Benefits of productised services

Increased scalability and efficiency

Enhanced customer experience

Streamlined operations and cost-effectiveness

Identifying Productisable Elements in Your Services

• Sales process – explain benefits and pricing
• Onboard clients – Add them to CRM, Set up a project in the project management platform, provide client access to tools for review and sign off on designs
• Kickoff mtg – understand brief collect brand assets guidelines etc
• Design work
• Design review – Gather feedback/approval from the client
• Delivery – Send final files/assets to the client

No-Code Tools: Enablers of Transformation

Key Areas to Productise with No-Code Solutions

• Workflow Automation – Use tools like zapier to automate repetitive tasks
• Data Management and Analysis – Building custom dashboards without coding
• Client Communication and Engagement – Develop interactive client portals

The Future of Service Productisation with No-Code

Digging Deeper into the Remarkable Excel Formula Bot

Earlier this week I wrote about the remarkable Excel AI formula bot which turns English descriptions into Excel expressions. For example:

Input: return the smallest non-zero value in a range

Output: =MIN(IF(A1:A10>0,A1:A10))

but in the last few days I have discovered it is even more remarkable than I first though, for two reasons.

1. it is based on a general Artificial Intelligence (AI) algorithm and not one tuned to Excel
2. the idea to create the bot itself was hatched in “conversation” with the same AI

General Excel AI (and SQL AI)

Initially, I presumed that David Bressler, the bot’s creator, had written both the front end UI and the Excel formula generation AI engine.

However, on digging deeper it turns out that it is the the general purpose GPT-3 engine at OpenAI.com which produces the formulas.

This is truly remarkable because this algorithm (to my knowledge) was never trained specifically on Excel. Nonetheless it generates sophisticated Excel expressions e.g. this screenshot is from the OpenAI playground.

It can also do the reverse and explain an existing formula:

But more than that, because it is general it extends to other areas, for example SQL:

And even “understands” the difference between different dialects of SQL:

(even thought I mis-spelled Postgres)

All of this is available in a straightforward API which the Excel Formula Bot calls to get its results.

If you want to play with the OpenAI playground, you can sign up for a free account at https://openai.com/api/.

Perhaps we should produce a connector for Squirrel365 !

An Excel AI Idea Generated by an AI “conversation”

I learned of the role of OpenAI in the Excel formula Bot from a post by its creator Davie Bressler. In the post David explains the genesis of the whole project. (It is mentioned on the front page of the Excel bot itself but in my excitement I must have missed that initially.)

Basically, David was in “conversation” with the OpenAI bot and it suggested the germ of the idea of  the Excel Formula Bot and provided an available domain name for it as well as the underpinning AI technology.

It all makes my head spin a little so perhaps better to read it in David’s own words.

No Code Strikes Again

So it turns out that the Excel Formula Bot is a no-code project pulled together in next to no time from a number underlying technologies through web based APIs.

A glimpse of the future and a story close to our hearts here at Squirrel365.

An AI Bot to write Excel formulas ???

Over the weekend I received the weekly newsletter from the chaps at Full Stack Modeller (it’s a quick and quirky read for anyone in the spreadsheet/modelling world, written alternately by Giles Male and Kenny Whitelaw-Jones, I thoroughly recommend it).

In this issue there was a short section about “Excel formula bot”, a free, AI-powered bot written by David Bressler that turns a plain English description into an appropriate Excel formula.

Kenny’s example was: “remove the last four characters from any text label” which returned him the formula =LEFT(A1,LEN(A1)-4)

Must be a lucky hit I thought. How wrong I was!

This bot is a truly remarkable (and slightly scary) piece of technology as you can see from my experiments below.

Experiments with the Excel Formula Bot

First I just checked a variant of Kenny’s example by asking “get the text between the dot and the comma in another piece of text” to which the reply was:

=LEFT(RIGHT(A1,FIND(“,”,A1)-FIND(“.”,A1)),FIND(“,”,A1)-FIND(“.”,A1))

which is correct but perhaps a little long winded, however a slight change in the request gives a better answer:

Input: get what is between the dot and the comma in a piece of text

Output: =MID(A1,FIND(“.”,A1)+1,FIND(“,”,A1)-FIND(“.”,A1)-1)

Then I remembered a post I had commented on last week about getting the smallest non-zero value in a range, so I typed “return the smallest non-zero value in a range” and got

=MIN(IF(A1:A10>0,A1:A10))

Which was at least as good as all the suggestions which had been made in the post.

Now I was hooked. I launched one of my favourites; “return the annual growth rate from two 5 year values”, to my surprise, it got this right too replying:

=((B2/A2)^(1/5))-1

extraordinary! It gets even more useful if you ask the general question “return the annual growth rate from two values” which gives:

= ((Ending Value/Beginning Value)^(1/Number of years))-1

So then I thought, “what about Squirrel365 specific ones?”, for example in Squirrel it is often useful to be able to toggle a value in a cell, and sure enough:

Input: toggle a value

Output: =IF(A1=1,0,1)

What about VLOOKUP?

Finally I went to the old stalwart which everyone knows but can never remember; VLOOKUP:

Input: the value in the third column of a B1:F17 matching A1 in the first column

Output: =VLOOKUP(A1,B1:F17,3,FALSE)

But some people prefer INDEX and MATCH so I added “using index and match” to my request and sure enough I got:

Input: the value in the third column of a B1:F17 matching A1 in the first column using index and match

Output: =INDEX(B1:F17,MATCH(A1,B1:B17,0),3)

But what about XLOOKUP? The same is true, changing to “using XLOOKUP” gave:

Input: the value in the third column of a B1:F17 where the first column of B1:F17 matches A1 xlookup

Output: =XLOOKUP(A1,B1:F17,3)

Failing on Primes

In the short time I played with it, the only thing I could find to break it was to ask it about prime numbers:

Input: return the tenth prime number

Output: =PRIMENUMBER(10)

Input: is 2343 prime

Output: =ISPRIME(2343)

Which looked good until I realised that neither of these functions exist in Excel (or Google Sheets). Perhaps this is just the bot predicting a future release!

