Estimating Restaurant Workloads with Box Plots

Previously, I had posted on using box plots to estimate portions for kitchen prep.

In addition to estimating portions, a restaurant needs to anticipate how to staff each shift. As was the case with kitchen prep, box and whisker plots can help for staffing. By looking at the range of orders for shifts, you can estimate ‘how busy’ things might be. And by looking at net revenue, staffing requirements can be balanced with the cost of a higher head count.

Restaurant box plot - estimating workload

Unsurprisingly, Saturday evenings are the busiest and generate the most revenue. However, there is a wide range both above and below the median. You will want to staff more than other days, but how much? Depends on what you consider the greater risk. 1) Not having enough staff, and negatively impacting service and potentially repeat customers. Or 2) Being overstaffed and cutting into margins. (Or worse, cost exceeds net revenue.)

On the other hand, the range is quite narrow on other days, so estimating staff requirements will be easier. And Sunday is between the two extremes. Not as great a range as Saturday night, but still greater than weekdays. Unlike Saturday, the range is bottom heavy. (Data points are closer below and more spread out above. In this situation you can hedge your bets by sticking between the bottom and middle of the range.

Restaurant Kitchen Prep

How does a restaurant decide how many brussel sprouts portions to prep for any given shift? Prep too many and food gets thrown away, increasing costs. Prepping too many results in increased ticket times, which impacts service. Customers develop a negative impression as their orders are delayed.

Visualizing point-of-sale (POS) data can show historical purchase data to more accurately estimate needs. The box and whisker plot can display the variation in orders during different time periods. The box plot below shows how many brussel sprout portions, whether for a stand-alone order or as a side, were required for the morning and night shifts on each day of week.

Daily portions box and whisker plot for kitchen prep

A box plot provides more than simply looking at averages, which can be skewed by outliers. The bar portion of the graph shows how many portions were needed half the time. The center of the bar (where the color changes) shows the middle (median) portions needed. The bracket shows the top quarter and bottom quarter portion needs. By looking at the middle point and how far the box-and-whisker stretches from the middle, you estimate whether the portions required are more likely to be the lower quantities or the higher ones.

From this example, more of the shifts are bottom heavy. In these cases, it might be best to prep for a quantity around the median value. The Sunday morning shift, however, splits down the middle and covers a wide range. Whether you go with the lower or higher value depends on whether wastage or increased ticket times is of greater concern.

Another bit of information the box plot can provide are outliers. These are the data points that fall outside of the box and whisker. In this plot the Saturday night shift are all outliers, making it difficult to estimate the correct portion to prep. In this case, you might try looking at a plat that includes more POS data.

Same Information, Different Visualizations, Different Insights

With quantum science as the background, how important different visualizations of the same information can provide more (or less) insights into a subject.

https://www.quantamagazine.org/a-new-map-of-the-standard-model-of-particle-physics-20201022/

Attribution for Content

Here’s a good article about attribution models for conversions. These sort of discussions tend to be focused on the marketing attribution. Just as important is figuring out how to attribute your content’s contribution to conversions. What page on your web site was a visitor on when they converted? What other pages were they on prior to a conversion?

https://www.indicative.com/indicative-blog/how-to-choose-the-right-attribution-model-for-your-needs/

Restaurant Forecasting

Toast has a great blog that covers the many ways that you can use sales data to make business decisions. For example, there’s this article about sales forecasting https://pos.toasttab.com/blog/restaurant-sales-forecast

The article describes how you might calculate sales based on capacity and average revenue from a main dish. This is a good way to get an approximate idea of sales for busy times and quiet times.

However, you have many different dishes at different price points/profit and sales can fluctuate based on many variables, aside from busy and not busy time/days. There exists a statistical analysis that can help!

A regression analysis takes into account many different variables: time, day of the week, nearby events in the area, whatever you think is relevant and important. After you feed a bunch of data from your sales system (Toast or other) into this sort of analysis, you can get an idea of how these variables impact sales. Knowing that you can get an even more precise forecast of sales and for inventory, as well!

So for example, this analysis might say that Sundays result in $1000 more sales than weekdays and when there is an event at the local theater that will be $2000 more in sales but if it’s February that will be $500 less and so forth. You can than make some more precise (but not perfect!) forecasts for sales.

And you can do it on an menu item level—what sells when, given the above variables. And with that forecasting information, you can optimize inventory. If you know how much is selling when, you can order accordingly.

What's a good bounce rate?

Here’s an article that provides several different bounce rate benchmarks for different industries, web site types, and marketing channels.

