How Small Teams Can Run Low-Cost Behavioral Interventions

Last updated October 14, 2026

In an experiment by the White House Social and Behavioral Sciences Team, later analysed by Shlomo Benartzi and colleagues in Psychological Science, a behavioral intervention to increase retirement saving produced about $1,600 in extra savings for every dollar the government spent on it. Tax incentives, measured the same way, produced less than a hundredth as much savings per dollar.

Behavioral interventions do not need to be expensive or hard to scale. Small teams can use the messages and forms they already send to diagnose one behavior and test one change against a comparison group, keeping only what works. Two free frameworks cover most of the method: COM-B to find out why people are not acting, and EAST to design a fix.

At The Decision Lab, we run behavioral science projects for clients ranging from Fortune 500 companies to government agencies. The changes that move behavior in our work are often cheap to make, such as a rewritten call script or a new cabin announcement. Most of the cost sits in deciding which change to make and in proving that it worked, and this guide shows how a small team can do a scaled-down version of both.

For a single, high-volume behavior your team controls, such as sign-ups from a reminder email, a first test can usually be designed and run in about 30 days with tools you already pay for. Organization-wide change and low-volume problems such as B2B deal conversion take longer, because a month rarely produces enough cases to compare.

1. Define one specific behavior: who should do what, by when, and how you will count it.
2. Diagnose why it is not happening, using the COM-B model.
3. Generate possible fixes with the EAST checklist.
4. Rank the fixes on expected return and ease of running them, and pick one or two.
5. Test the chosen fix on something you already control, such as a reminder email or a sign-up form, against a comparison group.
6. Keep the fix if it beats the comparison group, and revise or drop it if it does not.

Behavioral interventions have a reputation for being resource-intensive and hard to scale. In most projects, though, the bulk of the money goes into diagnosis and evaluation, and the change itself is often a rewritten letter or a different default setting.

Michael Hallsworth, John List, Robert Metcalfe and Ivo Vlaev ran two field experiments on UK tax reminder letters with more than 200,000 taxpayers, adding sentences about what other people do to the standard letters. They describe the trials as cost-free to implement, and estimate that the tax authority collected £9 million more over 23 days than it would have with the original letters.

Our own work follows the same pattern. For a top Canadian insurer, we used machine learning to analyse more than 170,000 hours of sales calls and find where conversations went wrong. The fixes were targeted script changes and agent training, and the first pilot raised sales by 12% compared with a control group. A small team cannot repeat that call analysis, but it can run a lighter diagnosis with a few interviews and the data it already holds, and the script change at the end costs the same.

Low cost does not guarantee good value. Avishalom Tor and Jonathan Klick reanalysed the Benartzi cost-effectiveness data in Review of Law & Economics and argued that once every cost is counted, nudges do not consistently outperform traditional tools such as tax incentives. We take this as a reason to measure each change instead of assuming it pays off.

COM-B gives a small team a free way to turn a vague problem, such as "people aren't signing up," into a specific kind of barrier. Susan Michie, Maartje van Stralen and Robert West built it after reviewing 19 existing behavior change frameworks in Implementation Science, and found that none covered the full range of ways an intervention can work. Their model holds that a behavior happens when people have the capability, the opportunity and the motivation to do it. Around that core they mapped nine intervention functions, including persuasion and environmental restructuring, and the paper is open access.

To run it yourself, take your one target behavior and list every reason someone might not do it, drawing on a handful of short interviews and whatever drop-off data you have. Then sort each reason under the COM-B headings. If most reasons are about capability, clearer instructions or a short demo will probably do more than a persuasive message. If they sit under opportunity or motivation, changing the process or the feedback people receive usually does more than another round of communication.

We use COM-B in client work as well. When one of Europe's largest airlines asked us to reduce the litter left on planes between flights, we drew on it to rewrite cabin crew scripts and onboard announcements without adding work for staff.

Once you know the barrier, EAST gives you four prompts for designing a fix. The UK Behavioural Insights Team published the EAST framework as a free paper in 2014, drawing on its own trials and the wider research, and its principles are to make the behavior Easy, Attractive, Social and Timely. Each one points to changes a small team can make without new budget.

• Easy: cut steps from the process, or make the option you want the default. The Behavioural Insights Team cites automatic enrolment in workplace pensions as a case where a default sharply raised uptake.
• Attractive: make the action stand out and feel personal, for example by putting the reader's name and their specific deadline at the top of a message.
• Social: tell people what most of their peers already do. In the tax letter trials above, messages about what others actually do worked better than messages about what others think people should do.
• Timely: reach people close to the moment they decide, and repeat the prompt. In a megastudy of flu vaccination text reminders, the most effective messages were sent on more than one day.

EAST says little about capability, so we pair it with COM-B. Use COM-B to decide what kind of barrier you face, then use EAST to brainstorm specific fixes for it.

A short ranking step stops a small team from spending its one testing slot on the wrong idea. With the insurer above, we ranked candidate pilots by expected return and feasibility, and also by spillover, meaning whether a win in one part of the business could carry into others. For a Fortune 500 HR technology company raising AI adoption, we designed 20 candidate programs and used structured, evidence-based criteria to choose eight to pilot.

