BH Technology Book a consultation

ROI for AI

AI gives you hours back. Where you put them decides what you earn.

Hypothetical, illustrative example. Assumptions are shown in the revenue section.

  1. Hours back100 hfreed each month
  2. Useful activity40 hput into new proposals
  3. Sales result5 clients wonfrom 20 extra proposals
  4. Possible profit$15,000gross profit only if all 5 are served, before AI costs

Step 1 · Capacity

One person moves faster. A department frees more hours.

Measure it

  1. What changed?
  2. Was the time used?
  3. What limited the upside?

One person · typical setup example

Overview

Example values for one person who prepares standard sales proposals on their own accounts. A person approves every proposal before it goes out.

Hypothetical example · same weekly workload

  • 30 hToday
  • 20 hWith the setup
  • 10 hBack each week

One person completes their usual batch of work in fewer hours, with review and upkeep included.

How one proposal moves

  1. Deal in HubSpotProposal stage reached
  2. Approved sourceTemplate, pricing and examples in Drive
  3. Agent draftsGrok Bot drafts and organizes recurring steps
  4. Person checksClaude compares the draft with approved examples
  5. Person approvesOwner signs off and sends
  6. RecordedNote and file link saved in HubSpot

Who does what each day

  • Grok BotReads the approved queue or an exported deal and drafts proposals and recurring steps
  • ClaudeWrites longer pieces and reviews drafts against approved examples the owner provides
  • CodexBuilds and tests small scripts and reports in VS Code or a repo
  • HubSpotHolds deals, tasks and customer status
  • DriveHolds approved docs and working files

One deal, start to finish

  1. A deal in HubSpot reaches the proposal stage. The owner exports the deal details or adds them to the approved queue.
  2. Grok Bot reads that entry plus the approved template and pricing from Drive, then drafts the proposal and a checklist of recurring steps.
  3. The owner hands Claude the draft and approved examples as files. Claude flags gaps and differences. The owner checks facts, scope, pricing and any exceptions.
  4. The owner approves and sends the proposal. The note, file link and next task go back into HubSpot.
  5. When the routine needs a fix, the owner uses Codex in VS Code to update the script and tests it before the next run.

The owner stays in charge. They hold every account, review each draft and approve what goes out. Tools hand work over by file, link or repo commit.

Entry $30 Grok Premium + $20 ChatGPT Plus + $20 Claude Pro = $70 a month. Heavier tiers $300. New Workspace adds $16.80. CRM and extra usage are separate.

USD before tax. Sources xAI, ChatGPT, Claude.

Go deeper

Department of 10 · typical setup example

Overview

Example values for a department of 10 after rollout, preparing standard sales proposals on company accounts. Three people pilot it first. An assigned owner approves every proposal.

Hypothetical example · same weekly workload

  • 200 staff-hoursToday
  • 160 staff-hoursWith the setup
  • 40 hBack each week across the team

The full department after rollout completes its usual batch of work in fewer hours, with review and upkeep included.

How one proposal moves

  1. Deal in HubSpotStage and owner set
  2. Shared approved contextTemplates, pricing and examples from Drive
  3. Postgres queueProposal tracked by ID with an assigned owner
  4. Agent preparesGrok Bot drafts the proposal and evidence pack
  5. ChecksClaude review against approved examples
  6. Person approvesAssigned owner or manager signs off
  7. RecordedHubSpot status and run log updated

The stack

Records
HubSpot is the source of record, Supabase Postgres tracks the queue, Google Workspace holds approved docs
Agents
Grok Bot Teams seats on company accounts
Builders
Codex through ChatGPT Business, in VS Code and a repo
Reviewers
Claude Team for results and playbooks

One deal, start to finish

  1. A deal in HubSpot reaches the proposal stage and a work item enters the queue with an assigned owner.
  2. The queue service gathers the deal record, the shared approved template, pricing and examples.
  3. Grok Bot prepares a proposal draft and an evidence pack that lists the sources used.
  4. The owner hands the draft and evidence to Claude for a check against approved examples, then checks facts, scope, pricing and exceptions. A manager handles anything unusual.
  5. The owner approves and sends. The approval is logged and the status goes back to HubSpot.

People and checks

  • Team trainingThree pilot operators learn the routine first. All ten join only after the pilot shows evidence.
  • Quality checksEvery draft gets a human check before it goes out. Managers approve templates and exceptions.

The service delivers the working setup. BH Technology maps the workflow, cleans the source data, sets permissions per identity, builds the queue service, tests it in the pilot and trains the team. Each part is scoped and quoted separately.

Example monthly subtotal for licenses and the database is $545 (10 Grok Bot seats, 3 Codex through ChatGPT Business, 3 Claude Team, Supabase Pro). With new Google Workspace it is $713. Usage, CRM, hosting and the setup service are separate costs.

USD before tax. Sources xAI Teams, ChatGPT, Supabase.

Go deeper

Measure it

What changed?