An AI Bot for a Complete Excel Profitability Model?

Primes notwithstanding, I was truly staggered (and as I said slightly scared) at what the bot could do.

How long before I could ask for “a model to predict 15 year profitability of our network in the face of the acquisition of one of our large customers by a competitor” and get a model with all appropriate drivers in it returned in a few seconds. Before seeing the Excel Formula Bot I would have been very sceptical, now I am not so sure. Perhaps AI is coming after our financial modelling jobs after all.

If you want to try it out and amaze/scare yourself, then you can find the bot at:

https://excelformulabot.com/

It’s OK if you’ve made wrong decisions in a project for the right reasons

Making decisions in a project is not always plain sailing

Right at the start of the development of the Squirrel designer, we set ourselves some very aggressive timescales to launch version 1.0. To help us achieve these timescales, the development team agreed on some foundation principles to follow. One of these principles was to use “best-in-breed” frameworks and 3rd party components where possible. This would save us writing everything from the ground up. It would allow the development effort to focus on integrating these components along with our own code giving us a win-win on two fronts:

1. we wouldn’t have to invest time in re-inventing the wheel; and
2. we could still build a strong and scalable product.

Problems appearing out of the woodwork

As the project entered the quality assurance and alpha testing phase, we faced some challenges with one of the main 3rd party libraries we were using for the embedded spreadsheet. Problems started to appear. Whether down to bugs in the product or our non-standard way of using the component, we were having to come up with workarounds or work closely with the vendor’s support team.

Identifying the root cause

The 3rd party component was designed to be used as an embedded spreadsheet within a website — useful for performing “Excel-like” actions and importing and exporting the spreadsheet. We needed that functionality in Squirrel, but we also wanted to connect the values in cells to our components so that the spreadsheet updated the component and vice versa.

Herein lay the problem: the 3rd party library wasn’t designed for that. Although it had some of the APIs to achieve what we wanted, the more we used it the more we were straying into uncharted territory for the framework. The knock-on effect meant support issues were being logged with the vendor and fixes were estimated to take six months. This was an issue we needed to address, especially because the spreadsheet is such an integral part of Squirrel. We needed to do something to significantly increase our response times to these spreadsheet bugs.

Problem-solving in technology projects

It was time to pause and reflect. At the start of the project, we estimated it would take approximately six months to build our own spreadsheet component. This estimate was mainly derived by the sheer volume of unknowns that we’d have to work through and solve. However, as the project progressed, our understanding of what we’d need and how it would integrate with Squirrel, evolved. In October 2019 (three months before our go-live date) the decision was made to make a u-turn and to remove the 3rd party spreadsheet and build our own.

The ideal project process versus reality

Cons of rewriting the spreadsheet

This should have been a difficult decision to make. Remember how we purposefully chose an external spreadsheet at the outset, to save ourselves time? However, when looking at the bigger picture, the decision to rewrite this part of Squirrel became a no-brainer:

• We were spending more and more time fixing bugs or finding workarounds.
• We didn’t have the level of control we wanted over the turnaround time of issues logged with the vendor.
• We were uncomfortable with not having absolute control over the support and workings of such a key component of Squirrel.

The revised estimate for writing our own spreadsheet component came in at around four weeks. We green-lighted the development immediately. We had a new round of beta testing starting imminently and we wanted to get the new spreadsheet in before that so we could put it through its paces.

The decision to build our own spreadsheet would lose us four weeks of time on fixing spreadsheet bugs. We also ran the risk of introducing new bugs with our new solution.

However…

Pros of re-writing the spreadsheet

Because we’d know the codebase inside and out, this gave us the confidence that any new bugs could be fixed quickly and efficiently. It was a worthwhile trade-off and the benefits have been seen many times since the launch, including:

• A quick turn-around for creating missing Excel formulas.

Previously this would have required us to raise a support ticket with the vendor. We’d then wait for them to prioritise the development and make a release. Only after this could we implement, test, and deploy in Squirrel. Now we can create the formula and deploy it in the next release of Squirrel.

• Significantly reducing the size of a data model.

We noticed that the size of the data models in the exported projects was excessively large for complex spreadsheets. Previously, we spent large amounts of time understanding what the framework needed in order to strip out the superfluous data. Instead, we were able to immediately rationalise the data and its structure to be more efficient. And in parallel, we changed how the data was compressed to make it more efficient.

It was so efficient we were able to turn a 28Mb spreadsheet into a 600Kb data model (a 98% reduction in size). And this only took 4 hours to code and implement!

Squirrel’s spreadsheet component

Happy customers mean happy developers

With the spreadsheet now firmly in our control, we can provide accurate estimates for how long bugs and feature requests are going to take. We can schedule when the code changes will occur, and more importantly when these changes will be released in the product. Looking back at our initial foundation principles, the decision to leverage best-in-breed frameworks was the right direction to go, at that point in time.

Squirrel still has 3rd party libraries in it but we are continually reviewing the benefits that these frameworks provide compared to the responsiveness to change we require.

Writing our own spreadsheet at the start of the project would have been the wrong decision. However, changing our minds later in the project and building our own spreadsheet was absolutely the right decision to make too. It just proves that a project u-turn in software development can reap rewards.

Want to learn more about Squirrel? Keen to start building apps from your spreadsheet? Take a look around our website to find out more.