Five decisions and what came of them.

Making an AI feature defensible in a regulated product

Company Sage ships AI into legal business formation, where a confident wrong answer is a compliance problem rather than a rough user experience. I defined the AI governance standards those features were held to: prompt design, hallucination prevention, and response validation.

More

I am claiming a discipline here, not a number. Governance work shows up as incidents that never happened, which is hard to put on a page and easy to overstate.

Buying the partner platform instead of building it

The partner portal had reached production without product vetting. Partners were live on it with no pipeline visibility, no clear view of what they were being paid, and bugs that were never going to be fixed on the base platform.

I made the call to implement Salesforce PRM out of the box rather than build a bespoke partner platform, and moved the partner system of record off the business platform onto the PRM. The platform kept three jobs: cart validation, commission calculation, and Stripe payout processing. Everything else about the partner relationship moved.

Before After

How it was delivered

Partners migrated with their referral history, payouts, and status visible to them for the first time. In the three weeks after release I triaged and cleared more than twenty P1 to P3 defects through to deployment.

Taking over a partner program that had never been measured

I picked the partner program up a month into the role, at close to zero, and owned it through twelve months of compounding growth.

• <1% → 12%Partner conversion, in 60 days
• 3 days → <1Partner onboarding time
• +35%Partner satisfaction

Those came off specific work: rebuilding the referral and integration workflows across the partner funnel, redesigning partner onboarding and the systems underneath it, and running a touchpoint program covering partner onboarding and communications. I owned the program across the period it grew, and partner recruitment, sales effort, and timing moved it too.

The harder problem was that nobody could say what the program had actually done. Referral counts, real partner revenue, and owed-but-unpaid commissions were all unknown, so every argument about the program was being made on inference. I built a reconciliation framework joining converted paid intake carts and their payment identifiers against partner attribution records and discount codes. It surfaced orders that had been paid but never recognized for payout, and a backlog of owed commissions that had been invisible until then.

What I did with it

What I’d do differentlyGet the data first. I was making the case for this program on qualitative feedback and inference while its own numbers sat unreconciled. The audit should have been the first thing I did, not the last.

Ticket creation that happens in the conversation

A problem gets described in a channel, in detail, by the person who hit it. Then it waits. A PO, BA, or PM has to read back through that conversation, pull out what matters, and retype it into Jira. A ticket raised late in the day might not exist until the next one.

I built Slack workflows and integrations that create the ticket from the conversation that reported it, and let people update it without leaving the channel. Release communications fire on the same rails: when epics and features close in the development lifecycle, the announcement goes out automatically instead of waiting on someone to write it.

• <5 minReport to triage
• 15 minTriage to assignment
• 96%P1s inside agreed SLA
• 24 hrsRelease comms delay removed

Fifteen or so workflows and at least ten integrations, across product development, IT services and support, and interdepartmental communication. The release automation took roughly eight hours of work out of each cycle on top of the delay it removed. These were standing items in QBRs and in customer advisory sessions, which is where the numbers above come from. The SLA figure is from Company Sage; the workflows and integrations span both.

It has been running over a year and still is. That is the part I would point at: it outlived my involvement in it.

Standing up Slack and SSO for a 100-person company

Through Prime Solutions Group we install, configure, and administer Slack for Vox Technologies, and integrate it with the client’s authentication and SSO systems.

• 100+Employees on the workspace
• 9Systems integrated
• 3–5 moRollout

SSO is the part that changes how the day feels. Nine connected systems without it means nine sign-ins and a stall every time someone crosses a boundary mid-task; with it, the boundary stops being visible. It also collapses new-hire setup into one step rather than a checklist someone works through on a first morning.

We have not instrumented the time saved, so there is no figure here for it. The rollout scale is measured; the productivity claim is not.