# Performance projection
Vibe.co · 2026 · Lead Product Designer · Project
Source: https://albancarmet.com/project/performance-forecaster
Shows what a budget increase would deliver, and applies it in one click, at four entry points.
In one line: Budget recos applied 2× more than others
Team: PM, 1 engineer and me · Duration: 1 month · AI: Ideation and prototyping

## At a glance

- Problem: Budget recommendations were generic and almost nobody applied them, while advertisers kept asking how much they should spend.
- What I did: Turned an existing projection model into a module: see what more budget buys, then apply it in one click. AI served to prototype and to compare message variants.
- Result: Budget recommendations are now applied twice as often as other types. But the A/B test showed no difference with or without the forecaster.
- Limit: Projected-versus-actual tracking isn’t in place yet, and the first reading points to an optimistic curve.

## The company

Vibe is a self-serve advertising platform for connected TV. Brands of every size, from fast-growing direct-to-consumer labels to large advertisers, buy ads on the major streaming channels, target their audiences — including Walmart shopper audiences — and measure the sales they drive, the way they would on search or social.
- Sector: Ad tech · connected TV
- Customers: Thousands of brands, from DTC to large advertisers
- My time there: Since Dec. 2023 · Lead Product Designer

## The product: The platform

Everything on Vibe revolves around the campaign, and three steps come up throughout this page.
- Upload creatives: A creative is a video ad. Advertisers upload theirs to the platform.
- Launch a campaign: A campaign sets a goal and is split into strategies: who to reach, with what budget.
- Measure: The dashboard and reports show spend, reach and the sales driven.

## Process

AI as an exploration ground, to prototype fast and multiply message variants. The decisions stayed human.
- Converge: Low interaction data, the question customer success teams heard most, and budget simulators on Google and Instagram. (without AI)
- Prototype: Manipulable prototypes instead of static screens: the curve and slider tested in minutes. (with AI)
- Word: Several ways to word and rank the gain and the recommendation, compared in context. (with AI)
- Decide: What to show or hide, how far to simplify, the tone of the alerts and the final wording. (without AI)
- Ship and measure: Built with one engineer, then an A/B test on the apply rate. (without AI)

## The problem — Budget advice had stopped working

Raising the budget was one generic recommendation among others on the dashboard, and Apply was almost never clicked.
Yet “How much should I spend?” was the most frequent question in customer follow-up meetings. Advertisers wanted a precise amount and what it delivers.

## The approach — Turn an existing model into a recommendation people trust

The projection model already existed on the data side, so we went straight into design and validated through prototypes. Handling the slider in a prototype is what shaped the four alert states: below current, toward recommended, at recommended, above.

## The trade-offs — Five decisions, and what each one cost

Where the module lives, how much the number should promise, and what waits for later.

## What shipped — One projection, four entry points

Advertisers see what more budget buys, then apply it in one click, where they already are.

## Constraints

- Predict without over-promising: The model extrapolates from each strategy’s recent performance, so the figures had to stay cautious estimates, presented as such (“≈”, “modeled on your recent performance”).
- Enough data per strategy: A projection only makes sense after enough days of delivery. Many recent or low-volume strategies couldn’t get a recommendation — hence the “You’re all caught up” empty state.
- Move past generic advice: Raising the budget had been one fixed recommendation among others, with identical wording, that nobody read anymore. It had to stand out by resting on figures specific to each strategy.

## Signals and research

No formal discovery: internal signals converged, and the projection model already existed on the data side, so we went straight into design and validated through prototypes.
- Data — Recommendations ignored: Interaction with Dashboard recommendations was low and declining, and Apply was almost never clicked. Generic content no longer triggered action.
- CSM — “How much should I spend?”: The most frequent question in follow-up meetings. Advertisers wanted a precise amount and what it delivers, not “increase your budget”.
- Benchmark — Google and Instagram already do it: Search and social platforms show budget simulators with projected results. Advertisers are used to them and expect the same reasoning on CTV.

