Shadow Mode: preview the AI, then trust it.

Pacvue, Perpetua, and Teikametrics all run trials. Fourteen days, thirty days, demo-required. Inside the trial window the seller is supposed to evaluate whether the tool deserves a credit-card commitment. In practice nobody evaluates an autonomous bidder properly in fourteen days — the data window is too short and the seller is too distracted — so the trial converts on momentum rather than evidence.

Mirox does not run a trial. We run a tier. Shadow Mode is free, forever, no card. The AI reads your real account, makes real decisions, writes real traces, but nothing moves to Amazon. You can sit in it for a week, a quarter, or until the AI proves out — at your pace, on your evidence.

This is the third pillar of Mirox and the one that took the most argument to keep.

The trust-required problem

Every autonomous PPC tool faces the same chicken-and-egg shape. The product earns its value over weeks of real bidding. The seller has to authorise real bidding before any value accrues. So the seller is asked to trust on faith — and the tool is asked to perform with no margin to recover from the first bad week.

The category answer was the demo: a sales call, a sandbox account, screenshots of someone else's results. The demo is a poor proxy because the seller's own account is the only data set that matters. The category's second answer was the trial — but a trial is the same trust-required structure, just with a timer instead of a meeting.

Shadow Mode is the structural answer. The AI runs on the seller's real account, in read-only mode, for as long as the seller wants. The trust question becomes "do you trust the decisions you have already seen this AI make, written down, on your own data" — which is a different category of question and is answerable.

A tier, not a trial

The reason this is locked as a tier and not as a 14-day or 30-day trial is brand-level, not engineering-level. The tagline "Preview the AI. Then trust it." only works if the preview window is genuinely the seller's call. A 14-day clock re-introduces the trust-required pressure we explicitly reject — at day twelve the question is "should I pay" again, not "is the AI right."

The trade-off is real. A free-forever tier costs us compute that a trial would not — the sixteen agents run against the seller's account whether or not the seller is paying. We have spent the engineering hours to make that economic.

Specifically:

• Shadow accounts auto-pause after fourteen days of inactivity. One-click resume. The tier never ends; the compute pauses while the user is not looking.
• A total seat cap on the public Shadow tier when it opens in Q3 2026 (starting at 200 active seats; raised as conversion data accrues). New signups join the waitlist when the cap is hit.
• Lower agent cadence on Shadow versus Live. Paid tiers run continuously. Shadow runs at one cycle per day. Enough for daily decision traces; less than ten percent of paid-tier compute.
• A Haiku-dominant model mix on Shadow. The trace surface stays full quality; the internal cost falls roughly four to six times per invocation.

These levers compound. Per-account daily cost on Shadow ranges from five to fifteen cents in expectation, and zero for paused accounts. The free tier is not a promotional loss-leader. It is a cost-engineered position.

What the simulation actually does

Inside Shadow Mode, Mirox executes the full bidding loop end-to-end with one substitution: the SP-API write call is replaced with a write to a parallel trace store.

Concretely, on a Shadow account:

• The full sixteen-agent stack runs against your live campaign data.
• Tactician computes a target bid for every keyword the system would have acted on.
• Strategist evaluates campaign stage and proposes stage transitions.
• Sentinel reads your days-of-cover from the SP-API feed and applies inventory throttles.
• Semantic computes asymmetric relevance scores against your product detail pages.
• Shield runs every safety gate it would run on live mode.
• Every output is written to a decision trace exactly as it would be on a paid account.
• The only thing that does not happen is the API call that changes your bid.

The result is a parallel record of what the AI would have done, hour by hour, for the entire window. The seller can read traces in the dashboard, export to CSV, and compare what the AI proposed against what their current campaigns did. The gap is the value of the tool, on their data, at their cadence — not at a salesperson's.

Why we eat the cost

The economics question is real. SaaS free-to-paid norms are 2–5 percent. Shadow Mode at scale only pays for itself if conversion is meaningfully higher than that.

We made that bet on three grounds.

Brand wedge. The category's three best-known tools all run trials. Shadow Mode is the single most-mentioned reason early prospects ask us a second question. Giving up the wedge to save the compute would cost more in customer acquisition than the compute itself.

Sales-cycle compression. A demo lasts an hour and produces no data. A 14-day Shadow run produces a trace per agent per day on the seller's own account. The decision conversation at the end is qualitatively different — the seller has already lived inside the product. The compute is buying us a higher conversion rate against the same number of qualified leads.

Honesty insurance. A free Shadow tier means we cannot oversell. If our pitch claims the AI finds €X of waste and the trace shows €X/3, the seller catches it before signing. That keeps our marketing tethered to what the product actually does, which is a brand asset for the next decade.

If conversion runs below one percent at scale, the levers we will reach for are tighter cadence and a verified-seller gate (must connect Amazon and show some minimum monthly spend), not killing "forever." The wedge is expensive to give up and cheap to keep.

Public Shadow vs founding cohort Shadow

A clarification, because the calendar is staggered.

The public Shadow tier opens in Q3 2026. Free forever. Tier not trial. Auto-paused on inactivity. The waitlist is open today at mirox.pt/waitlist.

The founding cohort uses Shadow as a fourteen-day calibration window before mandatory transition to live bidding at Day 15. That is a different product configuration, defined by the signed founding-operator agreement, and exists in part to harden the live-bidding stack before public open. If you are an EU seller spending €5K+ per month and you want to skip the waitlist, the cohort is open through summer 2026.

Public Shadow stays free, forever, trial-free, for everyone else. The cohort is not a contradiction of that promise; it is a different signed agreement for a different role.

What Shadow Mode is not

Two things are worth being explicit about.

Shadow Mode is not a forecast. The traces are decisions Mirox would have made at decision time, with the information the system had then. They are not a prediction of what the campaign would have done if those decisions had been live — that would require a counterfactual model we deliberately do not ship, because every "what would have happened" estimate quietly turns into a marketing claim.

Shadow Mode is not a coaching tool for sellers learning Amazon PPC from scratch. The traces assume the reader knows what ACoS, TACoS, and a placement modifier are. If you are still building that vocabulary, the trace will look like jargon. The product gets better the more PPC fluency the seller brings.

Where this pillar leads

Of the four pillars, Shadow Mode is the one that converts buyer skepticism into evidence. It is also the one that costs the most to operate and the one that competitors are most likely to copy first. Both of those facts argue for keeping it.

Watch the AI decide before you trust it. That is the whole pitch. Read the transparency post for what the traces actually contain, then join the waitlist at mirox.pt/waitlist.