I am the Chief Custodial Technician at Rent Application, a former options trader & passionate QBASIC developer. I like country music, bourbon, Words With Friends, ice cream, Crossfit, Louis CK, and Zooey Deschanel.

How To Bid For CPC Campaigns (aka “Stop Doing Adwords Wrong”)

There Is A *Right* Answer

For each keyword in a CPC campaign, there is one single, *correct* amount to bid for a click. All other amounts are wrong. Many people bid the wrong amount. The consequence of this is lost profits.

Most online CPC bidding advice says things like:

  1. “Bid however much you would want to pay for someone to come to your site”
  2. “If you’re unsure what bid to start with, try setting a max. CPC bid of US$1″ (Really, that is what Google says. Thanks for nothing. )
  3. “Bid less than max bid = conversion rate x profit per sale”

However, most advertisers don’t know:

  • how much a click SHOULD be worth for them
  • what their conversion rate is
  • how to use campaign results as feedback to further optimize their campaign

The most important points are that at any given time there is a *right* bid that:

  1. Maximizes total profit,
  2. Can be experimentally estimated

This amount is not static and changes over time, but it’s possible to constantly correct towards the optimum bid.

cover_photo

 

Profit Per Keyword

Google gives advertisers the ability to group keywords together and apply the same bidding strategy across a group. This is great for manually managed campaigns because the amount of effort it would take to individually, manually optimize each keyword in a large campaign is unreasonable.

Thankfully though, computers are really good at doing repetitive data analysis very quickly. This approach to CPC campaign management is one where every keyword in the campaign is managed completely independently, by a computer program (this is the basic approach, but in practice, it’s sensible to group similar keywords with insufficient volume to avoid data sparsity).

Profit Per Conversion

profit_per_conversion2

This graph should be uncontroversial. It is a line graph of the amount of profit earned per conversion (y-axis) vs the CPC (x-axis).

A. An advertiser who bids $0 will earn the total amount of their conversion value for each conversion that is made through the advertising campaign.

B. An advertiser who bids $N will earn the conversion value – the cost per click (CPC) times number of clicks it takes to receive a conversion (1/conversion rate) for each conversion that is made through the advertising campaign. Profit per Conversion = Conversion Value – (Cost Per Click) * (1 / Conversion Rate)

C. The x-intercept is the Breakeven Point, which occurs when Conversion Value = (Cost Per Click) * (1 / Conversion Rate). This is the point where the amount of money spent on clicks exactly offsets the value earned through a conversion. Spending more than this leads to negative ROI campaigns. Spending less than this is positive ROI.

Note: This is the actual CPC, not the max bid. In practice, you just use the max bid as a proxy for the actual CPC, which will be lower than your max, and assume there is some secret function that Google uses to convert your max bid into a realized CPC.

 

How Many Clicks (or Conversions) Can You Buy?

buyable_clicks

Google is in the business of selling adspace in open auction. All other things constant, by bidding more money, an advertiser can get better placement and more clicks.

A. Bidding zero will give an advertiser no exposure and therefore no clicks.

B. Increasing the bid will increase the number of clicks the advertiser receives. This increase is non-linear.

C. There is a ceiling on the number of clicks that can be bought for any particular keyword or query (even if you had a billion dollars to spend tomorrow, you couldn’t buy a million clicks for the keyword [elephants in pant suits never eat donuts] because there isn’t enough search volume to sell you that many clicks).

 

Total Profit Per Keyword

total_profit

Multiplying the information from the preceding graphs gives us the total profit earned for a particular keyword.

For example:

Conversion Rate = 5%
Conversion Value = $10.00

For this particular example keyword, increasing CPC linearly increases total clicks purchased such that every 5 cents higher the CPC bid is results in 100 more clicks over the measurement time period.

  • 0 clicks * 5% CR * $10 Conversion Value – 0 clicks * $.125 CPC = 0
  • 250 clicks * 5% CR * $10 Conversion Value – 250 clicks * $.125 CPC = 93.75
  • 500 clicks * 5% CR * $10 Conversion Value – 500 clicks * $.25 CPC = 125.00
  • 750 clicks * 5% CR * $10 Conversion Value – 750 clicks * $.375 CPC = 93.75
  • 1000 clicks * 5% CR * $10 Conversion Value – 1000 clicks * $.50 CPC = 0
  • > 1000 clicks == >1000 clicks at >$.50 == lose money

A. zero spend == zero clicks == zero conversions == zero profits
C. this is the max profit CPC price. paying this amount for clicks balances total traffic and price per visit in a way that maximizes profit.
B & D. there are 2 ways to make less than the maximum amount of money. B: buy less traffic and pay less per click or D: buy more traffic, but at a higher cost per click. B is safer because small errors don’t result in losses.
E. the other way to make no money is to buy a lot of traffic but spend the total conversion value on generating traffic (breakeven CPC ( $10* 1 conversion per 20 clicks) == 50 cents/click).
F. lose money — spend more money on traffic than you recover through the conversions. whenever you hear someone say “we tried adwords, but it didn’t work for us”, this is where they were on the curve.