How to measure hours freed

  • Time one defined task from start to finished and approved, including review and corrections.
  • Log enough real examples, before and after, over the same kind of period.
  • Multiply the minutes saved by the number of comparable tasks completed.

Illustrative example · one weekly client report

Before90 min

After60 min

DraftReviewFixes

(90 − 60 min) × 20 reports a week = 600 min = 10 h released

Hours released are the starting point. They gain value once the team applies them to useful work.

For the setup examples · time one real task from start to approved finish, with review, fixes and upkeep included. Compare it with how long the same task took before.

Step 2 · Leaders first

Equip the department head first

Start with the head of the department where improvement matters most. Their own tools and workflow improve first. Better visibility and decisions then shape the whole team’s work, and the team learns the workflow firsthand from the head.

Select any part of the map to read more.

Start here Equip the department head first Their own tools and workflow improve Read more
Information See the work sooner Owner, status and next step in view Read more Action Decide and coach Priorities and coaching from the record Read more
Whole team Roll out what works The head teaches a workflow they’ve already used Read more

Equip the department head first

One person can often see a larger percentage gain sooner. It takes less time and infrastructure to set up their tools, files and routines.

Start with a department head in the area where improvement will matter most. A sales, operations or billing leader can use AI in their own work, see team activity sooner and give faster feedback. That gives one person’s setup a wider effect and prepares the leader to choose tools, train the team and hire where needed.

The head’s own rate of improvement stays with the head. What reaches the team is the quality of the priorities, reviews and coaching that come from it.

How one person’s gain reaches the team

Back to section

How one person’s gain reaches the team

Two scopes, two percentages

In the illustrative comparison, a 50% gain for one person adds half that person’s usual output. A 25% gain for a department of ten adds a quarter of the department’s usual output. With comparable starting workloads, the team can add more work overall at the lower rate. That wider rollout takes more time, money and shared infrastructure.

Why the head comes first

Applying the head’s personal rate to every employee would be a mistake. A faster individual task changes one person’s output. A leadership decision changes what several people work on, how they do it and which standard they follow. When the head sees the work sooner and decides better, that reach shows up in the team’s priorities.

What the head owns at this stage

The head works from existing records, current priorities and team context. They review activity, adjust priorities or assignments, coach from the record and write down the workflow that worked. The result is a better-informed decision and a workflow they can explain and teach. A personal gain does not guarantee a department result, so the team is measured on its own terms.

Back to Equip the department head first

See the work sooner

Much of what a leader needs already passes through their own workflow in pipeline updates, job status, open requests and billing records. With AI organizing that information, they can see sooner which work is moving, which work has stalled and who may need help.

A useful view helps the head spot what needs attention before the issue becomes harder to correct. That depends on each record being tied to its owner, its status and the next step.

What a useful activity view needs

Back to section

What a useful activity view needs

What each item should show

  • Who owns the work
  • What was completed and when
  • The current stage or status
  • The next action and whether it is documented
  • Any context needed to read the record correctly

Who checks what

AI can organize the records and flag gaps. The department head checks each signal and decides which items go on the review list. The responsible employee confirms that the record is accurate. In a billing team, the view might list each open invoice with its owner, its status and the missing item holding it back.

Reading gaps carefully

A missing activity may mean the work was not done. It may also mean the work happened and was never recorded. A stalled item may reflect a true lack of movement, a customer or internal dependency, or incomplete context. The head confirms which one applies before acting.

Back to See the work sooner

Decide and coach

A signal is where a decision starts. The leader checks the context before deciding. If work appears stalled, they can review the last action and speak with the person responsible.

  • Pick out the specific work that needs attention.
  • Set the immediate priority with the person responsible.
  • Coach from the actual record, such as the proposal, job or invoice in question.
  • Adjust assignments or timing when the team’s workload shifts.

From a signal to a decision

Back to section

From a signal to a decision

The decision sequence

  1. Review the flagged activity.
  2. Check whether the record is complete and current.
  3. Understand the surrounding context.
  4. Decide whether to follow up, reassign, reprioritize or leave it alone.
  5. Explain the decision using the source record.
  6. Record the next action and review the result later.

What AI prepares and what the leader decides

AI can prepare a summary, surface gaps, organize records or draft a possible next step. The department head verifies the facts, interprets the context, assigns responsibility and coaches the person doing the work.

A billing example

In an illustrative case, a leader sees an invoice marked incomplete. Before assigning follow-up, they check whether required information is missing, whether another department owns the dependency and whether the record shows the latest status. Then they assign the right next action and coach the responsible person from that record.

When the same signal means different things

A missing record is not automatically poor performance. A completed task is not automatically quality work. The context decides which action fits.

Back to Decide and coach

Roll out what works

A leader who has used the workflow firsthand knows what it takes. They have seen where it saves time, where it needs a human check and which steps matter most. The team then learns a way of working their leader already uses.

