Agentforce is a business case before it is a build

Most conversations about Agentforce start in the wrong place. They start with what the agent could do, which channels it could sit in, and how quickly a pilot could go live. Those are fair questions. They are also the second set of questions.

The first set is commercial. What is this agent for, who owns the result, and how will we know it has paid for itself? When those answers are missing, the pilot usually works and the programme still stalls. The demonstration impresses, nobody can say what it is worth, and funding for the next stage becomes an argument about belief rather than evidence.

Our view is simple. An agent is an investment in changing how a piece of work gets done. It deserves the same discipline as any other investment.

Why the build goes first, and why that is a problem

Building first is tempting because it is tangible. A working agent in a sandbox feels like progress. A business case feels like paperwork.

The trouble is what a build-first approach leaves undecided. It rarely settles which cost or revenue line the agent is meant to move. It rarely names the executive who will answer for that movement. And it almost never records how the work performs today, so there is nothing to compare against later.

Each gap is cheap to close before the work starts and expensive to close afterwards. A baseline cannot be measured in hindsight. An owner appointed after go-live inherits decisions they did not make.

Three things to settle before design starts

We ask leadership teams to agree three things before any design work begins.

1. The case. One specific piece of work, described in business terms. Not "an agent for service" but "resolving order status and delivery enquiries without a handoff, for the customers who contact us most often".
2. The owner. One named executive who is accountable for the outcome, not for the technology. If the agent handles service enquiries, the owner runs service.
3. The measure. One primary number that the owner already cares about, with a baseline taken now and a target agreed in advance.

If a team cannot agree these three in a single working session, that is useful information. It usually means the use case is too broad, or that the organisation has not yet decided what it wants from agents.

What a good agent business case contains

A good case is short. Four or five pages is enough. It should cover:

• The work today. Volumes, effort, cost, error rates and customer impact, as far as they can be measured. Estimates are acceptable if they are labelled as estimates.
• The change. Which steps the agent takes on, which stay with people, and where the handoffs sit.
• The benefit. What improves, by roughly how much, and on what assumptions. A range is more honest than a single figure.
• The cost. Not only licences and build effort, but data preparation, process redesign, training, oversight and ongoing ownership.
• The risk. What happens when the agent is wrong, who notices, and what it costs to put right.
• The stop conditions. The results that would lead you to pause, narrow the scope or stop altogether.

The last item is the one most often missing. A case with no stop conditions is a commitment, not a case.

If nobody can say what would make you stop, nobody has really decided why you are starting.

Costs that tend to be left out

Agent cases are often optimistic on cost because the visible items are the small ones. The larger costs sit around the agent rather than in it.

Data is the usual surprise. An agent that answers customer questions needs accurate, current, permissioned information. Getting there can be the largest piece of work in the programme. Change is the second. People whose work shifts need new roles, new measures and time to adjust. Oversight is the third. Someone has to review what the agent does, tune it, and answer for it, every week, for as long as it runs.

None of this is a reason to hesitate. It is a reason to count properly, so the case survives contact with reality.

Starting smaller than you think

The strongest first cases are narrow. One process, one team, one measure. Narrow cases produce evidence quickly, and evidence is what earns the next round of investment. They also teach the organisation how to run an agent, which is a capability in its own right.

Breadth can follow. A portfolio of agents built on proven cases is far easier to fund and govern than a broad platform ambition that has yet to show a result.

When we advise on Agentforce, this is where we begin: with the case, the owner and the measure, agreed before anyone opens a design tool.