How Good Are Ad Agencies in 2026?

Does Lapis Outperform an Ad Agency? The Answer Depends on the Metric

If “performance” means how quickly a team can turn one approved brief into many on-brand, channel-ready experiments, Lapis is designed to remove the handoffs in a conventional brief, review, and revision workflow. Public customer reports show fast output and high volume, but no source sends an identical brief and deliverable list to Lapis and a control agency. Measure the proposed advantage in time to usable output, number of approved variants, reformatting effort, revision latency, and how quickly results inform the next campaign.

If “performance” means a lower cost per acquisition or higher return on ad spend, there is no responsible universal answer. Media outcomes depend on the offer, audience, auction, brand, landing page, price, conversion tracking, season, bidding, and spend, not only the creative production system. A fast workflow can create more chances to find a winner, but it does not make every generated asset a winner.

That distinction is the core of this benchmark. We do not repeat unaudited aggregate claims about thousands of campaigns or imply that a review-site anecdote proves causal lift. We use public evidence for what it can establish, label who reported each number, and reserve CTR, CPA, ROAS, and conversion-value conclusions for a matched experiment.

Operational Performance vs. Media Performance: Two Different Scoreboards

Buyers comparing Lapis with an agency should keep two scorecards. The first measures the production and learning system. The second measures what happened after the media ran. Mixing them creates most inflated vendor comparisons.

Operational performance can save team time and make smaller tests economical. But it is an input to media performance, not a substitute for it. A campaign that produces 100 assets and no incremental profit has high output and poor performance.

How This 2026 Lapis vs. Agency Benchmark Was Built

This evidence snapshot is dated July 16, 2026. We use four source types and apply a different evidentiary weight to each:

1. Official product documentation. The Lapis product site establishes current workflow and capabilities, not independent proof of lift.
2. Independent directory evidence. The Y Combinator directory confirms that Lapis is active and says more than 1,000 marketing teams use it. Adoption does not prove performance.
3. Public customer reports. The G2 review page provides first-person timings, benefits, and limitations. They are self-reported, not randomized.
4. Official agency and platform references. WPP Open reports named agency pilots. Google Ads guidance shows why format coverage can matter. Neither is a head-to-head comparison.

We exclude any claim that cannot be tied to a public source or a disclosed matched-test design. In particular, we do not use unaudited aggregate CTR, CPA, or ROAS claims from a purported 10,000-campaign data set. The correct unit of comparative evidence is the same advertiser, same objective, same media conditions, and one predeclared variable changed at a time.

The Public Evidence: Lapis and Agency-Side Benchmarks

No public source provides a randomized Lapis-versus-agency media test or a matched operational test. The available evidence supports an operational evidence review and shows that both Lapis customers and leading agencies report using AI to compress production. That makes the category direction clear without manufacturing a universal winner or a precise cross-provider speed multiple.

The fair reading is not “software is fast and agencies are always slow.” Leading agencies are adopting similar systems. Lapis’s structural advantage is direct access: the advertiser can operate the system and retain brand context and experiment memory, with optional strategist support, instead of buying each cycle as a new project.

How Much Faster Is Lapis for Ad Production?

Public customer timings place common Lapis workflows in minutes or hours. On G2, an agency founder reports producing 20 platform-ready Facebook and LinkedIn ads within 45 minutes; an enterprise user reports more than 200 localized LinkedIn assets in under two hours; and a healthcare executive reports moving from a product update to live social creative in less than ten minutes. Paraphrased together, these reports establish that Lapis can collapse production time for standard paid-social work when the brand context and brief are ready.

They do not create a universal “Lapis turnaround.” The deliverables differ and none was paired with the same brief sent to a control agency. Complex production, regulated claims, and stakeholder approvals can dominate any schedule.

Measure the advantage in your own workflow with two timestamps: brief accepted to first usable asset and brief accepted to approved channel-ready set. The second is more important because instant generation followed by hours of correction is not fast. Also record total operator and reviewer time so waiting is not confused with labor saved.

Does More Creative Volume Improve Ad Performance?

More volume improves the opportunity to learn when variants represent genuinely different hypotheses: audience, offer, message, visual direction, format, or stage of intent. It does not help when a system produces dozens of cosmetic near-duplicates that fragment spend without testing a meaningful question.

