Improving M&A with machine learning

I’ve recently been asked by a customer how they can apply machine learning (ML/AI) to their M&A activities and I thought it might be prudent to share my thoughts. Comments and contributions are all very welcome.

Machine learning is good at spotting complex patterns. The obvious big prize would be a tool that suggests which company you should invest in (or not). But that is easier said than done. And if anyone creates that, they would do better by just using their own tool and make a fortune, without sharing/selling it. 

If you had success/fail data of 100 000 acquisitions with a limited set of clearly measurable input parameters (not just for weather, but also for economy, management style, product quality, people characteristics and culture...) - then you might stand a chance of building the model. But otherwise it’s better to break down the problem into smaller chunks.

The specific customer who asked me the question, acquired a couple of businesses per year, so they don’t have nearly enough data to help them here.

One dimension of the problem they flagged was in building the pipeline of potential acquisition targets. In their case, they also had no visibility of the targets they did not know about, so using ML (from their perspective) to find that would also be hard. (I understand that some accounting or auditing firms do have sufficient visibility of a large number of businesses and have indeed built models to help predict M&A target values.)

Instead, I think ML can play a strong role in helping bring systems, businesses and data together. It starts with just good old data… bringing that together.

Here is how I would approach integrating the systems from many acquired businesses into my core business, with minimal disruption:

If acquisition n has no (or poor) internal systems, I would rip it out and provide them with your core systems and tools where they lacked. (If your core systems are poor too, then please don’t migrate more to it! Pick a new one for the new acquisitions - even if it’s just to test a new potential system.)

However, more often than not rip & replace is not going to be an option: If the existing systems were crucial to their operation, I’d start by taking a data feed extract from these disconnected systems and write it into a central shared data store. This is non-disruptive. All existing systems should still work as before, it just happens to expose the data in a fresh, modern system.

Over time, this would build a central place with consistent product, production, sales and people data, which will become crucial in informed decision making. Once you have this, ML would really come into its own right, helping with cross selling product recommendations, demand forecasting and price setting.

In one customer, I’ve seen the managing director hire one smart developer who did this. She pulled mainframe data into a modern database. No-one in the business had to know or change tools, yet word spread from the finance director right through to shop floor workers that there was a much simpler, better and easier way to get stock levels or hourly sales reports or forecasts by just looking at this new app.

Let’s take one step back: I know that in practice such imported data is rarely clean data. This is an interesting opportunity to apply ML at the integration stage, as it can spot anomalies in data and help with merging data cleanly.

But once a new system with fresh, relevant data is in place, it opens the doors to all sorts of new opportunities.

Initially, it’s a read-only version of the data. Write backs can happen by first writing back to the original systems (manually or in an automated way), but many find that by this stage, it becomes easier to build a table in the new system to capture fresh updates. Then eventually the old back-end system becomes redundant in areas until it can be fully retired.

As with all change: People love it only when they choose it. We love a new car/home/phone when we choose it. But when we feel like it’s forced on us, then we’ll happily fight hard to resist it. (Just think about some of the political and health stories in today’s news!)

I’ve seen some very innovative ways to help with this technology transition for the workforce. In one customer, I’ve seen a massive roll-out of new disruptive tools replacing old industry tools under the banner of “it’s just a pilot - let’s experiment and see what’s good and bad… leave feedback here”. Positioning as a pilot meant people adopted it quickly without feeling threatened.

Another story: In London, taxi drivers are renowned for having “The Knowledge”, knowing all streets in London, not needing GPS or satellite navigation. When one big taxi company introduced navigation systems years back (to track and ultimately plan more efficient dispatch which was impossible in the old radio call systems), the drivers objected strongly. The key to introducing the new system was simple: Turn it into a competition, gamify things. “Can you beat the sat nav?” Overnight all the drivers wanted to show the technology who’s boss.

While you are bringing data together from different acquisitions, there are many (not M&A specific) ML opportunities in most businesses that you could employ immediately, without needing full integration . One instance that I see coming up frequently in our manufacturing customers is using computer vision to do visual inspection to ensure things work as they should. One customer uses an electronic eye to check that frozen pizzas on the conveyor belt look yummy, another counts footfall in town centres, another checks Covid-19 compliance in their warehouse, or detects fights in trains, or that the package looks right before shipping. And it’s really simple to build a model like that.

And a last story of data migration: A few years ago, when I was a product manager at Microsoft, we moved from one version of Exchange Server to another. The problem was the migration of our inboxes, full of big emails with PPT files attached. The migration was outsourced to a consultancy company and they sent us notification of the downtime and migration weekend when it all had to happen. Here’s the smart thing: They offered rewards for smaller inboxes and a bottle of champagne for completely empty inboxes. Come Monday morning, nearly every desk had a bottle of bubbly on it and the migration work was made exceptionally easy. 

I encourage you to think about your M&A activities and where you can apply artificial intelligence. Start with the data you have.

And don’t forget to also apply real human intelligence.