How to actually get AI adoption over 90% across the organization
How do you actually get AI adoption over 90% across the organization?
At Credal, we’re used by hundreds of businesses looking to drive AI adoption across their Enterprise. We’ve seen how companies succeed, and how companies fail, in making this adoption happen, and we’ve learned what works across industries. I wanted to share a few of those learnings with you:
1. Make someone accountable
Companies, especially enterprises, stand to experience truly extraordinary efficiency and productivity gains from effective AI adoption. The average Credal user alone reports increased satisfaction and engagement at their workplace, and dramatically increased productivity. One fascinating fact we’ve seen is that Credal users, at our customers’ organizations, are 16% more likely to be promoted than the median employee at their organization, and 27% less likely to be impacted by an organizational RIF.
With productivity gains like this to be realized, it’s vital that someone is accountable for realizing those gains. In practice, implementing the right tools, educating employees, discovering key use cases, and securing executive buy-in requires effort. Having an accountable person in IT or on the AI platform team is vital to make sure that these initiatives are driven through to completion in order to actually succeed.
At Credal, we’ve seen various groups take accountability for AI adoption, but there are a few common patterns:
1. IT (which is in charge of employee productivity initiatives in about half of all customers, but usually in customers below 1000 people).
2. AI Platform / ML Platform Team (Engineering) (this is almost always the case at larger, 1000+ person companies)
3. AI Task forces (usually comprised of AI enthusiasts and business representatives from many different parts of the business)
4. None - individual business leads make decentralized decisions about what tools to use (rare, but we do see it)
5. Engineering - within the existing Engineering team, someone has as part of their portfolio general AI adoption.
In practice, from what we’ve seen, any of these groups can be successful as long as it’s clear who is accountable, and they have a clearly defined goal of driving adoption. Even when it is decentralized, as long as it is clear that each business lead is accountable, AI adoption can still be quite fast (although it can be slower where adoption is very decentralized, as tools and data become siloed and less interoperable).
2. Give your people access & license to experiment (with guardrails)!
AI adoption works best both top down and bottom up. Although it's important for business leads to communicate the business’ priorities to the employee rank and file, in practice, if employees aren’t given a license to experiment with use cases, AI adoption will go much slower.
This sounds so obvious, but funnily enough, we encounter many organizations who notionally say they want to “drive AI adoption” and “transform the experience of working at their company”, but in practice, they aren’t willing to let users experiment and instead insist on validating a very small number of highly specific use cases before giving employees access to experiment with tools.
This is, of course, usually well-motivated - they worry about distracting teams with side projects or with tools in pilot mode. Oftentimes, IT leaders are worried about disappointing users given the amount of hype and overpromising there is in the industry.
But we find in practice, that after a few days of fairly basic validation, by far the best way to validate if a tool is actually useful or not is to let users experiment and discover what works. Although not everyone at a company will be excited to play around with new technology, even in non-tech industries, around 10-15% of employees will be early adopters and experimenters. These employees are often the ones who find novel ways to deploy AI into workflows. For more information on the typical AI adoption curve, see our guide.
The key is that AI enthusiasts must be equipped with technology to build tools that everyone else on their team can use. This 10% can then become evangelists and educators for AI as they experiment, learn, teach, and then ultimately share production-ready AI applications, often built with no code at all. They get celebrated for their innovation, their teammates become more productive, and the whole company becomes inspired to find more use cases for the technology. Win-win-win.
The graph below shows the adoption curve of one our first ever enterprise customers. Of course, the product is much better now than when this enterprise first started using Credal, and adoption curves are therefore much steeper now, but it shows a clear relationship for how, as the builder personas increase, that creates a huge flywheel driving more users, and then more builders, and then more users over time!
(Note that the ability to build applications in Credal only became available a few months after we launched, which is why the % building line is shorter than the % using line)
By contrast, when access to AI applications is gated to an extremely small community of people, these viral flywheel effects from learning and building on top of each other is greatly reduced. At the same time, one or two individual testers can often become critical bottlenecks for entire teams’ usage, and holidays or sickness can actually slow down the entire enterprise's ability to adopt and realize the value of AI.
Another dimension that can vary is between buying a single platform that can support users building multiple use cases, vs buying a number of point solutions to solve each individual problem. Both have their pros and cons, but our recommendation is to do both. Buying a platform as your first port of call that provides a good low/no code experience as well as APIs will get the enterprise off to a flying start.
That gives users who want to move fast the flexibility they need to build their own apps easily for all the low hanging fruit. Eventually, you might find certain use cases not well supported by the platform offerings, and so at that point looking at specific point solutions (the most popular ones we’ve seen are Cursor (enterprises are typically more familiar with Github Copilot, but Cursor is much preferred by most of the world’s best software engineers) and ElevenLabs for speech to text, but there are many for different use cases).
3. Create a governance program, and make sure people know about it
Communicate your governance program very clearly. One of the most fascinating things we’ve seen at Credal, is that organizations that have a robust (and sensible/business minded) governance program for how they are planning to adopt AI responsibly (i.e. without compromising data security, product or customer experience quality) actually drive much faster AI adoption that ones that do not.
That might seem counterintuitive, since typically such governance and security initiatives are perceived to slow the spread of new tech, not accelerate it. But in practice, what we’ve found is that users who are not sure what they are allowed to do, will bias towards doing nothing at all, for fear of breaking rules or causing the company some nebulous harm. Getting security right in your AI platform therefore is an accelerant to your company’s AI adoption efforts. For example there are many AI platforms that allow users to connect docs and chat with them. But if those platforms don’t fully support the permissions of the underlying source systems (e.g. Google Agentspace, that supports permissions only if they are fully defined with the Enterprise’s iDP. Any native permissions defined in Confluence, Jira, Sharepoint etc are not supported), then users will be unable to build agents or AI applications that require such documents to function effectively. It also becomes impossible to connect private slack channels - most of an enterprise’s HR or sales data. As a result, a robust security and governance foundation to an AI platform is actually an accelerant to adoption.
