not a poll and not a prediction of what real customers think. Kapari explores the range of possible reactions to help a decision, on a simulated panel, never on the real population.
You describe the decision. A panel of voices reacts. You read the range of reactions before you announce it, not after.