Maple Spark Labs
Friction First

Method and templates

How we find where AI earns its place, test it with the people who do the work, and take it to production, with templates you can use inside your organization.

Modernization is a loop, not a project.

Operations don't get modernized in one program. They get modernized one workflow at a time, in turns that each leave the next one cheaper. AI makes each turn faster and reaches steps that rules never could. A Sprint is one turn of this loop on one workflow.

  1. 01UnderstandWatch the work where it happens. Measure where the time goes.
  2. 02SimplifyRemove steps, hand-offs and re-keying before any tool is chosen.
  3. 03DigitizeCapture the work once, in a form the system can use.
  4. 04AutomateRules where rules work. AI where the input varies too much for rules. A person approves what matters.
  5. 05MeasureSame numbers before and after. Without them, the result is an opinion.
  6. 06RepeatThe next workflow, or the next step of this one, with the baseline the last turn left behind.

The principles

  1. 01Start with the friction, not the tool.

    Map where people lose time before naming a technology.

  2. 02Modernize the system around the worker.

    The person doing the work stays in the design. The tool changes to fit the job, not the other way round, and the fastest person in the room is usually right about where the time goes.

  3. 03Production is the finish line.

    A pilot that never ships is a cost, not a result. Exit criteria are agreed before it starts.

  4. 04Design the controls with the product.

    Review, logging, refusal paths and cost limits are features in the first version.

  5. 05Measure before and after.

    Capture the baseline first. Without it, the result is an opinion.

  6. 06A named person owns each decision.

    AI can draft, check and summarize. It doesn't get to decide.

Common patterns worth testing

These are common places where AI can help in small and mid-sized operations, and where a Sprint looks first. Each is a hypothesis to test with your people, alongside the honest possibility that a form or a process change does the job.

Documents into the system of record

Invoices, purchase orders, claims and forms extracted into accounting or ERP, with a person reviewing the exceptions.

The most repeatable win, and the easiest to measure: minutes per document, before and after.

Orders and requests by email, phone and paper

Captured once, checked by a person, and in the system of record without re-keying.

AI reads the messy inputs; a form and a process change often carry the rest. The worked example shows a Sprint where AI earned its place on the step rules couldn't handle.

Faster quotes and responses

The request that waits days for a price, a schedule or an answer, turned around while the customer is still deciding.

Usually a mix of pulling the right information together and drafting for a person to approve. Measured in turnaround time and won work, not just hours saved.

Routine intake by voice or chat

The routine majority of calls, requests and questions handled automatically, with clear rules for what goes to a person.

Glue between the tools you already have

One repetitive process joined up across email, your CRM and your accounting software.

Mostly rules, a little AI, and something your team can own.

Drafting with approval

Replies, summaries and documents drafted for a person to check.

Where most businesses already are, and rarely worth a Sprint on its own.

Search over your own procedures

Answers from your SOPs and documents.

Real demand, but the failures are quiet, so this one needs the evaluation set most.

And sometimes: a form, a checklist, or a change in who does what. The Map says so when it's true.

Want this run with you, all the way to production?

See the Sprint