## Decisions and trade-offs

- An integrated module rather than a separate “Guided budget” mode. Decided with: My PM and the engineer. Rejected: “Manual / Guided budget” — Radio buttons switched between the existing manual budget and a budget guided by the recommendation. Kept: An integrated forecaster — The recommendation and the curve sit right under the budget field, with no mode switch. Why: One more interaction for the same result, and a visual weight out of proportion with the campaign’s other settings. What it cost: Nothing within the shipped scope.
- The alert at the top rather than at the bottom. Decided with: My PM and the engineer. Rejected: Alert under the slider — The message came after the curve. Kept: A dynamic alert at the top — Placed before the curve, it changes with the slider: below current budget, between current and recommended, at recommended, above it. Why: A more logical reading order, less visual noise, and the emphasis stays on the slider. What it cost: Nothing. Above the recommendation, the message stays positive and quantifies the extra gain, never discouraging a higher budget.
- A tooltip that qualifies the impact rather than a bare number. Decided with: My PM and the engineer. Rejected: Number only — The tooltip showed only the value, with color codes that didn’t clearly read as positive or negative. Kept: The forecaster tooltip — Three states — positive in green, negative in red, neutral — show the gap to current budget and the tested budget, e.g. “+25.1K sessions at $625/day”. A “New $X” badge marks the tested point. Why: A bare number wasn’t telling enough. The sign, the color and the matching budget make the impact readable at a glance. What it cost: Nothing.
- Keep the Agent’s card rather than rewrite it. Decided with: My PM and the engineer. Rejected: A projection “Scale card” — The same component as the existing card, reworded to name the strategy and explain the spend-to-results curve. Designed, not shipped. Kept: The existing card — The Agent keeps its “Scale your results” card, with Approve and Reject. Why: The change was wording only, and every other agent card would have needed the same rewording — not enough substance to justify it. What it cost: Wording consistency between the Agent and the other entry points.
- Ship the projection rather than the teaser for new campaigns. Decided with: My PM and the engineer. Rejected: A teaser for new campaigns — In campaign creation, inviting advertisers to launch so they would later get projections. A Dashboard version would announce that results and recommendations come after some delivery time. Kept: Projection only — The forecaster only appears once a strategy has enough data — on the Dashboard, in the Campaigns list and in campaign creation. Why: Limited development time: we kept what had the most impact, and the teaser needed extra effort for a return judged too low. What it cost: Guidance for new campaigns: they get no projection, and nothing tells them they will. What makes up for it: On the Dashboard, a fallback module shows the 7-day spend and impressions trend when there is nothing to recommend.

## What shipped

- Performance forecaster on the Dashboard (The best scaling opportunity, front and center.): problem: The budget recommendation was buried among generic cards nobody read anymore. Solution: A dedicated module explains why, with an estimated weekly gain and a curve, then applies the budget in one click.
- “Ready to scale” in the Campaigns list (A signal in the Performance column.): problem: Nothing in the list showed which campaigns could absorb more budget. Solution: A chip next to the existing optimizations, with the estimated impact on hover and a direct path to the strategy.
- Forecaster in campaign creation (Test a budget before committing to it.): problem: Advertisers set their budget without seeing what it would deliver, or why to raise it. Solution: An alert at the top follows the slider, and the curve’s tooltips quantify the impact in real time.
- Recommendation in the Agent (Approve the recommendation in the conversation.): problem: Advertisers who work through the Agent had to leave it to act on a recommendation. Solution: The Agent surfaces the same recommendation as its existing “Scale your results” card, to approve or reject.

## Results

Budget recommendations are now the most applied kind. But the A/B test didn’t show the forecaster getting more of them applied.
(A/B test mid-August → early September 2026, then follow-up to late October)
- 2×: Budget recommendations applied within 14 days, vs other recommendation types
- No difference: Apply rate with vs without the forecaster in the A/B test (23% vs 25%, not significant)
- ~15%: Of spending strategies have an active budget recommendation
- ~½: Of the projected gain actually reached by scaled strategies (provisional)
A CSM put it plainly: many advertisers don’t scale simply because they have no more money to spend.
What worked:
- The recommendation now rests on a figure specific to each strategy, and can be applied in one click wherever the advertiser works.
What didn't work:
- The A/B test showed no measurable effect on the apply rate, nor on the share of accounts raising a budget — on a short window and a limited sample.
- A first reading suggests the projection is optimistic: scaled strategies gain, but about half of what the curve announced.
- Coverage stays narrow: only about one spending strategy in seven has enough data for a recommendation.

## Still open

Make the projection right before making it more visible. Projected-vs-actual tracking isn’t in place yet, and the first reading points to an optimistic curve. New campaigns still get no projection: the direction considered is a reliability score that strengthens as data accumulates.

## What AI did on this project

AI as an exploration ground: prototype fast and multiply message variants, to simplify a mechanism that looked complex at first.
- Interactive prototyping: Manipulable prototypes instead of static screens: the real behavior of the curve and slider could be tested in minutes.
- Message variations: Several ways to word and rank the gain, the gap to current budget and the recommendation, compared in context.
- A concrete example: Handling the slider in the prototype shaped the four alert states (below current, toward recommended, at recommended, above). We kept only the messages that make the curve understandable without reading it.
- What stayed human: What to show or hide, the acceptable level of simplification, the tone of the alerts and the final wording.