 

Implementation

It’s important to note again that you can’t look up the information to set your CPC bids or calculate it theoretically — it has to be determined experimentally. There’s nothing you can do to just start with the perfect campaign. You will make your best guesses, experiment, then use the feedback to estimate the shape of the curve.

That means you will start by running suboptimal campaigns, then use the information to get closer to optimal. The better your initial guesses, the less money you’ll spend in the “curve discovery/estimation” part of the process.

If you already have conversion rate data, you can start by applying the site global conversion rate as your estimated conversion rate for each keyword, then that gives you point E, the breakeven spend CPC.

If you don’t have any conversion rate data, now’s as good a time as any to start buying traffic to see what the conversion rate is. You may lose money doing this, but you should treat this expense as a cost of information. Once you have some basic conversion rate data, you’ll be able to estimate your breakeven spend CPC. Being cheap here is a mistake. You pay less for worse data. Since this is information you want to base your campaign on, consider it an investment in your future.

Once you’ve estimated point E for the keyword, your goal is to test various CPC levels. To do this, vary your max bid over several time periods, and at each CPC level, recording the profit generated by that keyword. Assume that the profit vs CPC curve qualitatively fits the model above, then use the collected data to estimate what the curve looks like.

Do this around 1/3 E, 1/2 E and 2/3 E. In the example above, if you have a a $0.50 breakeven CPC price given your global (as opposed to keyword) conversion rate, try 16 cents, 25 cents, and 33 cent CPC’s to see what the corresponding profits are. Use these results to estimate the maxima for each keyword in your entire portfolio independently.

Each keyword in the portfolio should convert at a different rate. This process will help estimate the CPC amount that maximizes the profit for a keyword while accounting for the different conversion rate for each word.

Additionally, if there’s reason to believe that there are different Conversion Value or LTV’s for different keywords, you can use this information as well to optimize. However, in practice you need a lot of conversions for this data to be reliable and except for the highest volume sites, this will be difficult to execute, so it’s easier to just use the same keyword. Obviously this depends on the kind of site you have: An electronics retailer that bids heavily on product name keywords to sell those should probably use the actual profits generated by the keyword (TV’s are more expensive than USB cables). A SaaS product that has one price can probably just plug in their product price.

 

Maintenance

Because the correct bid price for a keyword is a function of changing inputs outside of your control, for each keyword you should be constantly re-assessing whether the “optimal price” estimate is correct.

To do so, slightly change bid prices after you have enough conversion data at a specific bid price. As the CPC’s change you will be constantly traversing up and down the curve to either: (a) discover that the shape has changed, or (b) confirm that the shape is still as you estimate.

You should stay NEAR the optimal bid price as you probe the curve for information in order to take advantage of what you’ve learned. When the shape changes, you’ll adjust your bids to the new maximum.

 

Alternate Curve Shapes

Not all keywords can or will be profitable (lots of competition for ads, low conversion rate, low conversion value).
Some will be always profitable (your searchers are already looking to give you money, high conversion value, high conversion rates, no competition in the keyword space).

The following 2 are just special cases of the general shape described above.

Always Lose

Joe’s company “Get Face Punched, LLC” sells a coupon for $1000 that entitles the holder of the coupon to be punched in the face by Joe. He advertises on Google Adwords. He’s disappointed in his 0% conversion rate.

always_lose

For a product that converts too poorly, no bidding strategy can turn a profit. Every dollar spent is a dollar lost. Joe can’t turn a profit because he just never converts. He can fix this by changing other non-bidding elements (the product, the landing pages, the ads, etc) to change the shape of the Profit vs CPC graph.

Similarly a product whose Conversion Value is too low won’t be able to turn a profit — It’s hard to pay money for traffic and recoup it without collecting any money back.

In the case of free products, they may operate at a short term loss, but the operator needs to know what price they are willing to pay for a conversion, whether or not it corresponds to future financial returns.

