# Computing Optimal Pay-Per-Click Bids In 19 Easy Steps

This post describes how to compute optimal bids in paid search.

- Suppose [1] you’re a single-channel [2] online B2C retailer [3] seeking to maximize profit [4].
- Suppose you sell a single gift product [5] via a single page website [6].
- Suppose there’s only one term [7] relevant to your offering: “gift” [8].
- Suppose you run “gift” on exact match [9] on Google [10].
- Suppose [11] that on average 2.25% of visitors to your site purchase, with an average order of $350. [12][13]
- Thus, your sales per click are .0225 * 350 = $7.88.
- Suppose this $7.88 SPC is relatively stable all day, every day,[14] all year. [15]
- Suppose your cost of goods for your product is $125, and it costs you on average $25 in other variable costs (pick, pack, ship, dunnage, phone support, etc) to get it out the door.
- So COGS is 125/350 = 35.7% and your variable costs are 25/350 = 7.1%. [16]
- Suppose you have no returns. [17][18]
- So your effective variable margin is 1 – .35 – .071 = 57.1%.
- Suppose you (the marketing person) don’t have to worry about fixed costs. [19]
- To maximize profit dollars [20], give half the effective variable margin to Google, [21] and
- Keep half of the effective variable margin for your firm. [22]
- Half of 57.1% is 28.6%, so you should spend 28.6 cents of every revenue dollar on advertising. [23]
- So your target A/S ratio is 28.6%. [24]
- Your actual SPC is $7.88 and your target A/S is 28.6%, so you need your average CPC to be $2.25 cents per click.
- Suppose every other advertiser on “gift” has equally high Quality Scores, so max CPC bid alone dictates placement on the page [25], and suppose there is so much traffic on “gift” that max CPC bid essentially matches resulting average CPC. [26]
- With all these assumptions, you should bid $2.25 cents per click for this term on Google. Voilà!

Real life gets a bit more complicated, when you’re wrangling tens or hundreds of thousands of ads in your campaigns, bidding hourly to exploit time-of-day and day-of-week SPC variations, managing products coming into and out of stock, navigating Quality Score issues, competitive bidding, tail terms, match types, negatives, content, testing, second-tier engines, and on and on and on.

But this example captures the three essential steps of economic bidding: figure how much revenue a click produces; figure out how much advertising you can spend to get a revenue dollar; and use these two metrics to compute your maximum bid.

The rest is commentary.

**Footnotes **

- If footnotes make you itch, this is good time to bail on this post.
- So we don’t have to worry about search driving orders to the call center and to stores, and we don’t have to worry about catalog and mail driving orders to search. For more on tracking search into the call center, see Call Center PPC Tracking. For more on catalog driving sales to search, see Brand vs. Non-Brand Search. In real life, these are very, very important considerations.
- Search for B2B has important differences, and search for lead-gen and services has important differences. Not covering those here.
- Not all online retail advertisers seek to maximize profit dollars. For some, revenue goals trump profit. Others care about % efficiency or ROI. Some retailers advertise for branding. Others seek maximum sales within a fixed dollar budget. All different perspectives, all valid in certain situations, all in turn lead to different bidding approaches. This post is about bottom line show-me-the-money bidding.
- Conceptually, one SKU is enough for this discussion.
- Most sites have several pages where inbound traffic on a generic term could be directed. In real life, use landing page testing to determine the best URL for each term.
- Large term lists are essential. As a rule of thumb, we recommend testing between 3 and 10 keywords per SKU. So, for a retailer with 5k products, we may test 10k to 50k terms. Half or so may fall by the wayside due to low traffic or anemic conversion, leading to perhaps 20k active terms.
- “Gift” is a high traffic “head” term, with many searches each day. That’s intentional for this example. It is also about as untargeted as you can get, so performance will stink. That is a different topic for a different day.
- Match types and negatives matter a great deal. Using exact match in this example to keep things simple. In real life, match types are
*not*a small issue, and need to be optimized carefully. - All search traffic is not created equal: traffic from different engines behave and convert differently for the same phrase, sometimes almost half an order of magnitude. Sticking with Google here just for this example.
- In real life, conversion and SPC are the most important thing
*NOT*to suppose.

