Retargeting for Startups: What Works, What It Really Adds, and What It Costs You

1. Blog
2. Demand & Channels

By Saroj Jha · October 4, 2026 · 10 min read

Retargeting adds a real but modest lift, and dashboards can credit it with far more because many visitors would have come back anyway. Keep it a small, capped share of spend, act within days of the visit, and prove what it adds with a holdout.

Retargeting shows ads to people who have already visited your site. It looks like the most efficient ad spend a startup has: low cost per conversion, high return on paper. That is exactly why it deserves scrutiny.

Below are the five steps I would work through with a founder before putting more money into it. One example company runs through all of them.

Running case: Example Co. Fictional

Example Co. is a seed-stage B2B SaaS startup whose software automates the month-end close for finance teams at companies with 50 to 500 employees. Its site gets about 4,000 visitors a month, mostly controllers reading its blog about closing the books; a smaller number visit the pricing and demo pages. A customer is worth about $18,000 a year and the sales cycle is about 45 days. Its retargeting dashboard reports a high return, and the founders want to double the budget. Example Co. and all its numbers are illustrative.

Step 1: Know what retargeting really adds

Retargeting works, but less than platform dashboards suggest. Randomized experiments, which compare people who could see the ads with similar people who could not, find real effects:

• An online retailer’s display retargeting campaign “lifted website visits by 17.2% and purchases by 10.5%”^[4].
• At an online home-improvement retailer, retargeting caused 14.6% more users to return to the website within four weeks^[3].
• A working paper covering more than 2,000 advertisers found retargeting lifted advertisers’ baseline conversions by 4.6%^[1].

So retargeting helps. The question is whether it helps as much as you are paying for.

Example Co., step 1 Fictional

The team resets its expectations. It expects retargeting to add a modest number of demo requests from people who already know the product, not the large return its dashboard shows.

Do this

Write down the lift you would need to see for retargeting to pay for itself, before you look at any dashboard.

Step 2: Don’t trust the dashboard; set up a holdout

Platforms overstate retargeting because the people you retarget were already interested. Many would have returned and bought without an ad, but attribution credits the ad anyway. A study of 663 advertising experiments on Facebook found that for lower-funnel outcomes such as purchases, the median lift in the experiments was 5%, while the two non-experimental methods it tested put the median at 24% and 64%^[2]. Retargeting is likely the extreme case, because its audience is the most likely to convert anyway.

The fix is to measure with a holdout: keep a random share of visitors out of retargeting and compare. My post on incrementality testing for startups shows how.

Example Co., step 2 Fictional

The dashboard credits retargeting with 40 demo requests a month. The team keeps a random 20% of visitors out of retargeting. After two months, visitors who could see the ads request 2.4 demos per 100 and held-out visitors 2.0 per 100. On about 3,200 exposed visitors a month, that gap suggests about 13 extra demo requests a month, not 40. With only 1,600 held-out visitors so far, though, the 95% range runs from about −12 to +38 (Figure 3), so the holdout keeps running. All numbers are fictional.

Do this

Turn on a random holdout before you change the retargeting budget, and leave it running for at least one full sales cycle.

Free tool

Check whether your holdout gap is bigger than chance

The free A/B Test Significance Calculator reads a holdout like any two-group test, and its sample-size planner shows how long the holdout needs to run before the gap means something.

Step 3: Keep it small, and spend most on reaching new people

A 2026 geography-based field experiment for a bed-in-a-box mattress brand, run on Google and Facebook, found “large and consistent returns on prospecting ads across both ad networks, while retargeting ads yield significantly lower returns”^[5]. It is a single brand, but it matches the logic: retargeting can only reach people you have already reached. In B2B, John Dawes of the Ehrenberg-Bass Institute estimates that if firms switch providers about every five years, only about 5% of buyers are in the market in a given quarter^[6], and many of them may not have visited your site yet.

In practice, treat retargeting as a small, capped line in the budget that grows only when a holdout test shows it pays.

Example Co., step 3 Fictional

Instead of doubling retargeting, the team caps it at 15% of paid spend and puts the rest into reaching finance leaders at its named accounts who have not visited yet. The cap rises only if the holdout keeps showing extra demos. The 15% is a fictional choice, not a benchmark.

Do this

Set a cap for retargeting as a share of paid spend, and write down what holdout result would justify raising it.

Step 4: Match the message to the visit, and act within days

Show a useful next step, not a copy of the page they left. A field experiment with an online travel company found that dynamic retargeted ads, which show the exact product a person viewed, were “on average, less effective than their generic equivalents”. They stopped underperforming once people’s browsing suggested their preferences had formed, for example after visiting review sites^[7].

For startups, that suggests:

• Early visitors (blog readers, first visits): a helpful guide, a case for the category, or social proof.
• Late visitors (pricing page, demo page, trial sign-up): the specific offer, a comparison, or an answer to a common objection.

