How to Run HADI Hypothesis Testing for Client Campaigns: A Performance Agency Guide

How to Run HADI Hypothesis Testing for Client Campaigns: A Performance Agency Guide

HADI hypothesis testing gives performance agencies a repeatable system for running structured experiments across client campaigns. Instead of ad hoc tests that produce one-off results, HADI - Hypothesis, Action, Data, Insights - turns every experiment into documented knowledge that improves decision-making across your entire client portfolio.

Key takeaways

Fewer Than a Quarter of Advertisers Run Controlled Tests Regularly

Per Meta research, fewer than 25% of advertisers run controlled tests regularly - at agencies, the proportion is likely lower because pressure to deliver results across multiple simultaneous accounts creates an incentive to act on instinct rather than wait for structured data.

Test Results That Live in Slack Threads Die With the Project

The structural failure of ad hoc agency testing is not that tests don't happen but that results live in individuals' heads, Slack threads, or unlabelled spreadsheet tabs. The same experiment gets repeated from zero on a different client three months later because no searchable record exists.

HADI Testing Turns One Client's Result Into the Starting Hypothesis for Another

HADI testing at agency scale treats each client campaign as part of a shared experiment portfolio. A confirmed hypothesis on Client A - such as UGC video outperforming polished creative for cold audiences - becomes the starting hypothesis for Client B with a similar profile, not a from-scratch test.

Four Fields Are the Minimum for Any HADI Record to Produce Transferable Knowledge

The minimum is: Hypothesis (if X then Y because Z), Action (specific test parameters), Data (quantified outcome vs pre-defined success metric), and Insight (the transferable rule). An experiment missing any of these four fields still produces knowledge that dies with the project.

A Well-Maintained Hypothesis Board Surfaces Tested Responses in Thirty Seconds

The highest-leverage moment for a HADI board is when a client's ROAS drops and the account manager can search the hypothesis backlog rather than starting fresh. A well-maintained board typically surfaces one to three tested responses to any common performance problem within 30 seconds of searching.

The operational problem agencies don't talk about

You're running paid campaigns across eight clients simultaneously. One client's Meta ROAS drops. You test a bid strategy change. It works. Three weeks later, a different client hits the same problem - and nobody on the team remembers what fixed it last time.

This is the default state for most performance agencies: experiments happen, but the results live in individual account managers' heads, Slack threads, or unlabelled spreadsheet tabs. When a test works, the learning dies with the project. When a test fails, the agency repeats the same mistake on the next client.

Research by Meta indicates that fewer than 25% of advertisers run controlled tests regularly. At agencies, the number is arguably lower - because the pressure to deliver results for multiple clients simultaneously creates an incentive to act on instinct rather than wait for data.

HADI hypothesis testing is the operational fix. It forces every experiment through the same four-stage gate, makes results searchable, and creates a growing library of confirmed and failed hypotheses that every account manager on the team can draw on.

What HADI means (and what it doesn't)

HADI: A four-stage experiment framework standing for Hypothesis to Action to Data to Insights. Each stage produces a documented output before the next stage begins.

HADI is not A/B testing. A/B testing is a method for collecting data. HADI is the management system around the test: what question you're asking before you run it, what data you'll look at when it's done, and what decision the data produces. You can run an A/B test inside a HADI cycle, but HADI also works for bid strategy changes, audience swaps, landing page tests, creative rotations, and budget reallocations - any change where you're testing a causal claim.

HADI is not a sprint methodology. It has no fixed two-week cadence. A cycle can run for 5 days (a quick creative test) or 6 weeks (a funnel conversion test requiring statistical significance). What matters is that each cycle is formally closed before the next one opens.

Why agencies need a structured system - not just good instincts

A single account manager with strong instincts can manage four clients reasonably well. A team of six account managers handling forty clients cannot. The moment your agency scales beyond one person, informal knowledge transfer breaks down.

Structured marketing experiments address three failure modes that grow with agency size:

1. Siloed results. An account manager discovers that extending the attribution window from 7-day click to 28-day click changes reported ROAS significantly on one client. That finding should immediately be tested on every comparable client. Without a shared experiment log, it never is.

2. Failed tests repeated. Without a hypothesis backlog, different account managers test the same idea on different clients, get the same negative result, and document it nowhere. The agency spends budget on the same failed experiment three times.

3. No benchmark for future decisions. When a client asks "how do you know this bid strategy works?", the answer should be "we tested it on four comparable accounts and documented the results" - not "we've had good results before."

For a deeper look at how agencies can structure their entire analytics operation, see the marketing analytics for agencies framework.

1. Set up a HADI board per client (not per team)

The first structural decision is scope: one HADI board per client, not one board for the whole agency.