Bounce Rate Benchmarks: What’s a Good Bounce Rate, Anyway?

What caught my attention from the article: "Bounce rate can be a deceiving KPI, appearing good or bad depending on how you evaluate and segment data"

Good coverage of bounce rate benchmarks, but definitely read the conclusion.

Designing the Most Effective Menu

With the importance of online ordering and the challenge of reduced capacity, you might want to simplify your menu by highlighting your popular and most profitable items. With toast tab and other systems, you can pull data to identify those high margin items that are contributing the most to your profit.

Graphically, those items could be charted:

All items are neatly ordered from most profitable to least profitable items. You could just remove low profit items and keep the high ones.

But you could miss some opportunities to maximize profit.

A matrix-based analysis could uncover those opportunities and drive menu item and marketing decisions. This analysis was first popularized by the Boston Consulting Group (BCG) for corporate product manages. Many variations have since been applied to different entries, including menu design.

The Toast blog talks about menu design using this concept.

A menu item matrix would something like the graphic below (based on random data for illustrative purposes).

Each section of the matrix represents a category of menu items.

In the upper right are your star items. If your online ordering platform allows you to feature items at the top of the page, these items would go there.

The lower left are the dogs (low volume, thinner margins). These are candidates for dropping from a simplified menu. But be careful that you don’t remove items that are favorites of your regular customers that are also purchasing more profitable items.

The other quadrants are where you might find some of the opportunities for more profit. As an example, you might look at the high profit ratio/low volume items and pair them with a star to increase sales. A cluster analysis (explained here) of orders and similar analyses can show you what menu items are already being ordered together. These sort of analyses can also provide insight into your regular customers and their profitability.

A menu item matrix can be a powerful decision-making tool to makeover or simply tweak your menu to be more effective and driving more profit.

See also

• e-commerce analytics for restaurants

• user testing can help your online menu

Measurements of Success

You can take different approaches to measuring the success of your online activities.

Traffic-based.

You’ve got high click-through rates on your emails and other marketing channels, so you’re generating a lot of traffic. You’re essentially interested in achieving scale, measured by Unique Visitors, Visits, or Pageviews. Perhaps, bounce rate is considered and you want it to be low, with more pages per visit viewed.

This approach is fine for an ad-based website that is getting compensated for pageviews. If you’ve ever been to a top 10 or 20 list, you’ll have seen this approach in action, when you’re required to browse to another page for each list item.

Basic Engagement.

However, if your website is intended to engage users to act or become more informed about a topic, you may want to consider measuring the quality of the traffic coming to your site, not just traffic numbers.

Some ways to measure engagement:

• time on page/site (visitors are reading the article) -
  OK but number of visitors that spent a minimum amount of time on your site is even better

• low bounce rate (visitors are interested in related content)
  OK but be careful. You can have a low bounce rate if most visitors hit pages looking for something of interest but don’t find it. Take a look at what single page visitors are doing.

• downloads (you’re offering an asset of interest to your target audience)
  OK but you don’t really know what was done with the download. Did the visitor read that PDF or not?

• exit links (you’re helping visitors find a helpful resource outside your site’s domain of expertise).
  OK but similar to downloads, you won’t know what the visitor did on the external site. (Note that this metric counts clicks on links that you provided, not visitors leaving your site by closing a tab or entering another URL in the browser, etc.)

Note the caveats for each individual metric. You can better gauge engagement by combining multiple measures into a single engagement score.

Personalization. With an engagement score, you can identify what content (and how it is promoted) leads to higher engagement scores. If your implementation allows it, you can use the personalization features of your CMS to promote content for individuals according to what content they tend to be more engaged with.

Outcome-based

Engagement is good, but it can be misleading. Are visitors spending a lot of time on your site because they find value in the content? Or are they spending a lot of time on your site, because they’re struggling to find what they need or want? It could be that single page visits are delivering more than multi-page visits.

Outcomes, come in to play here. Some outcomes will be discrete:

• purchased a product or subscribed to a service

• became a member

• donated funds to a project.

Other outcomes will be less tangible (particularly with social mission web sites). For example, visitors became more informed about a topic that impacted their lives

With discrete outcomes you can directly analyze web behaviors to understand what behaviors lead to the desired outcome. Less tangible outcomes require a survey or some other mechanism to get direct feedback from visitors and an ability to identify what web behaviors are aligned with their responses. For example, through a survey you can ask visitors if your site helped them to better prepared for their retirement.