At one of our earliest pilot customers, we saw Credal’s adoption jump from 14% of the company to 80% in a few months, all because the company simply announced its AI governance program, clearly outlining what was allowed and what was not allowed. Bear in mind that technically this whole time, the company had purchased Credal and was paying for it - but due to IT hesitating to announce a broader official policy, users did not even know if they were allowed to use Credal, resulting in much lower adoption than clearly the organization had appetite for.
Given this trend, it's not surprising that regulated industries have struggled to drive AI adoption. In one study, nearly 80% of financial services firms see AI as vital, but only 32% had governance programs.
4. Meet users where they are
One thing will never change - users hate having to learn new tools. The more that users are given access to AI tooling inside platforms they already use - like Slack, Microsoft Teams, Salesforce, Zendesk, Retool apps etc, the faster adoption will be. What we’ve found is that while these tools often act as a “gateway” to realizing how useful AI can be, in the end, users will go wherever they need to, to get their work done. But the key is helping them discover the tooling organically, rather than creating initial barriers to discovery by putting walled logins and steep learning curves in front of their first, delightful experience. For Credal, the way we manage this is by making it completely seamless to deploy into external tools like Slack.
5. Pick AI-first partners/vendors
As an AI-first vendor, we are of course a little biased here - so feel free to gloss over this section if you prefer - but we truly believe in the advantages of AI native vendors. Since the technology is going to move fast, you want your partners and solution providers to move fast as well. Legacy enterprises or tech companies that pivoted into AI products have very, very rarely shipped the best, AI native products, that really fully unlock the capabilities of AI for end users. Often these companies look impressive out of the gate - because they have built a suite of products that they can repurpose to look relevant to AI adoption - but in practice, the value of the tools is heavily reduced by the fact that AI is not the central part of how their business got started, and so their teams are stuck maintaining legacy codebases and unable to ship basic features that users want. Without wanting to put too fine a point on it, we’ve heard so many buyers frustrated about Microsoft AI products’ like Copilot, due to its price point and limited value. Equally, companies like Glean and Coda had significant existing businesses, which they then tried to pivot into AI companies. While Glean had a significant installed user base and was able to leverage that into a headstart on go-to-market, its product still lacks the basic features that virtually all AI native companies have, like letting users switch models to use the best model for each use case, executing code to generate and download files as needed.
Credal enables users to pick the model best suited for them, on the day they come out from the provider! This is very typical for AI native providers, whereas legacy providers of AI tooling, like Glean, Microsoft Copilot, Google etc, have significant limitations on which models you can use and how seamlessly you can move between them.
By contrast, in Glean’s documentation, they explain that you have to choose a single AI provider and that they only really support a small number of models within that provider.
6. Teach your employees, and even better: let them teach each other!
AI, especially generative AI, is a technology that is still travelling extremely fast. What didn’t work last month, may suddenly work today. What may have been a complex process that only highly sophisticated users could manage last week, can suddenly become accessible to anyone this week. We see models or APIs that achieve State of The Art (SOTA) for at least one of the three most critical attributes (model intelligence, model speed/cost, model personality) shipped at a rate faster than once a month.
Every new state of the art model unlocks a range of new potential use cases, and can often even create a range of new best practices for how to get the most out of AI. There is simply no way for time-strapped employees to totally keep up with all these changes themselves. Even for L&D or education teams, its unrealistic for them to be able to maintain a curriculum in the face such an onslaught of innovation from the research labs and the AI industry more broadly.
But that doesn’t mean they should give up! Companies that host regular ‘show and tells’, that broadcast the applications that are working, that host hackathons, and prompting contests and actively encourage employees to tinker and celebrate those finding real-world applications and value creating use cases, achieve 3x the adoption rate (# of months to reach 90% of employees using AI every month). Ultimately, a decentralized education system that lets your employees discover new use cases and teach each other drives much more real world value, much faster. L&D teams of course still have a critical role in making sure this education happens, and producing training materials for in-production AI applications, but trying to keep fresh an extremely centralized curriculum without taking advantage of the creativity of your employees makes it difficult to keep pace with the rate of innovation in the industry.
One hackathon hosted by our customer almost single-handedly converted 32% of the invitees into builders and an additional 50% into users of AI.
In summary, driving AI adoption at the enterprise, like any enterprise change management process, requires intentional strategy and effective effort. Start by picking the right people and teams to lead the initiative of getting AI enablement into production at your organization. Once done, start getting tooling into people’s hands as quickly as possible. A secure platform that gives access with guardrails to everyone and enables the most innovative employees to get building will drive the fastest adoption. As you get tools into people’s hands, provide clear guidance on what is and isn’t permitted, and ideally build those guardrails directly into your AI platform and point solutions (something that Credal.ai supports natively)! Then as folks start to discover valuable use cases and workflows with AI, encourage them to experiment through hackathons, share what’s working through lunch-and-learns, and get people’s creative juices flowing. Before you know it - you’ll have 90% of your organization using AI and finally feel like you are realizing the phenomenal productivity gains therein.
Want to learn more?
Have questions on the above or want to know how these apply more specifically to your needs? We’re happy to chat. Drop us a line at sales@credal.ai!
