Amazon PPC: agency vs software vs AI‑managed.
The true cost comparison.

If you spend under roughly $5,000/month on Amazon ads, use software and your own hours. At that scale, any managed fee eats most of what the ads earn, and the reps you get from managing your own account are worth more than the hours cost. Sellerboard for profit truth is $19–79/mo; a mid-tier PPC tool is $100–500/mo.[1]

If you're an eight-figure brand with 1P complexity, multiple retail channels, heavy creative needs, or a compliance war to fight — hire a real agency. A good one is genuinely worth a $2,000–5,000/mo retainer (quotes we've seen for large books run $9,000–11,000/mo[2]), because what you're buying isn't bid management. It's a bench: catalog specialists, DSP traders, creative, someone to out-persist Seller Support at 2am.

If you're in the middle — roughly $500K to $10M/year on Amazon, PPC-centric, no time to do it yourself — that's the gap the third category exists for. AI-managed means an AI agent does the daily execution (bids, negatives, budgets, structure) and human operators supervise it. Dr. PPC, our product, is $300/mo + 3% of ad spend, capped, month-to-month.[3] The nearest analog, aiHello's Managed tier, is $1,500/mo.[4]

That's the conclusion. Everything below is the evidence: real 2026 price ranges for all three categories, an interactive calculator that includes the cost everyone omits (your hours), a decision framework that names the cases where we're the wrong choice, and the autonomy ladder — a plain-language standard for decoding what "AI-powered" actually means on any vendor's page, including this one.

Search "amazon ppc agency vs software" and look at who holds page one: a PPC agency, a software suite, and a software tool's podcast.[5] Each of them is one side of the fight they're refereeing. The agency's page concludes you need an agency. The software company's page concludes agencies are "$3,000 retainers for a VA moving bids." Neither is lying, exactly — they're each describing the worst version of the other.

There's a structural reason the fight is this loud: the clicks are absurdly expensive. In 2026 keyword data, "amazon ad agency" costs about $141 per click in Google Ads, and "best amazon ads agency" runs up to $398 per click at top-of-page.[6] When a single click costs more than a month of software, nobody funding those clicks can afford to tell you the other side is sometimes right.

We should state our own conflict plainly: Dr. PPC is an AI-managed service, the third category in this comparison. We will make the case for it. But this page follows three rules the category's marketing usually doesn't:

• Every price is real and dated. Vendor list prices below were verified against public sources on 2026-08-19. Prices change; re-verify before buying.
• Each category gets its genuine best case, not a strawman. There are sellers reading this who should not hire us, and we say so, twice, with numbers.
• Claims about our own results come from real accounts, anonymized, with the caveat that every account is different. Nothing here is a guarantee.

One more piece of context for why this question suddenly matters. AI running ad accounts stopped being hypothetical: analyst Brian Wieser estimates AI-driven ad buying at roughly 8% of US ad revenue in 2025, heading toward ~26% (~$142B) by 2030.[7] Amazon shipped its own approval-gated Ads Agent in late 2025 and an Ads MCP interface in February 2026.[8] The category is real. What's mostly missing — and what this page tries to be — is anyone independent explaining honestly where it fits next to the two options that already existed.

This is the biggest category by seller count, and the most mature. You subscribe to a tool, connect your Amazon Ads account, and manage campaigns yourself with better data and some automation. The range of what "automation" means here is enormous — from keyword research helpers all the way to goal-based systems that change bids without asking you — and we'll come back to that range in the autonomy ladder, because it's where most buyers get surprised.

Here are real 2026 list prices across the spectrum:

Notice the shape of that table: the "software" category quietly spans $19/mo to well past $1,000/mo. The upper half (Perpetua, Quartile, Pacvue-class platforms) costs agency-adjacent money while still, contractually, leaving the responsibility with you. That upper half is where "software vs agency" comparisons get muddy — you're paying $700–900+ a month and also supplying the strategy and the supervision.

The honest case for DIY software

Nobody will ever care about your account like you do. A founder running their own PPC learns their search-term reports, their seasonality, their true converting keywords — knowledge that stays valuable forever, whoever manages the account later. Software costs are transparent and fixed. You can start and stop without a contract. And below a certain spend, everything else is mathematically silly: if your ads spend $2,000/month and net you $1,500 of contribution profit, a $2,500 retainer is a machine for converting profit into invoices.

The cost nobody prints

Every software pricing page shows you the subscription and hides the hours. Managed well, an active Amazon account needs search-term reviews, negative harvesting, bid maintenance, budget rebalancing, campaign restructures, and event prep (Prime Day planning alone can be a week of work). Call it 4–10 hours a week for a mid-size account, done properly — our estimate from running accounts, and you can change it in the calculator below if yours differs. At $50/hour of founder time, six hours a week is ~$15,600/year — more than triple the software subscription it sits on top of. At what your time is actually worth while you're scaling a brand, it's worse.

