The challenge
The company's private equity owner engaged Midici to help bring AI into the rest of the organization. Engineering and product were already using AI heavily: every engineer had a license, and product and marketing were producing careful AI-assisted work. The other functions had not yet followed, and the CEO expected growth to strain implementation and support first. The company wanted to know where AI would expand capacity fastest outside engineering and product, and how to introduce it without unsettling teams that had recently been through change.
Our approach
We interviewed every member of the executive team and assessed every function where AI could play a part, using the same five-dimension rubric, based on observed behavior, that we apply in every engagement. At the CEO's request, we spoke with individual contributors only once internal messaging was ready, so discovery stayed at the leadership level and work with individual contributors moved into delivery. A preliminary readout midway gave the CEO the findings before the final report. Where the evidence could not support a score, the report marked it N/A.
What we found
Data and systems readiness was the common constraint
Data and systems readiness scored 2 out of 5 in every function where it could be scored. Implementation specialists spent about three hours preparing for each customer meeting, combining information from five systems by hand. Support could need six systems to answer a routine question, and there was no knowledge base. Unusual cases went to a Slack channel, where developers and product managers answered as they could. When we asked how much knowledge lived only in people's heads, the COO said:
“Lots … it's risky.”
Engineering adoption varied widely between engineers
Every engineer had a license and every engineer used AI to some degree, but one had generated roughly 100,000 lines of code with AI while others were near 2,000. There was no usage tracking, so the effect on productivity could not be measured. An AI-assisted fix by a junior engineer had corrupted data for every user and taken six to eight weeks to resolve, which explained the caution among much of the team.
The COO saw the value during discovery
During the COO's interview, we ran a working AI-assisted meeting-preparation workflow live. She applied it to her implementation team without prompting:
“You just did all your prep with the agent … done.”
Appetite for AI differed between teams
The implementation team asked directly for help. The support team was concerned about being replaced, because its professional identity came from solving difficult questions. We planned the rollout around that difference.
What we recommended
We recommended three flagship programs, each tied to a named business outcome:
- An implementation preparation agent, built on licenses the company already held. The hours it recovers go to proactive customer engagement, which supports the referrals and RFP wins the company grows on.
- A support knowledge program: one support system, a searchable knowledge base, and an internal AI layer, with a person always between the AI and the customer. It is framed around a career path from support into implementation.
- An engineering program to raise the baseline, built on standardized context, a shared skills library, pre-commit guardrails, usage tracking and visible recognition. The CTO's advanced agentic work continues as the company's research and development, while the program supports the rest of the team.
The report also proposed a read-only AI connector to the company's own platform. It would base internal work on how each institution actually uses the product, and could become a differentiator for customers adopting AI themselves.
The report also set out where to hold back. Sales and account management were left out, since the function depended on one leader who was already at capacity. Automated responses to customers were deferred until the team had shown it could supervise them. HR and finance were deferred to a later wave. The one tool purchase we recommended came with a clear condition for cancelling it.