If your analytics implementation allows you to group individual responses with overall behavior, you can gain some key insights. What types of website engagement are leading to desired outcomes? What marketing channels lead to desired outcomes? What content/articles/assets are resulting in those desired outcomes?

Optimizing Outcomes. If you can identify what web behaviors and content consumption leads to desired outcomes, you can then develop and promote web content/assets accordingly. For personalization, it’s possible you can identify visitors that are engaging in web behaviors that aligns with a particular outcome and promote content/assets to them accordingly.

E-Commerce Analytics for Pubs, Bars, and Restaurants

Now more than ever digital ordering has become a crucial stream of revenue. The dramatic increase take-out and delivery of food and drink has resulted in an overwhelming amount of information and data. It’s become all the more important to make sense of online sales data to shape the online menu that will deliver.

A platform like toast tab offers information (what’s being ordered by who) that can provide useful insight if you know where to look.

As a start, it’s great to be able to see what menu items and categories are performing well. You’ll want to promote those. And it’s easy enough to see what menu items are performing well versus others; however, if you can identify combinations performing well, you can create bundling specials and promote them. Applying certain statistical techniques to the Menu Item Details report can give you that. Your best combo might include an item that doesn’t excel on its own!

For example, you might identify a group of customers that like the house beer paired with a particular burger and side. You can bundle and promote that combo in your Toast menu. Perhaps, offer special pricing for that combo.

Likewise, you can figure out the tap line-up based on the revenue from orders that include specific beers, rather than how beers hold their own. You might generate more revenue from a beer that sells at a lower volume if customers often purchase that beer with another menu item.

Some restaurants and pubs have started offering subscription-based fulfillment, similar to HelloFresh or Blue Apron, except it’s a theme relevant to the establishment. It could be coffee, pasta, beer—any number of things. By digging into the analytics, you could possibly identify subscription packages: this coffee with this pasta and this desert, for example.

User Testing Can Help Your Online Menu

You can’t have your full menu available for digital ordering. So what food, beer, wine, cocktails do you offer? One approach would be to try some items out—whatever you think might work well—and then track the results and tweak the menu according to what is going well and what is not.

That would be a good approach, and you could do more. What would that initial menu look like?

You likely have some good intuition based on your experience. What if you could validate that intuition before releasing? Software companies often use data analysis to figure out what product users would most likely purchase. In many consumer industries, testing is used to identify product configurations that are more likely to succeed than others.

A menu is essentially a product line. Each food and drink item on the menu have different attributes like a product.

One approach, conjoint analysis, can help identify what users may prefer as a group. Not just do I prefer mayo over mustard. But given several options, one group of your customers would rather have a hamburger with mayo and a side of fries over a hamburger with aioli and a side of tater tots. Another group prefers an impossible burger with pickles and a side of sweet potato fries over a bean burger with pickles and a side of tater tots, etc.

What combination of items do large numbers of your customers prefer?

You can find a good starting point through surveys and online focus groups.

Conjoint Analysis

You can ask people to select preferred product attributes or blog topics or ask people to rank them. One of the challenges of surveying users is that what they say doesn’t necessarily match what they do.

Conjoint analysis is an approach designed to mitigate this effect. Rather than asking users to focus on an isolated attribute, a conjoint survey presents users with attribute bundles. A bundle could include screen size, color, camera resolution, and price. In a conjoint survey, users are given a series of choices between two (sometimes more) bundles of specific attributes.

For example, the first comparison might be bundle a) 5.6” screen size, gray color, 12mp camera, $700 or b) 4.5” screen size, black color, 16mp camera, $500. The user selects a or b and then is given another comparison, makes another choice, and so on. The total number of comparisons will depend on the number of attributes being tested, the different options for each attribute, and the number of user respondents.

Selecting bundles from different comparisons, as described, is a choice-based conjoint analysis. Another conjoint method is rank-based. Users rank several bundles from most preferred to least preferred.

An analysis of conjoint results reveals how much each attribute impacts the users preferences and how much each of the attribute options impacts the preference (either positively or negatively). When price is one of the attributes, conjoint provides the additional advantage of understanding user price sensitivity. By forcing choices between bundles of features, conjoint gets at the underlying, implicit value users place on the different attributes, which is likely more accurate than responses from a user that is directly asked about each attribute separately.

Conjoint is often applied to features sets of products (software, hardware, microwave, power drill, etc.). It can also be applied in other ways. For example, conjoint can be applied to online content to understand what formats, topics, lengths (summary or deep dive) that users might prefer.