A small team can use a lighter version. Score each idea from 1 to 5 on how much it could move the target behavior and again on how easy it is to run with current tools and people, then multiply the two scores. Start with the highest product, and leave anything that needs new software or a budget request for a later round.

Testing matters for small teams because intuition about which message will work is unreliable. Katherine Milkman and colleagues sent 22 different text reminders to 689,693 Walmart pharmacy patients and compared them with patients who received no texts. On average, the reminders raised flu vaccination rates by 2 percentage points over three months, meaning about 7% more people were vaccinated than in the no-text group. Neither experts nor lay people predicted which message would perform best.

A small team cannot run 22 versions, but it can run two. Most email and survey platforms have a built-in A/B split, and where they don't, you can assign people by something arbitrary, such as whether their customer number ends in an odd or even digit. Decide before you start what you will count and for how long, and leave both versions untouched until the test ends. Small groups can only detect large differences, so test a change big enough to matter, or run the test longer to reach more people.

A randomized test is the strongest evidence, but it is not always possible or necessary. When you cannot split people at random, roll the change out to one team or site first and compare it with a similar group that has not received it yet. That evidence is weaker, and still much more informative than comparing numbers before and after a change with nothing to set them against.

Pilots do not need to be large to be useful. Our Fortune 500 AI adoption pilot ran for one month with more than 100 employees against a control group. Confidence in exploring AI tools, measured before and after the pilot, rose 41% among participants and fell 26% among control-group colleagues over the same period.

Within that scope, the four frameworks fit into a single month for two or three people working on it part time.

1. Week 1: Write down the target behavior and how you will count it, then gather your existing data and five or six short interviews. Sort the reasons people give under the COM-B headings.
2. Week 2: Brainstorm fixes with EAST for the most common barrier. Score them on impact and ease, and choose the top one.
3. Weeks 3 and 4: Run the test against a comparison group, without changing either version midway.
4. End of month: Compare the results and decide whether to keep the change. A one-page note on what you tried and what you found makes the next round faster.

A small team scales behavioral work by reusing the method, so each new test costs less than the one before. Start with one channel, write down the protocol you used, including how you defined the behavior and how you split the groups, and hand the same diagnostic and testing template to the next team with a behavior to change. In our experience, the method travels better than any single winning intervention, because a message that works for one audience can fall flat with another.

Our largest projects follow this pattern. In our Fortune 500 AI adoption work, every program came with the protocols and guides the client's teams need to run it themselves, and the programs are now scaling to the company's full 20,000-person workforce. For the insurer, the first pilot was the proving ground for a permanent in-house behavioral science team that we designed, from hiring profiles through to its first live project. In both cases the client's own staff now run the work.

The DIY approach works best when the behavior is concrete and your team controls the moment it happens, such as a sign-up form or a sales script. It struggles in two situations. The first is when the barrier runs across many groups at once, as with AI adoption across a whole workforce, where the diagnosis needs survey work and segmentation that a small team rarely has time for. The second is when the expected effect is small, because detecting a change of a point or two usually takes thousands of people in each group.

Separately, any test involving vulnerable groups or sensitive personal data needs an ethics review before it starts.

This guide is meant to help teams tell a manageable first test apart from a problem that needs deeper behavioral research or formal experiment design. That second kind of problem is where The Decision Lab usually comes in. Teams that want to build their own knowledge first can start with our free library of explainers on cognitive biases.

What is the cheapest way to apply behavioral science in a small organization?

Start with a communication you already send, such as a reminder email or a renewal letter, and rewrite it using one EAST principle chosen after a quick COM-B diagnosis. Send the new and old versions to randomly split groups and compare the results. The main cost is staff time, since the sending tools are already paid for.

Do you need a randomized controlled trial to know whether a behavioral change worked?

Not always, though a randomized trial is the strongest evidence available. When random assignment is impractical, a staggered rollout compared with a similar group that has not yet received the change still gives a useful answer. A before-and-after comparison with no comparison group is the weakest option, because seasonal swings can look like an effect.

What is the difference between COM-B and EAST?

COM-B is a diagnostic model that sorts the reasons a behavior isn't happening into gaps in capability, opportunity or motivation. EAST is a design checklist from the Behavioural Insights Team for building a fix, by making the behavior easy, attractive, social and timely. Small teams get the most from using COM-B first and EAST second.

How many people do you need for a behavioral A/B test?

It depends on how common the behavior already is and how large a change you hope to detect. A large effect, such as a response rate doubling, can show up with a few hundred people per group, while a change of one or two percentage points usually needs thousands. Free online sample size calculators give a specific number.

When should a small team hire a behavioral science consultancy?

Bring in specialists when the behavior spans many teams or customer groups, or when the effect you expect is small enough to need careful experiment design and large samples. Work with vulnerable people also benefits from an outside design and ethics review. A good partner leaves your team able to run the next round of tests without them.