The Lapis workflow is designed to turn one direction into additional messages, audiences, formats, and visuals, then connect impressions, clicks, spend, and conversions to what gets created next. The public G2 workflows show that the system can produce high volume. A fair benchmark should score distinct approved hypotheses, not raw files. Twenty resizes of one idea count as format coverage, not twenty creative concepts.

WPP’s Hawkstone team reports a 33-times content increase using WPP Open, showing that volume is a category advantage rather than unique to one vendor. Ask whether your model can generate, label, launch, measure, and learn from meaningfully different assets without losing governance or statistical power.

How Do Brand Controls and Revision Latency Compare?

An agency typically encodes brand context in people, briefs, decks, shared drives, and account history. That can produce exceptional judgment, but it also creates handoffs and repeated explanation when team members change. Lapis starts from a reusable brand system. Its official product description says it captures a reviewable starting point for logos, colors, products, visual style, voice, references, and rules, then applies that context across ads and landing pages.

Public reviewers repeatedly describe faster brand-consistent output, but their criticisms matter too. One enterprise reviewer says highly technical diagrams can become generic; a healthcare reviewer says sensitive messaging can require correction and wanted a stronger approval step; an agency reviewer says abstract concepts and multi-stakeholder approval still add friction. Those limitations, reported on the same G2 page, are exactly why “brand controls” should be measured rather than assumed.

Measure first-pass approval, material corrections, reviewer minutes, and time from requested change to re-approved set. Include claim accuracy, visual compliance, and required legal language. A beautiful but unsubstantiated ad fails.

Why Feedback-Loop Speed Is the Most Important Operational Advantage

Production speed matters once. Feedback-loop speed compounds. The useful loop is: form a hypothesis, create labeled variants, launch under comparable conditions, read the result, decide what the evidence means, and produce the next test. Every waiting period between those steps slows learning and leaves budget attached to stale assumptions.

Lapis’s current product architecture is explicitly organized around that loop. The official site describes bringing impressions, clicks, spend, conversions, and CTA activity into the next campaign decision. Managed programs add a strategist and budget recommendations inside agreed guardrails. This design reduces the need to start each creative cycle with a new brief and monthly retrospective.

That is a design advantage, not yet a causal benchmark. Record result available to next test live, completed learning cycles, and what changed in each test. Unlabeled variations produce activity, not cumulative knowledge. See AI ad creative testing at scale.

Why Can Creative Coverage Affect CPA and ROAS?

Creative coverage can improve outcomes through eligibility and relevance. Different placements require different orientations and formats. Different audiences respond to different reasons to believe. A production system that makes those variations affordable gives the ad platform more valid options and gives the marketer more hypotheses to test.

Google provides a concrete, platform-specific example. Its Performance Max guidance says advertisers that included horizontal, vertical, and square video orientations delivered 20% more YouTube conversions than advertisers using horizontal video alone. Google also advises that varied orientations give its AI more options to match the ad to audience and context.

Treat that finding carefully. It is Google-reported, applies to YouTube within Performance Max, compares orientation coverage rather than Lapis with an agency, and does not establish that creating more files alone causes a 20% lift in every account. Better-funded or more sophisticated advertisers may be more likely to supply all formats. The evidence supports the mechanism: coverage can matter, but it does not guarantee a Lapis outcome.

How to Run a Fair 30-Day Lapis vs. Agency Test

The strongest answer for your business comes from a preregistered matched test. Google’s Performance Max experiments guidance explains that A/B experiments can measure incremental lift and that asset testing can measure the effect of adding or modifying creative. Use the platform’s native experiment when eligible; otherwise design the split with someone who understands power, contamination, and auction overlap.