An interesting article about metrics that don't measure the real world

Ran across this article that discusses how common GDP measures don’t necessarily measure optimal outcomes.

It’s time to retire metrics like GDP. They don’t measure everything that matters.

And we should be measuring what matters, rather than what doesn’t.

https://www.theguardian.com/commentisfree/2019/nov/24/metrics-gdp-economic-performance-social-progress

What is Strategic Web Analytics?

Moving forward with a web analytics strategy means that you’ve identified web metrics, aligned with your organization objectives, and are making tactical decisions based on an analysis of those metrics on an ongoing basis.

Let’s say your web site is getting lots of pageviews and visits. Does that mean your site is doing well?

It might be. If you’re, say, a publishing platform and your business model is based on selling advertising based on impressions. Then pageviews would be just fine as the metric you would care about.

But what if you were less dependent on advertising, and/or your organization had other objectives?

What if your business model was selling a subscription service or membership. Then your metrics would be related to your conversion funnel. You should focus on the web behavior, the actions and content consumption on your website, that correlate with people purchasing your service.

So the first thing you want to do, is decide what sort of attribution model you want to follow. Next, do an analysis to understand what web site content is contributing to your objective. Then you can adjust marketing and pathing on your web site to drive visitors to the high-converting content.

Or maybe, you’re a social mission organization and your aim is to educate or get donations. Then you might be interested in what web behaviors correlate with donations, but you might also need to directly ask website visitors if you’re meeting your educational/literacy outcomes. And to understand what web behaviors you need to encourage to make attain those outcomes, you’d need some way of isolating (in an anonymous way) what behaviors visitors who responded positively to your online survey were engaged in.

If the objectives of your website are a bit soft (literacy, helping visitors find a 3rd party services or organization that can meet a need) and not based on an explicit action like buying a product, you might need to employ a more nuanced approach to analytics the goes beyond traffic volume and generic web engagement (bounce rate, time on site, etc.)

Online Survey - using foresee or some other online feedback mechanism provides you with good outcome based data. You can directly ask, “did our site help you with X” This can provide you with some good directional data on whether you are doing the right things, but it does not directly tell you what online behavior led to the good (or bad) outcomes.

Cluster Analysis - a cluster analysis of online behaviors allows you to identify different distinct behaviors . It can be quite helpful to understand the different audiences on your web site, and what they are doing on your site.

Model based on assumptions - Another approach might be developing a model that incorporates behavior that you feel reflects the behavior of you target audience. For example, if you have a quiz that you think only your target audience would see through to completion, then that might be a metric you would use to characterize a target audience.

Indices - Once you have an understanding of your target audience and their web behaviors that lead to desired outcomes, you can put in place a measurement strategy that allows you to make tactical decisions on an ongoing basis. You to continuously improve, delivering positive outcomes to more people in your target audience.

What content are the people who spend a lot of time on your site often consuming? Through what promotion channel or search did they reach the site? What about the people you return to your site, but for short visits?

If you can overlay this information with outcome responses you can assess what % of your web visitors belong to your target audience and what web behaviors they are engaged in that lead to the desired outcomes.

User Discovery and Validation

You’ve implemented a new feature on your web site. You’ve published some new content on your site. And then the results are disappointing. Why?

Were you throwing spaghetti against the wall and seeing what stuck? Only to find that nothing stuck and you’re back to square one?

Why’d you do that?

Some validation through user discovery testing might have provided some guidance. What features to implement or not implement. Where content preferences might lie. What user experience might be best to deliver those features and content. Are you even trying to solve a problem for your target audience that they want solved?

It doesn’t require a cumbersome, long term research project to identify features and content that would connect with your market. User discovery and UX testing platforms exist, where you can get some sense of what might work or not — in a week or two. Provided you have a good understanding of your target audience, focus your testing on your market, and don’t try to be everything to everyone.

You can get good quantitative, survey based data complemented by qualitative usability testing. From that you can get an idea of whether you should pursue an initiative, or how you might pursue the initiative differently than you had originally envisioned. Or not pursue it at all, because it’s spaghetti that’s not sticking to any wall—regardless of how great and idea it might have seemed before you actually asked people they wanted it or now.

More Content Not Always Better

This article from the Content Marketing Institute shows the trend to publishing less frequently. Makes sense. Publishing five different articles with a slightly different spin, doesn't necessarily help visitors meet a need. Better to make one article, complemented by other content formats, easily accessible and findable. 

http://contentmarketinginstitute.com/2017/12/publishing-frequency-changing/

DIY Content Audit

Covers the basic components of a content audit, covering both attributes of the content and its performance. Gives a good foundation for a DIY content audit.