The method of traversing to find the optimal CPC bid will still work for a keyword like this.  As you continue to traverse to higher ground, the optimal CPC will turn out to be $0.00, and you will effectively never bid for this keyword. As a shortcut during the optimum CPC discovery stage, whenever the measured profitability of a keyword at a given CPC is negative, halve the CPC instead of doing small traversals.

Always Win

Sarah’s company “Insanely Comfortable Luxury Bed” sells $50,000 handmade beds made from angel wings and confer amazing sexual prowess on its purchasers. It’s a small market, but pretty much everyone who ever looks for it online has already made the buying decision. Her conversion rate is insanely high at 50% and there’s not a lot of competition for “angel wing beds make me a sex god(dess)” so her CPC’s are shockingly cheap.

always_win

In contrast, with a product that converts awesomely, the payout is massive, or competition is scarce, it’s actually hard to *lose* money. While there’s a theoretical breakeven point at extremely high CPC’s, the conversion rate is too high and there isn’t enough competition to ever drive costs high enough because Google will ultimately cap the effective CPC at the highest competing bid.

Additionally, since there is a cap on the number of clicks available for google to sell, for low volume keywords, hitting that cap means you can’t spend any further.

The same traversal method for finding the optimal CPC bid works here as well. You’ll continually bid more for clicks until you’ve reached the end and there are no more buyable clicks.  Bidding higher through Google adwords won’t result in higher actual CPC’s because you’ll have outbid the competition. Also, once you’ve purchased all available clicks, you won’t be able to spend anymore so the CPC bid will settle into the maxima at the top of the line segment.

 

Improving non-CPC-Bidding Problems Leads to Curve expansion

The correct amount to bid is a function of a lot of things, including:

  • the keyword
  • quality score
  • competition
  • day, time, place, device
  • conversion rate
  • conversion value (based on your landing page, your product awesomeness, other factors)

All these things are subject to change at any time.

For a given situation/world, there IS one single optimal bid price. That does NOT mean that for a given keyword there’s only one bid price.

It’s probable that unoptimized ads/landing pages will show that a lot of keywords have NO profitable bid opportunities. By improving conversion rate, quality score, ad copy, or any number of things, it’s possible to change the shape of the ad keyword profit curve, which necessitates recalibrating the keyword bid.

The specific bid price is just one small component of the process that can be optimized. It ensures that given your existing campaign, you’re allocating your ad spend in the most capital efficient way (The biggest possible piece of an existing pie)

However, changes to the campaign can expand can increase the available opportunities (making the pie bigger):

  • better ads/relevance lead to higher click-through-rates, higher Quality Scores, and lower bid prices for the same position, therefore growing profits and stretching the breakeven point to the right.
  • better landing pages can be tested to improve conversion rate, which decreases the cost per conversion and grows profits and stretches the breakeven further to the right.

 

Notes

  • In practice the curve isn’t so pretty and the falloff from the peak may not be symmetric.
  • If in doubt and uncertain of the actual shape for the keyword, its safer to underbid than overbid (lose money the other way, read about 1/2 kelly betting)
  • The shape of the curve can be improved, increase total max, by doing things on-site to improve conversion rate, improving your actual ads to also improve conversion rate, working on your quality scores, etc.
  • Google will curb your actual CPC — this plots actual CPC not bids. for example, i can’t pay 50 bucks for a single click of an obscure keyword because the bidding system will just have me outbid the highest bid. As a result, advertisers are protected from the most devastating failure cases (i.e. WAAY far to the right on the CPC/x-axis where the total profit would be hugely negative).
  • You need to be able to buy enough clicks in a steady enough environment in order to credibly estimate the shape of the curve. So low volume keywords will either be impossible to estimate, or you’ll have to combine them with other keywords to form a basket of keywords for which the shape can be estimated. This comes at the price of fine-grained keyword specific pricing control. If you bucket keywords that perform differently and estimate the shape of the aggregate click portfolio, you’ll end up with a shape that could be qualitatively different so you might not be appropriately estimating a peak (it could be multi-modal with many local maxima instead of a single global maximum).

Also, the corollary to this is that there’s a correct amount to spend, and it’s based on your correct CPC amount. It’s “however many clicks google will give you at the specified CPC price”. The budgeting advice that sounds like “As much as you feel comfortable spending in a day” is bad advice.  It’s the right advice for campaigns that are designed to be negative expected value. The right advice for campaigns that are positive EV (those that make money) is “spend as much as Google will let me”.

(thanks, http://www.onemotion.com/flash/sketch-paint/ for letting me make pictures.)