It always saddens me slightly when people get excited about the efficiency implications of artificial intelligence.
"Amazing! We can fire our [insert team here] team and the AI will do the job at 1/10 of the cost!"
We have seen recently that this kind of approach to implementing AI hasn't always worked out as planned (see Klarna, McDonald's, and CNET).
Every business has space for efficiency, sure. But a more interesting question in our team has always been: how can we use AI to grow, rather than to shrink?
Abstractly, AI-powered growth should have some pretty special qualities. AIs work tirelessly and are extremely intelligent, only increasing in IQ, and are only too happy to do whatever you want them to do, within reason, of course. If I offered you a new sales or marketing hire with those qualities at a salary of less than 1/10 of your company average, you'd probably hand them an offer letter on the spot.
Concretely, it's hard to get AIs to contribute to growth in a meaningful way, which is the problem we're currently grappling with. Here are some general points on how we're thinking about it.
Processes make businesses and growth
Not everyone will agree with this statement, but we're choosing to start from a place which assumes that most businesses of a certain size are a collection of processes which repeat ad infinitum and eventually, and hopefully, generate revenue.
Most processes don't contribute growth
That said, most organizational processes do not individually contribute meaningfully to growth over the short term. Most just keep the lights on. Invoicing, payroll, accounting, hiring/firing, supplier management and logistics are all required to serve customers and generate revenue, but they don't produce huge leaps in revenue or profit, or whatever your proxy for growth is. Traditional BPM theory would term process changes here as "efficiency optimization", which, as I said at the outset, is not what we're aiming for. You can use AI to improve 50 low-leverage processes in your business, and you still won't grow.
Instead, we want to laser focus the power of AI on driving results in the shape of outsized growth for our organization. So we can forget the idea that we want to reduce deviation, reduce mistakes, increase uniformity or do any other of the things which my former employer Accenture would have us do. Instead, we want to focus on the things which really move the needle.
Business nerds will spot the relation to Goldratt's Theory of Constraints here. If your objective is growth, you need to identify the throughputs which contribute to growth, and what is constraining them. If you try to remove constraints on anything else in the business, you're essentially wasting your time and effort.
Growth processes live in the revenue equation
The above two points suggest growth requires a sustained focus on the isolated processes which do, in fact, contribute growth. Discovering these and directing effort accordingly is key.
A revenue equation, or again, a profit equation, which may look very different, will vary from business to business. This is part of the challenge. But in our business, digital retail, there is a pretty clear value equation:
Revenue = qualified traffic x conversion rate x average order value x repeat purchase x availability
Each of the above has a set of processes controlling it. These may be termed our growth processes. Other types of businesses will likely have a different equation. But the key here is to understand your revenue equation, because that tells you which levers you can pull.
Not all growth processes have equal headroom
With the revenue equation above, any output can change the revenue number. You might be in the lucky position of being able to produce a 20% jump in all of them. But normally, not all component processes will be able to move numbers enough to significantly drive growth fast enough. Some have very limited upsides, and some, especially where there are power law distributions, for instance in marketing, have nearly exponential upsides.
One example is the fact that availability, taken as "stock availability", is a binary. Items are either available, or out of stock. Assuming your best-sellers are available all of the time, optimizing here is unlikely to get us to double our growth, even if we manage to reduce supply chain costs. Note: It's a fair observation that availability may also be taken to mean product range, which we're ignoring here as we can't create new products as quickly as we'd like.
Conversion rate, while not binary, also has an upside ceiling, because "qualified traffic" will never be "guaranteed-to-buy traffic". And so doubling your conversion rate from 2% to 4% will likely have a huge impact on your business. But the 4%-5% will be much harder, and eventually there's a ceiling where you'll face diminishing returns. Whether you're close to it will determine how valuable a tweak in the right direction is ultimately going to be.
Therefore the best growth processes to tweak, if you want to improve them quickly, will always likely have some characteristics which make them unique. For us, these are:
- It is close to money, so the observable effect on growth is fast and easy to measure.
- It repeats often, providing ample optimization opportunities and a larger sample size from which to determine results.
- It is highly variable and has high ceilings, so larger swings can produce larger upside outcomes. See the next section for this.
- It produces data which can be used to improve.
Can you process-engineer something which requires variability?
Creative outputs normally benefit from variance rather than standardization. If you try to standardize interior design, all rooms look the same. Therefore, processes which require creativity and originality must be wrappers over those things. They need to provide space for a creative process to exist without trying to control it.
Ideally they would also "feed" the process. Process leverage here is likely in providing the inputs and ensuring diversity, rather than trying to actually do the creating. By way of example, a room full of creatives benefits from customer research, which can be standardized as part of a process. They may also benefit from customer reviews, which can be collected in a process-driven way. And in a campaign retrospective, they may benefit from analytics, collected through a process. All of these can enhance creativity, while not trying to affect it.
Useful variability requires a ceiling
Finally, high variance must be coupled with a high ceiling: the ample growth headroom mentioned in the revenue equation above. What do we mean here? A button color may have a very high variance, pink or blue, take your pick. And the observed upswing might be huge. Blue could convert 18% better. And that can be meaningful. But only at massive scale.
In reality, for most businesses, there is a vast amount of noise which quickly reverses the 18% uptick the blue button provides us with. Think about it this way. That blue button arrives at the end of a process in which the consumer has seen an ad multiple times, clicked it, landed on the page, read the copy, scrolled the pictures, read the reviews, discussed with their husband, quickly checked Reddit and then, finally, decided to buy. There is an awful lot happening upstream of that 18%, and a lot downstream too.
Therefore, spending time and money to optimize the button may not be the best growth experiment. It's more likely that opening up demand further up the funnel will have a stronger net effect on the overall growth.
What this means
We now have a clear path thanks to our theory. We need to find processes that are close to the customer, which happen frequently, are highly variable and can be empirically measured. And they need to sit in:
- Traffic
- Conversion rate
- Average order value
- Repeat purchasing
- Availability
The final question is: how can we use AI's strengths to improve them? We'll cover that in our next post.