A department rollout takes more time, money and shared infrastructure. It may produce a smaller percentage increase while adding more output across the team. Build on what the leader has learned when the department is ready to adopt it.

What makes the team ready

Back to section

What makes the team ready

The rollout sequence

  1. The head uses the workflow in their own work.
  2. They identify the steps that produced a useful result.
  3. They explain the purpose and expected behavior to the team.
  4. Shared information, records and responsibilities are set up.
  5. Team members use the workflow in their own work.
  6. The head checks adoption and quality, beyond completion.

What can be shared and what needs an exception

The head can share the repeatable workflow, its definitions, worked examples, a review cadence and the records it requires. A shared template or a weekly activity review are typical starting points. Work with different inputs, approvals, dependencies or judgment calls needs its own handling.

Checking adoption and quality

The head leads the rollout and team members perform and record the work. Adoption means the team actually uses the workflow. Quality means the records are accurate, the next actions are clear and the work still fits the situation. A finished checklist with no useful work behind it does not count as adoption.

Back to Roll out what works

See activity

The head sees that Rep A sent 6 proposals this week, Rep B sent 5 and Rep C sent 1. Four of Rep C’s open proposals are separately flagged as stalled.

Two levels of visibility

The example shows the team activity count and, beneath it, the individual records that need review. The count shows where to look. The records show what is actually happening.

What to check

Each of the four stalled proposals should be checked against its record and context. The count alone does not explain why they stalled or which action is appropriate.

Back to section

Coach and set priority

The head asks Rep C to follow up on the 4 stalled proposals today and directs the whole team to use Rep A’s follow-up template.

How the example works

The leader turns a signal into a decision, assigns a specific action to a named person and picks out a practice that may be worth sharing with the team.

What to check

The example does not establish that Rep A’s template is right for every case or that Rep C’s lower count explains the stalled work. The head confirms the stalled records first and reviews how the template works for the team.

Back to section

Same team measure after rollout

The same team activity measure shows 12 proposals sent per week before rollout and 18 afterward. These values are illustrative.

Keeping the measure comparable

The two numbers can be compared only if the definition of a proposal sent, the team in scope and the counting method stay the same.

What the change shows

The count shows a change in recorded activity. It does not show revenue, proposal quality or that AI caused the change. Its role is to complete the leadership story of seeing the work, making a decision and spreading a practice.

Back to section

Back to section

Sales example

  1. 1. See activity

    Proposals sent this week

    • Rep A6
    • Rep B5
    • Rep C 4 stalled1
    Explore this example

  2. 2. Coach and set priority

    Rep C: follow up on the 4 stalled proposals today.

    Whole team: use Rep A's follow-up template.

    Explore this example

  3. 3. Same team measure after rollout

    Team proposals sent per week

    • Before12
    • After18
    Explore this example

Illustrative values. Team rollout goes smoother once the lead already works this way.

Step 3 · Revenue work

Point freed hours at work that earns

Sales

AI-assisted action
Drafts proposals and follow-ups
Activity measure
Proposals sent, follow-ups done
Business result
New clients and revenue booked

Delivery and service

AI-assisted action
Speeds onboarding and catches rework early
Activity measure
Days to onboard, rework rate
Business result
More work delivered and clients retained

Finance

AI-assisted action
Shows margin by service and staff hours per job
Activity measure
Margin per service, hours per job
Business result
Clearer pricing and staffing decisions

Tracked separately: payment speed, from invoice sent to cash received.

Worked example: Sales

  • 5 reps, 50 proposals a month
  • 4 h → 2 h per proposal
  • 100 h freed
  • 2 h per new proposal
  • 25% close rate
  • $10,000 per client
  • 30% gross margin
  1. 100 hfreed
  2. 40 hput into proposals
  3. 20extra proposals
  4. 5new clients
  5. $50,000booked
  6. $15,000gross profit if all 5 are served
$15,000 gross profitAI setup and running costsNet return

Hypothetical example. Confirm delivery capacity and quality before counting the profit.

Step 4 · Bottlenecks

Fix one area and the bottleneck moves

Before: Delivery unchanged

  • Sales wins5 clients
  • Delivery can start2
  • Waiting3

2 started, 3 waiting

After: Delivery equipped

  • Sales wins5 clients
  • Delivery can start5
  • Waiting0

5 started, 0 waiting

Before sending more work downstream: check delivery capacity, quality and real demand.

Billing, tracked separately: late invoices slow cash. Watch days to payment on their own.

Illustrative values.

Step 5 · Our assessment

How BH Technology assesses

  1. Map where money is madeSales, delivery, service and finance
  2. Read the dataTeam activity plus financials
  3. Pick the first priorityWhat holds back revenue or margin
  4. Equip the lead, then the teamStart with that department's head
  5. Measure the resultOutput, quality, actual cost and margin
  6. ReassessFind the next constraint

Back to step 1 with the next constraint

Find your first priority

We start with the activity that can change a business result.

Book a consultation