The second hidden cost is what unsupervised automation does when it's wrong. We've watched a rules-based tool keep raising bids while ACOS climbed, because one rule in a rule sheet was mis-set — automation faithfully executing a bug.[9] And in one account we took over, a well-known automation suite had made roughly 1,800 changes in a single day — with no change log a human could read, and duplicate keyword lists sprayed across ten campaigns.[9] The tool wasn't evil. It was unattended. That difference is this entire article.

The traditional model: a monthly retainer (typically $1,500–5,000 for small-to-mid brands, with quotes for eight-figure books commonly landing at $9,000–11,000/mo[2]), or a percentage of ad spend (commonly 10–20%), or both. In exchange, humans manage your account.

Here's something the software companies' "fire your agency" ads won't tell you: we run an agency. Dr. PPC was built inside a group whose agency teams manage $500M+ in Amazon revenue for 70+ brands. So when we describe what a good agency is worth, it's not a concession — it's the other half of our own business, and we know exactly where it earns its money:

• Multi-channel and full-catalog problems. DSP and streaming TV, Walmart, 1P/Vendor negotiations, international expansion, listing compliance wars, catalog surgery. None of that is "PPC," and no PPC tool will do it. When Amazon's systems wrongly suppress your best-seller three days before Prime Day, a dashboard does not call Amazon. People do — repeatedly, for weeks if that's what it takes.
• Judgment under ambiguity. Whether to defend a Best Seller badge with margin, when a rank push is worth running TACOS hot for six weeks, whether a refund spike is a product problem or deal-shopper noise. These are business decisions wearing a PPC costume.
• A bench, not a person. A real agency gives you a team with pattern-recognition across dozens of accounts. When something weird happens in your account, someone there has usually seen it before.

The honest case against

The structural problems are just as real, and they're the reason this article exists. The retainer is due whether or not the work happened. Fixed monthly pricing means account attention is rationed — a human portfolio manager covering several accounts gives yours a deep pass weekly at best, and the interval between touches is where money leaks. Pricing is almost never published, which tells you it's negotiated against your ceiling rather than their cost. Contracts commonly run 6–12 months. And the person who impressed you in the sales call is frequently not the person moving your bids in month four.

None of this makes agencies bad. It makes them expensive precision instruments — the right tool for problems complex enough to need a human bench, and a poor deal when what you actually need is disciplined daily execution of known plays. Most brands under ~$10M on Amazon need the second thing far more often than the first.

Definition first, because the label is already being abused: AI-managed means an AI agent performs the daily execution in your account — bids, negatives, budgets, structure, search-term harvesting — and accountable humans supervise its work. It is not a dashboard with an AI logo (that's software), and it is not a black box you can't question (that's autopilot). Both halves are load-bearing. Remove the agent and you have an agency with extra steps; remove the supervision and you have the 1,800-changes-a-day problem with better branding.

Why does this category exist now? Because the unit economics of expert attention changed. The expensive thing about great PPC management was never the knowledge — it's codified in a few hundred plays — it was the hours. An agent doesn't ration hours. It can pull every campaign, every search term, reconcile the full attribution window, and act on your account every single day, for less than a human costs to look at it once a week. The human's job shifts from doing the work to auditing it — one operator supervising agent-run accounts instead of hand-managing five.

What that costs, across the category as it exists in 2026:

A fair question: if the agent is so good, why charge a percentage at all? Because the percentage is the alignment. A flat fee is indifferent to whether your spend is productive; a small percentage of spend means the service grows only when you confidently spend more — and the cap means scaling your business doesn't quietly scale the invoice. Three percent was chosen to sit far below what the labor it replaces would cost at any spend level a mid-market brand actually runs.

And the skeptical question you should ask: "isn't this just autopilot software with a services markup?" The test is observability. Our agent writes down every change it makes and the reason it made it, an operator reads that log — not the dashboard, the log — and anything the data got wrong gets overridden by a person whose name is on the account. If an AI-managed vendor can't show you the change log and can't name the human who reviewed it this week, price it as software, because that's what it is.

One more honesty note, because this category's marketing is drowning in the word "autonomous": Amazon's own Ads Agent — built by the platform itself, with more account data than any third party will ever have — still gates every change on human approval.[8] When the house that owns the auction keeps a human in the loop, be suspicious of any startup telling you it went fully hands-off. We keep humans in the loop on purpose, and the ladder below is exactly where.

