I was listening to an episode of Behavioral Science for Brands when the hosts started talking about reciprocity. MichaelAaron Flicker and Richard Shotton, the authors of Hacking the Human Mind, were discussing a familiar idea: give someone something useful before asking them to buy, and the relationship may feel different.
My first thought was that we should test it at PaintScratch. A customer who uses the Color Hub has already done some real work to find the right paint for a vehicle. What would happen if, at that moment, PaintScratch offered the customer a small gift they could use toward an order?
A year ago, I probably would have put the idea in a backlog. It touched the website, our backend, Shopify discounts, transactional email, analytics, and customer privacy. Even a small experiment can take weeks when several systems and a long list of edge cases are involved.
This time, I brought the idea to the development tools we have been working with, including OpenAI Codex. Within a day or two, we had a useful first version. We spent more time after that testing the customer experience and making the experiment safe and measurable, but we were no longer waiting weeks just to learn whether the idea was practical.
That shorter path from idea to evidence is what interests me. The idea and the decision to pursue it came from us. Codex helped with the research, the existing software, the first build, and the details that followed while the idea was still fresh.
The idea behind the $5 gift
Robert Cialdini describes reciprocity as the pressure people often feel to return a gift, favor, or concession. You do not need to know the research to recognize the feeling. When someone helps you or gives you something useful, you are often more open to helping them in return.
One of the best-known examples is Dennis Regan’s 1971 experiment. A participant who unexpectedly received a soft drink from another person later bought more raffle tickets from that person. The effect appeared to have more to do with the favor than with how much the recipient liked the person who gave it.
That does not mean a gift automatically creates a sale. It gives us a reasonable question to test: if PaintScratch offers a small, relevant gift after helping someone find a difficult paint color, will that customer be more likely to keep shopping and place an order?
There is an important limit to what this experiment can tell us. We are changing more than the language around the offer. The customer also gets a real $5 discount and a slightly different path through the site. If sales improve, we will know that the complete $5 gift experience helped. We will not be able to say that reciprocity alone caused the change.
How the PaintScratch test works
PaintScratch helps people find the exact factory paint color for a vehicle. That can be harder than it sounds. A customer may be working from a paint code, a color name, a vehicle year, a photo, or a label that does not contain the information they expected. Once they find the right color, the Color Hub already gives them a way to save it and return later.
That existing Save My Color feature gave us a natural place to run the test. One group of visitors keeps seeing the normal button, Save this color. The other group sees Save this color + get a $5 gift.
The gift is a real $5 discount offered before the customer buys anything. It is connected to the exact color they just found, so it is useful in the context of what they are already doing. A customer can email the color and gift to themselves or use the gift right away. We ask for an email in either case so that we can limit the offer to one gift per customer, but choosing to use the gift does not subscribe anyone to marketing.
We also kept the terms simple. There is no fake countdown or unsupported expiration date. If the experiment is turned off or one of the supporting services is unavailable, the customer sees the normal Save Color experience. The test should never make the existing site less useful.
What it took to make the test trustworthy
Changing a button is easy. Running a test that can answer a business question takes more work. We need to know which version each customer saw, keep that version consistent during the visit, and connect it to a later order without putting an email address or gift code into public analytics.
We also have to plan for ordinary problems. A customer may click twice, an email may fail, or Shopify may not apply the discount automatically. Someone may return after the experiment has been paused. We need a daily limit on gifts, a way to stop the test quickly, and a safe fallback when any part of the system is not working.
Codex helped across that whole process. It found the existing Save Color code and the other systems the test would touch. It helped us compare possible approaches, build the customer paths, test repeated requests, check privacy and abuse risks, and prepare the measurement plan. When a test exposed a problem, such as a customer needing a manual code when the automatic discount route failed, we could fix it and test the full path again.
Producing code faster was useful, but the larger savings came from keeping the whole process moving. We could go from research to implementation, review, and correction without losing days each time the work crossed from one system to another.
We still had to make the important decisions. We chose what the customer would see, what information we would collect, what the system could do, where it should stop, and what result would make the test worth keeping. Codex made those decisions faster to carry out, not less necessary.
Fast follow-up applies to ideas, too
At Bombyx Labs, we talk often about the cost of slow customer follow-up. A person calls, sends a message, or asks for a quote while their interest is high. If the response takes too long, that interest cools and the customer moves on.
Something similar happens with ideas inside a business. A customer points out a repeated problem, someone on the team notices a better way to handle a task, or a podcast gives you a useful hypothesis. The idea goes into a document or backlog because there is no room to act on it yet. By the time the team has time, the details are harder to remember and something more urgent has taken its place.
After working with Codex for several months, we can follow up on promising ideas much sooner. We can ask agents to research the evidence, inspect how the current system works, and help us understand the likely cost and risk before we commit. When an idea makes the cut, we can often get to a working test in days rather than treating every experiment as a new engineering project that has to wait its turn for months.
For a small business, that creates more chances to learn. A team can test a clearer follow-up message, a better answer to a common question, a simpler buying path, or a small gift offered at the right moment. Most of those ideas will not become permanent features, and that is fine. A test has done its job when it gives the business enough evidence to make a better decision.
Moving faster still requires focus
There is a downside to making experiments easier to start: it becomes easy to start too many. An agent can research five ideas and build two early versions before the team has decided which problem matters most. That can replace an engineering backlog with a pile of unfinished tests that nobody has time to measure.
So we have to be more selective about what we ask the tools to do. Before we start, we still need to agree on the customer or business problem, the result that would change our decision, the possible downside, the way to turn the test off, and the person responsible for reviewing the outcome. Easy execution makes those choices more important because the old barrier of engineering time no longer filters as many ideas for us.
The $5 gift test had a clear question and a bounded cost, and it could be added without taking away the normal Save Color experience. Those qualities made it a good experiment to pursue. Other ideas can stay in the backlog until we can explain just as clearly what we hope to learn.
What we know so far
The PaintScratch $5 gift experiment has been built and tested in a controlled, non-production environment. We have confirmed that customers can be placed into the normal or gift experience, receive or use the gift, apply the Shopify discount, and return safely to the normal Save Color path when the test is off. We also confirmed that a repeated request does not create extra gifts or emails.
We have not proved that the gift increases sales. At the time of writing, the production experiment has not been turned on, so there is no conversion winner or measured profit increase to report. When the test does run, the main question will not simply be whether more people order. We need to know whether the gift creates enough additional gross profit to cover the discounts, refunds, cancellations, and product costs.
That distinction matters because a discount can make an experiment look successful while giving money to customers who would have bought anyway. More orders would be encouraging, but more profitable orders are what would make the treatment useful to the business.
A faster way to learn what is worth keeping
We develop and train AI workers that help small businesses grow their sales. Some of those workers speak directly with customers, answer questions, collect details, or help with follow-up. Others work behind the scenes, helping a business improve the systems that support those conversations.
For us, Codex has become one of those behind-the-scenes workers. It helps us turn research into a plan and a plan into a test while keeping the human decisions in human hands. The PaintScratch gift experiment is one example of what that changes: an idea that once would have waited in a long backlog can now reach a working, testable form while we still have the context to judge it well.
What matters is shortening the time between hearing a good idea and learning whether it deserves a place in the business. That gives us more thoughtful shots on goal, with clearer limits and better evidence about which ideas are worth keeping.