A/B Testing for Small Business: One Test a Week

A/B testing for a small business means changing one thing on a page, an ad, a sales script or an offer, sending half your traffic to each version, and keeping whichever one wins on a number you chose before you started. Run one a week and the business gets measurably better every month, instead of every time the founder has a new idea.

Why A/B testing beats your gut

In 2012 a Microsoft engineer had an idea for how Bing displayed ad headlines. Program managers rated it low priority, and it sat for more than six months. When someone finally ran it as an A/B test, it lifted revenue 12%, worth more than $100 million a year in the US alone. It was the best revenue idea in Bing's history, and the people paid to judge ideas had shelved it.

The same Harvard Business Review piece reports that at Google and Bing only about 10% to 20% of experiments produce positive results, and across Microsoft about a third work, a third do nothing and a third make things worse. These are companies with data science teams and a lot of traffic, and their experts still guess wrong most of the time.

Your instinct about your own homepage headline is no better than theirs. A founder who changes the page because it “feels stale” is placing a bet with two-to-one odds against, and never finding out whether it paid.

Optimization is the element that replaces opinion with numbers. Every change goes in as a hypothesis and comes out as a decision.

Which lever, pulled ten percent, changes everything?

Your funnel is a chain of conversion rates multiplied together: page visit to lead, lead to booked call, call to client. A ten percent lift at one step lifts the whole chain ten percent. Lift every step ten percent and the gains compound.

Say 1,000 people a month land on your page. Here is the math with round numbers:

Visitors

1,000

Page converts at 5%

50 leads

Half book a call

25 calls

One in five closes

5 clients

Each step lifted 10% (5.5%, 55%, 22%)

6.7 clients

Three small wins, a third more clients, and you didn't spend another dollar on traffic. That is why the readout for this element is trailing lift at each step: you track the improvement at every stage over time, not one headline number.

Where to start depends on which step is furthest off. For the page, Unbounce's analysis of more than 41,000 landing pages puts the median conversion rate at 6.6%, and 6.1% for professional services. PodLab's target for a Destination page is 8–12%. If your page sits at 2%, that is your first test, and our post on landing page conversion rate covers what to change. If you don't know your numbers at each step, fix that first with the six-number dashboard in KPIs for a service business.

Optimization missing, built and running

• Missing. Five things change at once, or nothing changes at all. The site gets a redesign, the ads get new copy and the price moves in the same month, so nobody knows what worked.
• Built. One hypothesis, one test, one decision. Every week. There is a test log, a metric chosen in advance and a rule for when a test ends.
• Running. Every week ends with something permanently better. Winners become the new control, losers get written down so nobody re-runs them, and the lift at each step trends up.

The diagnostic asks one question for this element: how many tests did you run last month on a page, an ad, a script or an offer? Four or more scores full marks. One scores less than half, and zero scores near the bottom.

How to start A/B testing this week, without help

1. 01
   Pick the weakest step

   Pull last month's numbers for visit-to-lead, lead-to-call and call-to-close. Test the step that is furthest below where it should be.

2. 02
   Write one hypothesis

   Use the form “If we change X, Y will rise, because Z.” For example: if we put the starting price on the page, booked calls that close will rise, because people who can't afford it stop booking.

3. 03
   Change one variable

   One headline against another. Video above the fold against no video. Same page, same traffic source, same week. If two things change, you have an opinion again.

4. 04
   Set the metric and the stop rule first

   Decide before launch which number decides it and how long it runs. Run whole weeks so Monday traffic and Saturday traffic both count.

5. 05
   Decide and log it

   Keep the winner, kill the loser, write one line in the test log: what you changed, what happened, what you'll test next.

How long should you run an A/B test? Longer than feels comfortable. A page that gets 40 leads a month can't call a winner after six conversions, because at that size noise looks like a result. With low traffic, run the test until each version has enough conversions that the gap is bigger than a few people's choices, even if that takes a month. Free significance calculators will tell you whether a gap is real; use one before you declare a winner.

What to test when your traffic is small

Big companies can test a button color because a 0.5% lift on millions of visits is real money. You can't detect a 0.5% lift on a few thousand visits. Test the changes big enough to move a number you can see:

• The offer

  Price on the page or not, a guarantee or not, one package or three.

• The headline

  Problem-first against outcome-first. This is the first line every visitor reads.

• Video

  A founder video above the fold against a static hero. Video carries the pitch when you aren't in the room.

• Reading level

  Rewrite the page in plainer words and test it against the current one. The benchmark is below.

• The ad hook

  The first three seconds of a video ad. Meta's ads manager has a built-in A/B test, so this is the easiest split you can run.

• The first follow-up

  Two subject lines on the email that goes out after someone books. Most email platforms split-test this for you.

On reading level, Unbounce found pages written at a 5th–7th grade reading level convert at a median 11.1%, well above the 6.6% median for all pages. If your page is written in your industry's jargon, a plain-English version is one of the cheapest tests you can run.

Sales scripts and offers on calls can't be split down the middle the way a page can. Alternate them by week instead and compare close rates. It's a weaker test than a true split, so run it longer, but it still beats changing the script whenever a call goes badly.

If a test needs a developer and a two-week sprint, you will run four a year. Set the page up so a headline or video swap takes ten minutes, and you can run four a month.

How PodLab builds Optimization

Optimization is the seventh of the eight elements, and the order is deliberate. You can't test a page that doesn't exist, and you can't read a test without the numbers from Targets & KPIs. So PodLab builds the destination, the dashboard, the marketing, lead routing, sales and nurture first, then installs the testing loop on top of them.

The Business Growth System ($18,500, a 90-day foundation) builds all eight elements in that order. From day 91, ExpansionLab ($3,000/mo for one video a day) runs it, and daily video gives the testing loop new hooks and new ad creative every week. AssetsLab alone writes 50 hooks, so the raw material for next month's tests exists before the camera turns on. EssentialsLab ($3,000) builds Destination, Marketing and Sales for founders under $1M; it doesn't include Optimization, and it shouldn't, because the first job under $1M is having something worth testing.

Video is what makes testing cheap for a service business. A page with a recorded pitch can swap its opening, its proof or its offer framing without a reshoot, and the Record Once, Sell Forever framework is built around exactly that kind of reuse. When the tests start winning and traffic can go up, the next element is Scale: growing without the founder's hours growing with it.