In B2B, retargeting on LinkedIn lets you reach visitors from target companies with content for their stage, once a website audience reaches LinkedIn’s minimum of 300 members^[8]. The LinkedIn ads post covers the setup.

Timing matters as much as the message. In the home-improvement experiment, 33% of the first week’s advertising effect happened on the first day^[3]. The value of retargeting fades fast after a visit. I found no recent peer-reviewed study that sets an ideal frequency cap, so the rest is practice: prioritize visitors from the last few days, cap how often each person sees your ads, exclude customers and people who converted, and end retargeting after a set window. On Google, any policy-compliant account can use a customer list for exclusions; targeting one takes 90 days of account history and more than USD 50,000 of lifetime spend^[9].

Example Co., step 4 Fictional

Blog readers see the month-end close checklist for controllers, not a demo request. Pricing and demo visitors see a comparison with closing the books in spreadsheets, plus answers to common IT and security questions. Both audiences exclude customers and open deals, and each person drops out after 14 days. These choices are fictional.

Do this

Split your retargeting into at least two audiences by the pages people visited, and give each its own message and a time limit.

Step 5: Check cookies and consent before you launch

The cookie picture has settled for now. Safari has blocked cross-site cookies by default since 2020^[10], and Firefox on desktop has kept each site’s cookies in a separate jar by default since 2022^[11]. Google changed course on Chrome: in July 2024 it said it would not deprecate third-party cookies; in April 2025 it said it would not roll out a standalone cookie prompt; and in October 2025 it said it would keep its current approach and retire ten Privacy Sandbox technologies^[12]. Chrome still allows third-party cookies by default; its Incognito mode blocks them^[13].

What the platforms’ own announcements show

Here is what the companies said in their own announcements and filings. It shows what they said, not how they made their choices.

Apple. Its WebKit team announced in March 2020: “Cookies for cross-site resources are now blocked by default across the board”^[10]. Its App Tracking Transparency rules name retargeting directly. Among its examples of tracking: “Sharing a list of emails, advertising IDs, or other IDs with a third-party advertising network that uses that information to retarget those users in other developers’ apps or to find similar users”^[14].

Meta. Its 10-K for 2021 described the effect from the ad platform’s side: “Apple previously released an update to its Safari browser that limits the use of third-party cookies, which reduces our ability to provide the most relevant ads to our users and impacts monetization, and also released changes to iOS that limit our ability to target and measure ads effectively”^[15].

Google. In July 2024 it wrote: “Instead of deprecating third-party cookies, we would introduce a new experience in Chrome that lets people make an informed choice that applies across their web browsing, and they’d be able to adjust that choice at any time”^[16]. In April 2025 it went further: “we’ve made the decision to maintain our current approach to offering users third-party cookie choice in Chrome, and will not be rolling out a new standalone prompt for third-party cookies”^[13]. In October 2025 it announced it would retire ten Privacy Sandbox technologies, including Topics, Protected Audience and the Attribution Reporting API^[12].

Most startup retargeting now runs inside the big platforms (Google, Meta, LinkedIn), using their own tags and server-side connections such as Meta’s Conversions API. The Meta ads post covers that setup.

Retargeting depends on tracking, so consent rules apply. This is orientation, not legal advice.

• European Union. The European Data Protection Board says tracking pixels and tracking information added to links fall under the ePrivacy consent rule, so they need consent unless an exemption applies^[17].
• California. Consumers can opt out of the “sharing” of their data for cross-context behavioral advertising, and businesses covered by the CCPA must honor opt-out signals such as Global Privacy Control^[18].

Set up your consent tool before you launch retargeting, and connect it to your ad tags. Google’s tags enforce this in the European Economic Area: a visitor reaches remarketing only after granting both the ad_user_data and ad_personalization signals^[19].

Example Co., step 5 Fictional

Before launch, the team connects its consent tool to every ad tag, honors Global Privacy Control, excludes customers and open deals, sets the 20% holdout, the 14-day window and a per-person frequency cap, and caps retargeting at 15% of paid spend. These values are fictional.

Do this

Run through the six checks above before the first ad goes live, and again whenever you add a new ad platform.

What changes by stage?

• Pre-seed — Usually skip it: too few visitors to matter. Google won’t serve ads to a list with fewer than 100 active visitors in the past 30 days^[20]. Spend on reaching new people.
• Seed — A small, capped retargeting budget for high-intent pages (pricing, demo), with a holdout group from the start.
• Series A–B — Segment by stage of interest, refresh creative often, and set the budget from holdout results, not attributed return.

What mistakes should I avoid?

• Judging retargeting by attributed return. It looks best exactly where it adds least.
• Retargeting forever. A large share of the effect comes in the first days.
• Showing the same product again and again. Offer the next step instead.
• Forgetting to exclude customers. You pay to advertise to people who already bought.
• Skipping consent. Tracking without the consent or opt-out handling the law requires creates legal risk.