As the team and its project portfolio grew, logging time became a weekly chore that consultants put off. Each week had to be rebuilt from memory and calendars, matched to the right Zoho project and task, and entered by hand. Late and incomplete timesheets meant unbilled client effort, lost revenue and poor visibility of consultant utilisation, and every new engagement made it worse.

An Agentic Discovery Workshop in August mapped where consultants were getting stuck and ranked candidate use cases by business value against effort. Time recording came out on top because the hard part is interpreting the week, which is exactly what an AI agent does well. Success targets were agreed up front, followed by a soft launch in September and full go-live for the 12-person cohort in October.

Timekeeper is an AI agent built on Amazon Bedrock AgentCore that lives inside the tools the team already uses. A consultant tells it what they worked on in a chat message, or asks it to read their calendar. The agent matches the work to the right Zoho projects and tasks, asks about anything unclear and proposes the entries. Once the consultant confirms, it records them in Zoho under their own access.

"This AI project lets us work through a single interface instead of switching between systems. We've gained time and visibility, and we turn our hours into revenue faster. That speed fuels our cash flow and empowers our strategic decision-making."

Each week, staff exported point-of-sale records from Yoco, analysed product movements and built the report in a spreadsheet, taking two to three hours. The report was a single weekly snapshot without period comparisons or brand and category trends, and every new question meant another export. Decisions waited on staff availability, and growth across the store and online channel was making it worse.

LSD Open ran an Agentic Discovery Workshop using its Understand, Analyse, Prioritise and Recommend method to establish the reporting baseline and find where slow insight was hurting purchasing and marketing. The agreed goals were zero manual report preparation and on-demand analysis between reports, at a running cost suited to a boutique retailer. When the POS provider's own integration could not support the required secure connection, the team built a custom read-only sales connector so delivery stayed on track.

The Sales Intelligence Agent runs on Amazon Bedrock AgentCore inside the team's existing Slack workspace. Managers ask in plain language which products are moving fastest, how this month compares with last, or which brands are growing. The agent works out the period and analysis needed, pulls live Yoco sales data and explains what it found, keeping context for follow-up questions. Every week it publishes the full sales report to Slack automatically.

"Sales Intelligence Agent that LSD Open has delivered has saved us hours of weekly reporting and gives us clear insight into what’s selling, helping us make faster, better-informed purchasing and marketing decisions."

Finch builds the software and digital platforms that other organisations use to deliver financial services, from marketplaces to customer onboarding and verification. As the business grew on AWS, understanding cloud spend became a job for both engineering and finance. Explaining a single cost movement took hours of specialist effort, and reviews happened only after a big change forced one. Avoidable spend surfaced late, and the effort grew with every new workload.

LSD Open started working with Finch on 13 August 2026. Two objectives were agreed up front: get from a cost question to a supported explanation within minutes, and replace ad hoc reviews with a weekly cycle. The service was designed to work where Finch's teams already collaborate, with people accountable for every spending and infrastructure decision. Onboarding was completed with the full Finch team on 21 October 2026.

The FinOps Agent works inside Finch's Slack. Anyone authorised in engineering or finance can ask why spend moved and get a clear explanation, drilling down from the overall bill to the individual areas behind the change. Each week it posts a cost comparison and a list of savings opportunities ranked by value.

The business runs a digital learning platform for neurodivergent primary-school learners, built around interactive lessons in the South African curriculum. Even carefully prepared lessons left some learners stuck, and repeating the same explanation rarely helped. Some needed smaller steps, a different analogy or an example tied to something they cared about. Younger learners often knew they were struggling but could not put it into a question. Without help at that moment, progress stalled and frustration grew.

Work started on 5 October 2026 with LSD Open's Understand, Analyse, Prioritise and Recommend discovery, run with the customer's business and technical teams. Together they agreed two outcomes: help at the point of difficulty without leaving the lesson, and explanations personalised to each learner. The AWS build kicked off on 27 October, and a soft launch with the customer's teams refined how the tutor teaches before it went live.

Learners ask for help from inside the lesson they are working on. The tutor already knows the lesson, the learner's progress and their learning preferences, so it starts from where they are. If an explanation does not land, it tries another way: smaller steps, a new example, or a link to the learner's interests.