Why per client? Because a hypothesis that tests well for a B2C ecommerce brand does not automatically apply to a B2B SaaS client. Audience sizes, attribution windows, seasonal patterns, and margin structures differ. Mixing experiments across clients in a single board creates noise that makes results harder to interpret.

Each board holds four columns: Backlog (hypotheses queued for testing), In Progress (active experiments), Completed (documented results), Archived (superseded or irrelevant hypotheses).

For agencies using Prooflytics, the HADI Kanban at /hadi-hypothesis-testing maps directly to this structure. Each hypothesis card captures the complete cycle from Hypothesis through Insights, and the board is scoped per client account so results stay separate.

The backlog is the most important column. A healthy backlog has 8-15 hypotheses queued at any time per client - enough to maintain experiment velocity without creating decision paralysis.

2. Write hypotheses in the correct format

The most common reason agency experiments fail to produce useful knowledge is an incorrectly written hypothesis.

A HADI hypothesis must:

• State a specific, falsifiable causal claim
• Name the metric that will confirm or reject it
• Define the success threshold in advance

Weak hypothesis: "Better creative will improve performance."

Correct hypothesis: "Replacing static images with 15-second UGC video creatives in the awareness campaign will increase CTR from 0.8% to ≥1.2% over 14 days at ≥£2,000 in spend."

The correct format forces you to commit to a number before seeing results. This prevents the most common agency testing error: deciding the test "worked" because some metric improved, even if the specific metric you were trying to move didn't.

For each hypothesis, document the trigger - what observation or data point generated it. "Client CPL increased 35% over three weeks" is a valid trigger. "A feeling that the audience is fatigued" is not. Every hypothesis should trace back to a data signal, a platform notification, a client brief, or a finding from a previous HADI cycle.

3. Define the experiment before touching the campaign

Once a hypothesis moves from Backlog to In Progress, the experiment design must be documented before any campaign changes are made. This is the Action stage.

For each experiment, record:

• Control: what the campaign looks like now (creative set, bid strategy, audience, budget)
• Variant: the specific change being tested
• Duration: minimum run time based on expected traffic volume
• Minimum spend threshold: the minimum budget required to reach meaningful data
• Decision date: a fixed calendar date when you will evaluate the result - not "when it looks ready"

The decision date is non-negotiable. Without it, experiments drag on indefinitely and get evaluated opportunistically - which introduces confirmation bias. Set the decision date at the start. If the data is inconclusive on that date, document the result as inconclusive and close the cycle. Do not re-run the same test until a favourable result appears.

For agencies managing ROAS-sensitive accounts, the ROAS floor rules framework gives the baseline thresholds needed to set your minimum spend threshold correctly per vertical.

Prooflytics

Make the call with the whole picture

Briefs are daily; the understanding compounds.

Start free trialSee pricing

14 days free · no credit card

4. Collect the right data - not all available data

The Data stage has one rule: only look at the metric you defined in the hypothesis.

This sounds obvious. In practice, when an experiment runs for two weeks, account managers tend to look at every metric in the account - and find evidence for whatever conclusion they already believed. ROAS improved but CTR dropped. CPC fell but conversion rate went down. There is always something that "worked" if you look at enough numbers.

Before closing a HADI cycle, record:

• Primary metric result: hypothesis metric only - did it hit the threshold?
• Statistical context: total spend, impressions, clicks, conversions over the test period
• External factors: any platform changes, budget edits, creative refreshes, or external events during the test window that could confound the result
• Verdict: confirmed / rejected / inconclusive

The verdict is binary for the hypothesis itself. A test that moved CTR from 0.8% to 1.1% when the hypothesis required ≥1.2% is a rejected hypothesis - not a partial win. Document it as rejected with the actual result noted. This matters when building a cross-client knowledge base: partial wins get misremembered as wins.

5. Write the Insights output for the team - not just for the client

The Insights stage is where HADI creates compounding value for agencies. Most teams skip it or produce a one-line note. The correct output is a structured record that a different account manager can read six months later and apply to a new client.

Each Insights entry should contain:

• What happened (1-2 sentences, metric result)
• Why it probably happened (your interpretation - flagged as interpretation, not fact)
• Transferability (which other client types or verticals this finding is likely to apply to)
• What to test next (the follow-on hypothesis this result suggests)

This output lives in the Completed column of the HADI board and is the raw material for cross-client learning sessions. When a new client joins the agency in the same vertical, the account manager should spend 30 minutes reading the Completed column of the most comparable existing client's board before building their first hypothesis backlog.

For agencies already producing structured weekly client deliverables, the HADI Insights output can feed directly into the white-label weekly report - giving clients a narrative of what was tested and what was learned, not just what the metrics did.

What the data shows about structured experimentation at agencies

The operational pressure most agency account managers feel is that running experiments seems slower than making direct optimisation calls. Why test when you could just fix?

