Winnipeg, economic diversity and risks from AI

I run a company that builds software to read documents and fill out forms. When I look at Winnipeg, I see a lot of potential customers. I also think the city should be asking some fairly uncomfortable questions about what companies like mine are selling.

Winnipeg has spent decades building an economy that does not depend on any one industry. Insurance, agriculture, transportation, aerospace, manufacturing, government, health care. That mix has served it well. A bad year for one employer does not necessarily become a bad year for the whole city.

But there is another way to look at that economy: what people actually do when they arrive at work.

Great-West Lifeco, Wawanesa, IGM Financial and Manitoba Public Insurance are all headquartered in Winnipeg. There are provincial departments, Shared Health and businesses handling the freight moving through CentrePort. Their purposes are different. A surprising amount of their paperwork follows the same routine.

A document arrives, often in a format nobody would have chosen. Someone reads it, finds the information that matters, checks it against the relevant rules and enters the result into another system. If something looks wrong, they investigate or send it to someone with more experience.

The document might be an insurance claim, a pension application, a customs declaration or a building permit. Each requires knowledge of its own field. Still, enough of the underlying work is similar that software developed for one industry can often be adapted to another.

That is where Winnipeg’s familiar argument about diversification becomes less reassuring. Several sectors could start reducing the same kinds of office work at roughly the same time.

The usual response is to talk about retraining. We should. But I would like to hear more about the decisions employers make before anyone needs retraining, when they are choosing a supplier and negotiating a contract.

Buying an existing product is often the sensible choice. It is available, the price is clear and somebody else is responsible for maintaining it. I would not expect an insurer to help develop every piece of software it uses.

A system that takes over a substantial part of its claims process deserves a different conversation.

The buyer could help develop that system, involving employees who understand the work and negotiating rights to use, modify or commercialize what gets built. That takes more effort than purchasing licences. It also gives the organization a chance to retain expertise and share in the value it is helping create.

Co-development can still eliminate jobs. Local ownership does not settle the employment question. But the terms of a purchase can influence whether Manitoba retains people, technical capability and future revenue as the work changes.

MPI and other Crown corporations, along with public bodies such as Shared Health, should be examining those terms. For major AI projects, I would want procurement teams to explain what capability will remain in Manitoba, who will maintain the system and what rights the buyer will have after paying to build it. The answer will vary by project. At least ask before signing.

Private employers have a responsibility here too. They should publish their best estimates of how these purchases will affect staffing, including whether they expect reductions through layoffs, attrition or slower hiring. Nobody can provide a perfect forecast. They can explain their assumptions and update them.

That would give colleges, workers and the province something useful to plan around. Waiting for a layoff announcement leaves everyone else trying to catch up.

Red River College Polytechnic and the universities should also look closely at the work of reviewing automated decisions. Someone has to investigate the claim that software cannot resolve, recognize when a rule has been applied incorrectly and take responsibility for the answer.

An experienced claims worker may already have much of the judgment that requires. Training should build on it, adding the ability to assess a system’s output, document an override and recognize when the technology is making the same mistake repeatedly. I would like to see a clear route into that work for people whose current jobs are changing.

My interest in this is hardly detached. Caseway co-develops AI products with enterprises and shares the revenue. I am arguing for an approach from which my company could benefit, and readers should know that.

Winnipeg nevertheless has a choice worth taking seriously. Its employers bring customers, institutional knowledge and people who understand complicated work. Those are valuable contributions to an AI product. They should count for something in the deal.

The time to negotiate that value is while suppliers are competing for the contract. It gets considerably harder after the software is installed.

Al Vigier is the CEO of Caseway, an AI company that co-creates products with large enterprises and shares the revenue with them. He spent seven years in the Canadian Army.