The Open-Exchange Pitch at Hearth & Home
Hearth & Home Furnishings operates 38 stores across six DMAs and will spend $900,000 in Q4. The current mix is 40% local TV, 25% radio, 20% paid search, and 15% paid social. Store traffic is down 6% year over year, e-commerce is 8% of revenue, average ticket is $1,400, and customers typically research for four weeks before buying, often online and then in a store. The incumbent agency proposes moving 60% of the budget into open-exchange programmatic display, guaranteeing a $45 cost per acquisition with conversions counted on a seven-day click or one-day view window. It also recommends removing frequency caps to maximize unique reach, and running all creative through a single ad server to save $30,000 in fees. Two vendor co-op programs worth $220,000 combined require brand-safe placements and pre-approved creative, and the money lapses if unspent by December 31.
1. Evaluate the agency proposal and restructure it. Which elements do you accept, which do you reject, and on what reasoning?
Read the worked answer
The proposal fails on four counts and should be renegotiated rather than accepted or rejected wholesale.
First, the $45 CPA is measured against a definition - seven-day click or one-day view - that is close to meaningless in a four-week, $1,400 consideration. Display sits far from the purchase, so view-through conversions will claim credit for orders that search or a store visit would have produced anyway. The guarantee is priced against inflated credit, not incremental revenue.
Second, open exchange conflicts directly with $220,000 of co-op money that requires brand-safe placements. Open exchange means unknown placements and adjacency risk. Restructure into a curated private marketplace or an allowlist with supply-path optimization, independent verification, viewability at 70% or better, and a firm exclusion list.
Third, removing frequency caps does not maximize unique reach; it recycles the same users. At this ticket and cycle, cap at roughly three to four impressions per user per week and spend the remainder on incremental reach.
Fourth, consolidating creative into one ad server to save $30,000 may break co-op reporting and vendor approval - a far larger sum at risk than the fee saved.
Revised shape: keep search fully funded because it captures the four-week research window; retain local TV if it is delivering reach inside store trade areas; take programmatic up to roughly 30–35% but inside a PMP with the guardrails above and geo-fenced to trade areas; ring-fence 10% for a test. Refuse any guarantee tied to view-through conversions - if a CPA guarantee is wanted, define it on incremental store visits or new-customer orders with a holdout.
2. Design the in-market test that would prove whether display drives store visits. Be specific about design, duration, and the limits of what 38 stores can tell you.
Read the worked answer
Use a matched-market design, because the unit of decision is stores, not users.
Define the outcome first: store visits where door counters are reliable, otherwise store-level sales in the trade area with an adjustment for online-assisted purchases. Pull 12 months of pre-period data per store and match pairs on volume, trend, and local demographics. With 38 stores across six DMAs, expect roughly 12–16 usable pairs. Randomly assign one store in each pair to the treatment (added display weight) and leave the other at business as usual.
Duration: eight weeks minimum, with the first two weeks treated as ramp and excluded, because a four-week consideration means the sales effect lags exposure. Measure with difference-in-differences against the pre-period and report a confidence interval, not a point estimate.
Be explicit about power. Thirty-eight stores is a small sample; only a large effect will be detectable, so pre-register the effect size you would act on and accept that a null result means not proven, not proven absent. Add cheaper corroboration: branded search lift by trade area, foot-traffic panel data, and a post-purchase survey asking recent buyers whether they recall the ads.
Guardrails: do not let treatment stores also receive extra search or social during the test, or the result becomes uninterpretable. Keep creative and co-op approvals identical across cells. Decide in advance what happens if the test reads flat - the honest answer is to cut the display budget, not to rerun the test until it passes.