Harvard Business Review research has found that brands running automated, structured experiments improve ad performance by 30-45% in the first two years, compared to teams using judgment-based optimisation alone. The compounding mechanism is the knowledge base: each documented test makes the next decision faster and more accurate, because the team is drawing on verified findings rather than recalled impressions.

For agencies, the compounding effect is larger than for in-house teams - because a confirmed hypothesis on one client account can be applied immediately across comparable accounts in the portfolio. An agency running experiments across 10 client accounts with a shared knowledge base effectively multiplies each test result by 10. An agency running ad hoc tests across 10 client accounts gets 10 unrelated data points that never compound.

Growth Rocket's research on agency experimentation culture found that without a centralised experiment log and governance structure, results are inconsistent, siloed, and rarely transferred across client accounts. The defining difference between high-performing agencies and average ones is not testing frequency - it is knowledge transfer infrastructure.

Prooflytics HADI hypothesis testing surfaces the hypothesis backlog, active experiments, and completed findings in a Kanban view scoped per client. When competitor intelligence shows a rival brand testing a new campaign format, the competitor-to-HADI one-click flow turns that observation into a hypothesis card immediately - without switching tools.

6. Run cross-client knowledge transfer sessions monthly

The structural advantage of per-client HADI boards only materialises if the agency has a mechanism for synthesising learnings across them.

Schedule a monthly 60-minute session with all account managers. The format:

1. Each account manager presents 1-2 completed HADI cycles from the past month
2. The group identifies transferable findings: which other clients could benefit from this test result?
3. Any confirmed hypothesis is added to the agency-wide hypothesis library with a transferability tag (ecommerce, B2B SaaS, lead gen, DTC, etc.)
4. Any new hypothesis generated by competitor observations or platform changes is added to the relevant client backlogs

This session is the highest-leverage 60 minutes in the agency month. It converts isolated client work into institutional knowledge. Over 12 months, the agency builds a library of confirmed and rejected hypotheses that makes onboarding new account managers faster, pitching new clients more credible, and optimising campaigns more systematic.

Frequently asked questions

What is HADI hypothesis testing in marketing?

HADI is a four-stage experiment framework: Hypothesis (a specific, falsifiable causal claim), Action (the designed experiment), Data (the collected results), and Insights (the documented learning). In marketing, it provides a structured system for testing campaign changes - bid strategies, creatives, audiences, landing pages - with documented outcomes rather than ad hoc optimisations. Each completed HADI cycle produces a searchable record that improves future decisions.

How many HADI experiments can an agency run at once per client?

One active experiment per campaign objective at a time. Running multiple simultaneous tests on the same campaign makes it impossible to attribute a result to a specific change. An agency managing a client account with four distinct objectives (awareness, traffic, conversion, retention) can run up to four concurrent HADI cycles - one per objective - without contaminating results.

How long should a HADI cycle run?

Minimum run time depends on traffic volume and the metric being tested. For click-based metrics (CTR, CPC), most campaigns need at least 1,000 impressions per variant before results are meaningful. For conversion-based metrics, you typically need 50-100 conversions per variant to reach statistical confidence. For low-traffic campaigns, 14-21 days is a practical minimum to smooth out day-of-week variance regardless of impression volume.

What is the difference between HADI and A/B testing?

A/B testing is a method for splitting traffic between a control and a variant. HADI is the management framework around a test: it defines what question you're asking before the test runs, what data you'll evaluate, and what decision the data will drive. You can run an A/B test as the Action stage of a HADI cycle. But HADI also covers changes that cannot be A/B tested - bid strategy shifts, audience exclusions, budget reallocations - by requiring a documented before/after analysis with a pre-defined decision date.

How does HADI hypothesis testing help with client reporting?

A completed HADI cycle produces a client-ready narrative: what was tested, what happened, and what the agency will do next. This replaces the most common agency reporting failure - presenting metrics without explanation. Clients who understand that their agency runs structured experiments with documented outcomes are more likely to approve test budgets, accept inconclusive results, and maintain longer engagements.

Bottom line

• HADI turns every campaign experiment into documented knowledge - confirmed or rejected, not "it seemed to work."
• Per-client HADI boards keep results separate; monthly cross-client sessions convert them into agency-wide intelligence.
• The hypothesis format is the most important discipline: name the metric, set the threshold, commit before the test runs.
• The decision date is non-negotiable: it prevents experiments from being evaluated opportunistically.
• Agencies with a shared hypothesis library onboard clients faster, retain them longer, and scale without losing institutional knowledge.

You can read independent reviews of Prooflytics on G2 and compare it to alternatives in the marketing intelligence category.

Start a free trial at Prooflytics - the HADI Kanban is available from day one, with competitor-to-HADI one-click import from your competitor intelligence feed.