Lessons from one too many model spreadsheets

Few things can be as helpful and potentially time-wasting as modelling your business with a spreadsheet and setting metrics. A clear model of how you make money is a great tool to help identify bottlenecks or risks. On the other hand, it is very easy for models and the team KPIs or metrics that result from them to be little more than noise. Previously, as a founder and in non-founder roles, I’ve had to build many models. I have learned a lot about how to set metrics, especially a long list of ways to do it poorly. This post will ideally help you to learn from some of my mistakes. I assume you know how to build a model in general; I’ll save Startup Modelling 101 for a future post.

If you lack predictability, focus on that first.

Building a good business model usually requires returning to first principles. This lets you strip back any embedded assumptions that float around in your team's heads and put them down in a spreadsheet. Sometimes, during this process, you’ll reach a point where the model looks good, but you have an eerie feeling that you have no idea if any of these figures are too ambitious or far too easy. 

At this point, you should compare what you’ve modelled to performance over the last one or two quarters. If you can’t, or if your measured values bounce around almost randomly, you have a predictability problem. It’s almost always a good idea to put someone to work trying to solve that as a priority.

If you want to build a viable model, you’re usually going to design linear or compounding growth on numbers from previous months. If you want that to succeed, you need confidence that, at a baseline, you can hit certain numbers.

This applies to any area of the business. Can you hit your sales targets? Can you acquire enough new leads? Can your product stay online as new customers sign up? When the answer to any of those is “I don’t know,” then you’re better off solving that than building a model you’ll probably miss in week one.

Importantly, here, predictably, it doesn’t always mean every month is bigger than the one before it. Almost every business has some seasonality. In the most extreme case, if you’re selling Christmas decorations, you’ll have one or two big months and probably not much activity. If you can reliably estimate and measure that, then you have predictably, even if your prediction is lumpy.

Do not try to model your business longer than necessary.

Once you have predictability sorted, you can start building out your model. It can be very tempting to predict years into the future and show your business achieving hockey stick growth. This is a waste of time unless you have an extremely predictable business. In startups, your business will rarely be that predictable. You might need a hockey stick chart for a pitch deck, but that can be generated with back-of-the-envelope figures. If you want a useful model that is not a time sink for the spreadsheet lovers in your business, keep it time-bound.

I’ve found in practice that modelling beyond a year becomes inaccurate, so if you have no other guidelines for length, start there. Another practice that is worth introducing to your model is a reconciliation process. Once you have a few periods of actual data, you can compare what you predicted to actual outcomes. With that data, you can adjust the remainder of your model. If you missed your sales targets due to a bad quarter, you can increase the growth targets in future months to catch up. If your growth estimates were too aggressive, you can re-calibrate with some actual data to back up your new assumptions.

Many people see changing a model as a failure of the model. While it certainly can be, if you set the expectation up front that you will re-calibrate as you progress, you save yourself from looking like you are constantly post-justifying your performance. I will say, however, that the older your business is, the less your model should need adjusting, save for significant macroeconomic anomalies in your industry. Your goal is to build a predictable business, so how much your model deviates from reality should decrease with time.

Models often lean towards lagging measures; day-to-day should include leading measures.

If you’re building a model, you should focus on modelling the actual needle-moving aspects of your business, at least initially. Start from measures like revenue and profit and slowly work backwards through the things that drive those metrics. However, once your model is complete, you will need metrics people feel they can impact daily. Metrics like gross profit or cash burn can feel very distant from people running the day-to-day processes in your business.

In the wider world, these are called leading and lagging metrics. Lagging metrics are the most concrete but only appear when all is said and done. The lag the activity in the business. Examples of lagging metrics are revenue, closed deals and profit. Leading metrics, on the other hand, are measures that can help guide critical day-to-day activities. They make for good ‘TV metrics’ because people can shift them up and down with action. The trade-off of leading metrics, however, is they aren’t always one-to-one correlated with the financial metrics of the business. For example, the number of people who hit your marketing website is an easy-to-track metric that might increase every few minutes. Ultimately, it's only an early signal of a potential new user. The user must still complete a sign-up and payment before generating cash for the business.

The right metrics can empower people, but the wrong ones can diminish them.

Once you finish your model, you will likely have a collection of day-to-day metrics to divide between teams. If you pick them well, these metrics can be hugely empowering. They can give a team a sense of purpose or a mission, something clear to rally behind. The wrong metrics can have the exact opposite effect. When you assign people metrics that they feel they cannot change, they lose the feeling of autonomy, which can significantly impact morale.

That’s not to say you should pick metrics based on people’s egos or feelings. At the end of the day, the metrics should reinforce the model, which should help the business succeed. You should be careful, though, to ensure the people who are on the line for these numbers feel they have the power to change them. That often means the process from model to metrics is best done collaboratively. You want a mix of people involved to strike that ideal balance between the business and the people.

On that note, be extremely careful when linking metrics to individual or team performance reviews. I am not saying don’t do it, but if your metrics are supposed to be stretch goals, make sure teams know that. Startups often have to take big bets to get where they’re going. You want people to know they’re taking the bet with you, and the worst thing you can do is inhibit their risk appetite by pinning metrics too close to their ego and livelihood.

When picking long-term metrics for a product or technology team, be wary of how your model dictates your roadmap.

If you aren’t in a technology-focused business, this one will likely be less relevant. Chances are, if you’re reading this, you probably are. One of the best things you can do as a technology startup is find a way to stay agile (lowercase A agile). In the early days of building your product, you need to move and adapt to what your customers actually need. This can often differ significantly from what you think they need.

When you build out a model and translate that into metrics or OKRs, be careful not to have those metrics overly constrain your product roadmap. You will want to pick metrics that signify that a problem has been solved, not ones that signal how it was solved. 

Putting this slightly less abstractly, a good example of a metric would be the percentage of users who sign up after their free trial of your product. You might have some hypotheses about what you could build to change that metric, but you have the luxury of adapting as you learn more about what works. 

On the other hand, if the metric were some measure of the completeness of a specific feature, like “add an AI chatbot to our product to increase conversion,” you run the very real risk of being tied to completing something that will have little to no real impact on the metric you really care about, conversion rate.

Remember that models and metrics are a window; you still need to do the hard work.

Metrics and models are helpful tools. They can provide great clarity and focus within a team or business. However, you still need to do the work. Setting targets on cold calls is great, but someone still needs to pick up the phone. It is very easy to get overly invested in modelling and spend time tweaking a spreadsheet, thinking it is some form of divine scripture. It’s not. Any model you put together is just a window into your business. No spreadsheet, no matter how complex, will do the hard work of solving your problems.

Do not make complicated combined metrics.

This one is a little bit of an oddball, hence why it's at the end. I have regularly been tempted by the prospect of formulating a very complicated combination of metrics so that the team or company has a single number for its north star. Usually, this is a weighted sum of several other numbers forming one overall key measure. This is appealing in theory but has never worked for me or anyone I’ve talked to who has tried to do the same.

What usually ends up happening is that the magical combo metric becomes disconnected from both the underlying model and the day-to-day. Yes, if the metric goes up, it's a sign things are going well, but it becomes very hard to understand why on both sides. In the modelling sense, it can blind you to holes in your plan when the overall number moves up, but critical pieces are falling over. In the day-to-day setting, knowing how to shift the number becomes impossible.

Instead, I recommend picking one to three directly measurable figures to use as a north star. You can change them every quarter if you want. The clarity you will get from having figures that people understand and can contribute to will far outweigh the fact that the three you choose may not cover every aspect of the things you want to change.