Follow along with this full video walk through of customizing a search to find rock climbing gyms. This guide offers an in-depth look at the process of building a brand new project from inception to searching the entire USA for every rock climbing gym. (~20 minutes)
After opening Find Businesses, create a new project by clicking New Project.
On the Project Starter form, name the project, include some example companies, and provided some search terms that you might use to find these example companies if using a search engine.
Once the Project Starter form is complete, click Create project and udu will set up the project and test your search terms to see how effective they are at finding the example companies.
After the setup, udu shows stats about what it thinks is the best search term.
It’s always a good idea to look at the Search Term Tool to see if there is a better search term than what udu has chosen.
It’s best to use the most common search term that is still very specific that finds all of the example companies.
In other words, If using a search engine, which of the search terms that found all of the companies (all 4 in this case) would you most likely use. “rock climbing gym” seems like a very good choice in this case.
Next, add some keywords to the Keyword Scoring Module. udu has already added the initial search terms as potential keywords. Most of the time, this makes sense, but if it doesn’t you can remove or change the scoring for these keywords.
You could add some keywords that you think are relevant, but it is generally better to use the Help me find keywords tool. From this interface, you can analyze all of the example companies to find what similar keywords they are using on their websites.
Once udu is finished analyzing the example company websites, view the results and add keywords that make sense.
Skip common words like ‘bring’, ‘group’, ‘time’, ‘month', ‘team’ and ‘check’. Instead, choose industry-specific keywords like ‘day pass’, ‘birthday parties’, ‘climbing instruction’, etc.
Since these keyword suggestions are coming from the example companies, you should click Add good. The Add bad feature is used later.
Adjust the Single instance and Frequency scores for the new keywords. Click ? to learn more about these. In this case, make both positive before the initial run of the project.
Now that all of the keywords have a positive score for both Single instance and Frequency, save the changes.
Next, it’s useful to add some Required home page keywords. These are terms and phrases that must be present on a company’s home page in order for udu to spend time analyzing them. This is an important time saving technique that will help reduce the amount of companies that udu analyzes. Basically, if the home page doesn’t have either “rock”, “climb” or “boulder”, there is a very strong chance that it isn’t actually a rock climbing gym. Don’t include the word “gym”. This is very intentional as it’s likely that udu will find a lot of different types of gyms with this particular search term. The word “gym” on its own will not be a good indicator of a good company match.
You’re now ready to do an initial run to see what kind of results you get. For the initial run, you should always choose Standard as the search type and Zip codes from example company websites as the Location. (Note that the current app lists this option as Suggested > Zip codes)
Once the run is complete, you can view the results. You can see immediately that the 5 example companies are already at the top of the list, which is a good sign. You should review several of the other companies and tag them as either good, bad or mediocre. Many times, for the initial run, it’s useful to go ahead and tag all of the results if it isn’t more than 25 or so. In this case, smacasheville.com is tagged as a mediocre company because it appears to be a parent company that just has a link to climbmaxnc.com which is already on the list.
Now that the other companies are tagged, you should generally start adding more keywords. This time you should focus on possible negative keywords and add them for individual types of industries that showed up in the initial results. Camps and publications were common among the bad companies that showed up. For both of these, you can use the same technique from before where you only add the keywords that seem specific to the industry.
With all of the new negative keywords added, repeat what you did earlier and make sure all of the Single instance and Frequency scores are set to a negative number.
Save those changes and then perform a rescore on the initial run.
Now view the results from the rescore. Since there are now good and bad companies tagged, moving forward, udu will always provide a Smart score and a Stats score. For this project we should focus on the Smart score. As you can see, all of the good companies are grouped together with high scores and the bad companies are grouped together with much lower scores. This is an indication that the model is in pretty good shape, so it is time to expand the geography a bit.
Choose a larger geography to test in. In this case, since all of the example companies are in North Carolina, choose a large city there: Greensboro. (Note that in the current version of the app, udu will suggest increasingly larger geographies for new runs in the Suggested section of the New run menu)
For any runs that are larger than 5 zip codes, you can add additional seats to the run to make it go faster. This particular user has paid for access to 4 seats.
Now view the results, and do the exact same thing as with the initial run, start tagging companies as good and bad. Since there were only 5 companies that showed up for Greensboro, we can tag them all. Generally, you will only need to tag a portion of the list if it is a long list.
Since another type of industry showed up (apparel), add some more negative keywords and follow the same steps as usual.
With the negative keywords all having negative scores, We’re ready to go ahead and run a rescore on Greensboro.
By viewing the results, we can see that we have a nice separation between the good and bad companies. This means, we’re ready to expand the geography again. This time run the entire state of North Carolina.
Now that the run is finished, start tagging the top portion of the list. It’s definitely not necessary to tag all 193 of the results. We are really just looking to tag the top results.
There were some new industries that showed up, so add some more negative keywords.
With all of the keyword tool runs completed, now follow the same steps as before and add keywords from each of them.
Sometimes a keyword tool run will not turn up anything that you think is useful, so it is perfectly acceptable to add your own keywords to the list. It’s likely that “boxing” on its own would be a good negative keyword to identify boxing gyms.
Now we need to make sure all the new negative keywords have negative scores.
Next run a rescore on North Carolina.
Then view the results of the rescore.
There are some companies that filtered towards the top of the list so tag these in order to train the model further.
With the new companies tagged, there are more industries that we need to find some keywords for.
A couple of good companies were scoring a little low. It would be good to add a few more good keywords so that they are scored higher in the results.
Follow the same process for adding keywords from the keyword tool runs.
Again, there don’t seem to be any keywords on the list that would be useful so we’ll add some of our own.
For the keyword tool run on the good companies, be sure to use the Add good button.
We’re also going to add a keyword of our own and then make sure that all the new keywords have their scores for both Single instance and Frequency.
Now run another rescore on North Carolina to account for the new keywords.
Then view the results of the rescore.
As before, we have some new companies towards the top that need to be tagged.
So add some more negative keywords for these new bad industries.
Let’s also try to find some more positive keywords for the low scoring good companies.
As always, go through the tool runs and add keywords.
And again, make sure all new keywords have scores for Single instance and Frequency.
Then run another rescore on North Carolina to account for the new keywords.
A blogging website is still scoring high. We don’t generally care about blogs for this type of search so we can add an Excluded websites module to account for blogspot.com.
Excluded website modules are not taken into account for rescores since they have nothing to do with scoring and only indicate whether or not a company will show up on the list at all. So, run North Carolina again from scratch to make sure that the results are still good. (Note that in the current version of the app rescoring will take into account the excluded website modules)
Now that the fresh run on North Carolina is complete, take a look at the results.
There is one company with a high score that isn’t tagged, so make sure it is a good one.
All the good companies are at the top and there is a nice separation between the good and the bad companies. The model is good, and it is ready to run on the 49 other states.