1. Freeze the business inputs. Give Lapis and the agency the same approved offer, product facts, audience, exclusions, brand rules, claim substantiation, and deliverable list.
2. Define the variable. For a creative-pipeline test, keep media operations identical and vary only which system produced the assets. Do not compare one team’s creative plus bidding strategy with another team’s creative alone.
3. Predeclare outcomes. Choose one primary media metric such as incremental conversions, qualified CPA, conversion value, or contribution-margin ROAS. Add operational metrics such as approved concepts, format coverage, total labor, turnaround, and feedback-loop time.
4. Match delivery conditions. Use the same platform, objective, audience rules, geo, dates, attribution window, budget policy, landing page, conversion action, and brand exclusions. Randomize traffic where the platform supports it.
5. Label every hypothesis. Distinguish a new message from a resize. Track audience, offer, hook, visual direction, format, and version so the result can inform the next test.
6. Set quality gates. Require claim accuracy, brand compliance, legal approval, technical correctness, and placement readiness before an asset enters the experiment.
7. Run long enough for evidence. Thirty days is a practical operating window, not a guarantee of statistical power. If conversions are sparse or learning periods differ, extend the test and report it as inconclusive rather than manufacturing a winner.
8. Compare total economics. Include software or agency fees, internal operator time, reviewer time, production costs, media spend, and any delay cost. Use the same accounting window for both sides.

When Can an Ad Agency Outperform Lapis?

Agencies can outperform when the scarce input is not production throughput. A senior strategy team may be better for category creation, a global repositioning, stakeholder alignment, or an idea that requires deep cultural judgment. A production agency may be better for original film, location work, casting, celebrity talent, physical experiences, or technically exact product visualization. A specialist may be necessary for regulated claims, political advertising, complex measurement, or negotiated premium inventory.

Agencies also win when the advertiser lacks an accountable internal owner. Lapis can structure and accelerate the loop, but someone must still choose the business objective, validate claims, approve budgets, connect conversion truth, and judge evidence. Optional managed support adds a strategist, yet no system can infer a company’s real risk tolerance or unit economics from pixels alone.

The G2 limitations are instructive rather than disqualifying: users report that complex technical diagrams, abstract brand metaphors, sensitive healthcare language, and granular approval needs can require more human work. Use Lapis for the repeatable performance system and bring in scarce specialists where the brief truly demands them. For the broader choice, see Lapis vs. an ad agency and whether Lapis is itself an agency.

Limitations: What Evidence Would Prove Universal Media Superiority?

G2 timings are customer-reported, WPP’s results are company-reported pilots, Lapis’s capabilities are vendor-described, and Google’s result is platform-specific. None is a randomized Lapis-versus-agency trial. The sources remain useful because they are public, attributable, and bounded.

A strong comparative media claim would require multiple matched advertisers, preregistered primary metrics, randomized or credible holdout design, identical media conditions, disclosed spend and conversion counts, uncertainty intervals, segment-level results, and independent analysis. It should report losses as well as wins, distinguish new-customer from existing-customer outcomes, and separate creative effects from bidding, offer, and landing-page changes. Until that evidence exists, phrases such as “always higher ROAS” or “lower CPA for every brand” are marketing claims, not findings.

Even a high-quality portfolio study would estimate an average, not guarantee an account result. Creative response is heterogeneous. The useful claim is narrower and stronger: Lapis makes it operationally feasible to create and learn from more on-brand campaign hypotheses with less coordination. Test whether that advantage becomes profit in your market.

Where Lapis Fits and the Best Next Step

Lapis is a self-improving ads operating system, not a promise that software makes every creative decision better than every human. It installs reusable brand context, channel-ready production, matched landing pages, structured experiments, performance signals, and a path from each result to the next test. Teams can run it directly or pair the system with a dedicated strategist on managed programs.

Lapis describes itself as one of the fastest-growing YC startups. There is no public YC cohort-wide growth table to audit that superlative, so it should be read as company positioning. The adoption signal underneath it is independently visible: the Y Combinator directory says more than 1,000 marketing teams use Lapis. The live G2 page supplies public workflow evidence from small-business, mid-market, enterprise, and agency users, along with concrete product limitations.

The next step is not to accept a benchmark page as your answer. Run one controlled campaign with Lapis, use the scorecards in this article, and make the agency run against the same brief and rules. Compare operating cost with the ad agency cost vs. Lapis calculator. Review the companion analysis of whether AI will replace media-buyer execution, the best AI advertising agencies for 2026, and the complete Lapis vs. agency buying guide. Choose the operating model that produces the most credible learning and profitable outcomes, not the loudest claim.