One area of disagreement: Bounce Rate by itself is not a good metric of performance. Visitors find your content and it meets their need. They leave, creating more bounces. Good? Bad?

And what bounce rate? Hard? Soft? Hard bounce rate is probably a better KPI as it takes into account engagement on your site. (If visitor hits a page and leaves, regardless if they do anything on the page, like select a link, that’s a soft bounce. Hard bounces only get counted when a visitor hits a page and leaves without interacting with the page in any way.)

You don't know unless you also consider Time on Page or Return Visits. They bounced after spending time on the page and found it valuable enough to return to your site later? Good.   

http://contentmarketinginstitute.com/2015/08/analytics-content-audits/

Assigning Monetary Values to Web Analytics

Step-by-step how to assign values to web behavior in Google Analytics, providing the foundation for ROI calculations. Especially, useful in those situations where you have overall revenue numbers but aren't able apply any attribution model for converted leads.

http://contentmarketinginstitute.com/2011/10/tracking-and-monetizing-your-website-conversions/

What is your Target Audience Doing on your Web Site?

Many people are visiting your web site. If you’re a commerce web site, they may be interested in your products. If you’re a blog, maybe their completely enraptured by every word you have to say about a particular topic, whether it be your latest travels or latest conspiracy theory about fluoride in the water. Or it might be people just passing through, who read your content, shrug their shoulders, and move on.

Most sites, can’t be everything to everyone (some e-commerce sites being a possible exception). You can’t appeal to the flat-earthers and the sane people at the same time.

From a web analytics perspective, it’s important to know the web behaviors of those you want to engage with the you site. If you are some sort of a whack-a-doodle peddling some sort of edge-case theory about the aliens living among us, you want to optimize your site for those people. It’s likely though that you may have other people visiting your site.

At any rate you have a certain audience in mind, the people that your web site exists for. You want to know what they are doing on your web site, and whether you are engaging them in the way you want, whether it’s spending time on your site, reading your content, donations, buying products/services, writing to their congress person, etc., etc.

You’ll want to isolate the web behavior of your target audience, so you know how to reach out to them, what content to path them through on your web site to meet their need.

And don't fall into the trap of awareness. “We should pursue this direction with content or site features, not because it’s directly impacting our target audience, but it's building awareness for customers/donors/engaged users of the indeterminate future. You can justify doing almost anything for awareness. But how do you measure whether you are being successful making the leap from awareness to active, relevant user? Some survey work, possibly, but unless you have unlimited resources, where do you want to expend your effort? The prospects you can directly impact now or the maybe prospects of the future?

Which Attribution Model Works Best

Provides pros/cons of different attribution models and for what situations they work best. I've implemented what I call a Last Touch/Any Touch model, which is basically the multitouch model with more weighting given to the last touch.

http://contentmarketinginstitute.com/2017/02/primer-attribution-models/

What is Content Strategy?

With the right content strategy, you'll know exactly what the word count for that blog post should be (it’s between 250 and 1500 words). Or if you should have a video instead of a web page of text. Or both. Podcast, anyone?

In my view (and many others) content strategy is how you attract the right people at the right time with the right content to make them more aware of your expertise and knowledge about a product, service, or cause. Often content strategy and content marketing are used interchangeably. (I have been guilty). Here are the basic steps of a content marketing strategy:

1. Identify target audiences from market research

2. Identify best channels (online and offline, but often mostly online) and best type of content (long blog posts, short blog posts, video, podcast, tweets, etc.) to reach those audiences

3. Develop core messaging that aligns with your brand and resonates with your audiences. You might have variations on your messaging for each audience. Hopefully, you're delivering content through your marketing channels that is personalized for each target audience.

4. Establish the tactics, including an editorial calendar, and implement the technologies that support your strategy

The idea behind content-driven marketing is establishing trust with a prospect by providing information that they are seeking and is relevant to what you have to offer as an organization. With the right content strategy, you capture their attention, and then with the right marketing automation strategy, you nurture your content-based relationship with that prospect until they become a customer, member, or donor.

That pretty much sums up content strategy as I see it. Now, I should mention that some people see content strategy and content marketing as separate disciplines with content strategy expanding beyond marketing to all of an organization's content being an asset to be managed (e.g., knowledge base for technical support) -- see what the Content Marketing Institute has to say.