Strip away the dashboards and the jargon, and competent Amazon PPC management is a loop that runs against your account on some cadence. Every model in this comparison — you with software, an agency team, an AI agent — runs the same loop. What differs is how often it runs, how much of the account it covers per pass, and who checks the output.

• Pull the data. Every campaign, ad group, keyword and search term, out of the Amazon Ads API or the console exports. DIY sellers usually pull weekly; agencies weekly per account; an agent pulls daily because pulling is free for it.
• Reconcile before acting. Amazon attributes conversions for days after the click, so yesterday's search-term report understates reality. Acting on raw recent data is the most common self-inflicted wound in DIY accounts — bids get cut on keywords that were actually converting, the conversions just hadn't posted yet.
• Harvest and negate. Move converting search terms into exact-match control; add negatives where clicks burn with no orders. This is the compounding work — skipped weeks never come back.
• Move bids toward profit, not ACOS. A 40% ACOS keyword can be your best investment (rank-building on a high-LTV product) or your worst (a low-margin SKU bleeding out). The decision needs contribution margin, which needs your COGS — one reason "set target ACOS and walk away" black boxes disappoint.
• Rebalance budgets. Money trapped in capped, unprofitable campaigns is invisible in most dashboards and obvious in a daily pass.
• Watch for the platform's surprises. Suppressed listings with live spend pointed at them, lost Buy Boxes, silently swapped images, deal-driven CPC spikes. These don't wait for your Tuesday review — the $588-a-week restoration in our receipts section was found this way.

Once you see the job as a loop, the three categories stop being ideologies and become staffing decisions for the same loop. DIY staffs it with your evenings. An agency staffs it with a shared human who runs it weekly. AI-managed staffs it with an agent that runs it daily and a human who audits the agent. Price each accordingly — which is what the calculator below does.

Three things fall out of this arithmetic at almost any slider position. First, DIY is rarely as cheap as it looks — at a defensible hourly value, the labor dwarfs the subscription. Second, the agency premium buys the most when your problems are least PPC-shaped — if all the retainer is doing is bid management, you're paying bench prices for one player. Third, the gap between models widens with spend, which is exactly why we publish a cap: past the cap, the 3% stops growing no matter how large the account gets.

A note on mixing: these aren't exclusive. Plenty of brands run DIY software on a stable core catalog and bring in help for the hard part; plenty of agency clients graduate specific workloads to AI-managed as the plays become repeatable. The categories are honest defaults, not tribes. The only genuinely wrong answer is paying agency prices for software-shaped work, or spending founder hours on work an agent does better at 3% of spend.

• 1. "What is your price, in writing, before we negotiate?" Published pricing means the price was set against the provider's costs. Unpublished pricing means it will be set against your ceiling. This one question sorts the market faster than any review site.
• 2. "Show me every change made in a client account last week, with reasons." Software: does the tool keep an intelligible change log? Agency: does the team log work, or will you get a slide deck of charts instead? AI-managed: this is the whole test — no readable log, no deal.
• 3. "Who exactly touches my account, and how many accounts do they carry?" The industry's open secret is load: a portfolio manager carrying many accounts gives yours minutes a day at best. Whatever the answer, it should be a number, not an adjective.
• 4. "What happens in week one?" The right answer in every category starts with diagnosis — structure, waste, tracking, COGS — not with activity. Anyone promising immediate scale before an audit is planning to spend your money to learn your account.
• 5. "What will you refuse to automate or won't do?" An honest provider has a boundary list (ours is the amber column above). A provider with no boundaries hasn't found theirs yet — you'd rather not be the account where they do.
• 6. "How do you measure incrementality, not just attributed ROAS?" Any answer that doesn't include some form of holdout, matched control, or AMC reconciliation means their number and reality will drift apart precisely when spend scales.
• 7. "How do I leave?" Month-to-month with your account, campaigns, and history intact is the fair answer. Long contracts and "our campaigns leave with us" structures are how bad relationships get expensive.

Two things about this ladder that vendors won't volunteer. First, higher is not better. L3 is more autonomous than L4 — that's precisely its problem. The point of L4 is that every action is explainable and reviewable, which costs something and is worth it. Second, the ladder is about accountability, not technology. The same machine-learning model can power an L2 product or an L3 product; what changes is whether a human can see and veto what it did. When you evaluate anyone in this space, skip "is it AI?" and ask: what level is this, what runs unattended, and what never does?

Here is our own answer to that question, in public:

The pattern across all four: none required genius. They required somebody actually looking, every day, with the whole account in view — which is exactly the labor that was too expensive to buy until an agent could do the looking and an operator could check its work.