If you’re seeking to maximize profit, SPC is your*key*metric. Averages and assumptions will kill you here. You good data and careful statistics. Also, in real life, AOV and conversion vary widely by term and engine. Here, as we’re running one term on one engine, we’ll ignore that essential point. - 2.25% conversion is much higher than typical in paid search, and is patently ridiculous for a generic term like “gift”. These numbers are illustrative.
- We picked “gift” here, a “head” term with high traffic. Correctly estimating conversion, AOV, and SPC on medium-traffic “body” terms and on very low volume “tail” terms is very, very important. At our firm that’s one of the most important ingredients to our secret sauce. Correctly estimating low-probability events has been an interest of mine since my doctoral research on the topic at MIT. RKG maintains collaborations with smart academicians at Rochester and Santa Clara to continue to hone our statistical algorithms. (We have some marvelous proofs for this, but sadly they will not fit in this margin.)
- SPCs vary by time of day and day of week and season of year. This topic is called “day-parting”, and doing it right is an important aspect of bidding. It is easy to do it wrong and get burned.
- Another place search practitioners and bid management platforms get burned is going into and out of holiday ‘crush’ seasons. We’ve seen data from multiple retailers and agencies showing staggering over- and under-shoot behavior when going through periods of fast change. Watch carefully. (Merry Christmas!)
- With only one product, there’s no heterogeneity in the margin structure. In real life, many retailers have products and/or product categories with different margin structures. Margin-based bidding is a big deal and not many folks seem to be doing it yet. There are several ways to bid by margin, from actual SKU margin data (we’re doing that for some clients) up to category margins (usually a reasonable approach, we’re using that for other clients).
- Wouldn’t that be great!
- If return rates and/or fraud rates are high and variable — they are for retailers in some categories — those can and should be folded into bidding through post-sale order-level feedback feeds.
- At day’s end, you do have to cover overhead and profit, but direct marketing calculations typically are done on a variable basis, unburdened. For a great intro to direct marketing math, check out Jim Novo’s stuff.
- Which, again, might not be everyone’s objective. We’re assuming here it is, see [4].
- When these numbers get big, Google starts sending
*you*holiday gifts. - See How Much To Advertise.
- From this, your firm covers overhead, profit, and taxes.
- If your margin structure is reasonably homegenous, A/S is a great proxy for profitability. If not, see [14].
- I have a nice bridge in Brooklyn to sell, if anyone is interested.
- For terms with high competition, market pressures force you essentially up to your max bid.

** Afternote**

Yep, for this post I’m trying Andy Hagan’s advice on picking titles to generate traffic. Good titles are not only highly important to blog traffic, they’re also key to natural search and to CSE feeds. Will watch how this works and report back…

Nice piece — giving it a stumble thumbs up.

Found it interesting even at its most simple this kind of sums are pretty complicated, makes you realise how much you take for granted the sums you do behind the scenes.

Excellent post and I second that Stumble. Your posts are always thorough and very interesting. A lot of newbies could learn lot just by reading your posts. Keep up the good work.

I typically only calculate optimal CPCs when setting up new programs using the same assumptions outlined above plus an assumed conversion rate based on the ‘bucket’ or part of the tail the keyword falls into as we have not yet gathered any statistically significant conversion data.

This SPC analysis is very interesting, but I am also curious; when did you decide to toss a CPA or ROAS based calculation for an ideal CPC out the window, if ever?

Hi Ryan –

Assuming an a priori conversion rate can often be dangerous. We recommend computing actual conversion rates, using appropriate formal statistical methods to handle sparse data.

And sorry if you missed the point of the post — we’ve not tossed out efficiency out the window. This toy example *is* an efficiency calculation — in the example given here, not ROAS, but equivalent. We run these economics and statistical analyses each hour on many million ads.

Cheers –

Alan

Thanks Alan for the follow up.

I would concur.

Cheers!

-Ryan

Love the PPC posts – sure wish Charlottesville was a commute away from Charleston :)

Loved your article Alan . . . very thorough. Although it takes a somewhat different route to get there, I’d be interested to know what you make of the calculator I’ve built at http://www.keywordmarketer.com/bmate1.html

Hi Gary –

As per the calculator at http://www.keywordmarketer.com/bmate2.html, there’s a deceptively simple field labeled “conversion.”

Predicting future conversion based on historic data, particularly for “long-tail” terms, is indeed the secret sauce of a good paid search bidding platform. The data are noisy, time-varying (time-of-day, day-of-week, season, etc), influenced by your actions (bids and CTRs) and by your competitors’ actions, etc. etc. etc.

It is pretty straight-forward to predict future conversion based on recent data for high-traffic terms… the true trick is getting the middle and long tail of the search phrase distro right.

Thanks for the link to your calculator!

Cheers –

Alan

Thanks a lot! It is a really useful post! An excellent explication, You are

great on this!

Alan,

Great post. It was useful to see extraneous factors tossed out so that the concept could be understood. My favorite part is the side benefit that comes along – you have to focus on all the assumptions you made to realize just how many different financial figures you have to know or calculate. It’s much easier for our firm when our client can tell us what a lead or sale is generally worth to them (especially each kind) so we can focus just on the numbers on the search end to complete the calculation. Again, well done!

Superb post! Here is the $Million Question. Is there a (good)Software Suite or Program that will monitor and optimize Bid costs on an ongoing basis, down to an hourly level? Oh, and can it include keyword tracking as well.

Somebody tell me there is something out there that does the job!